Analysis on IT trends and competitive strategies, with emphasis on micro processors, computer systems and networks. Based on latest news, backed up with real data, this site intends to provide a true and realtime picture of the fast changing IT landscape. This journal strives to be accurate on facts and sharp on criticisms. You may email your opinion to sharikou@yahoo.com or post comments here, be cool and intelligent.
Freelance journalist on IT matters. Some of my writings have been published on online IT journals. Any original content on this journal is Copyrighted, but it's free for non-commercial use. Any Trademarks used on this site belong to their respective owners. Some of the pictures are links. If there is any issue with the content of this site, please email sharikou@yahoo.com .
K10 on time to frag Intel.
posted by Sharikou, Ph. D. @ 5/22/2007 06:05:00 PM
154 Comments:
Who cares? By the end of this year Intel will have four 65nm fabs and two 45nm fabs operational. In 2008 they're bringing another 45nm fab online and completing the conversion of the 90nm fab in New Mexico to 45nm.
Intel can flood the market with low priced 45nm CPUs and force AMD's BK.
AMD BK Q2'08.
This is what Pat Gelsinger had to say about Barcelona:
Pat commented that the new quad-core Barcelona from AMD (AMD) was “Too Little, Too Late when compared with Penryn”, and that at best it was a catch up effort by AMD.
Source: http://www.nyquistcapital.com/2007/05/22/intel-on-amds-barcelona-too-little-too-late/
Pat Gelsinger is in an enviable position. Penryn coming this year, and another brand new architecture coming next year. The world's most advanced 45nm process this year. And perhaps most importantly, Intel has the desire to kill off AMD once and for all. With all this technology they will do just that.
AMD BK Q2'08.
Randy Allen said...
Intel can flood the market with low priced 45nm CPUs and force AMD's BK.
Like they did with the P4? AMD saw high profits during their "yard sale". Intel can flood the market with as much of their garbage as they want, but it will just sit idle of shelves like it has in the past.
Randy Allen said...
Pat commented that the new quad-core Barcelona from AMD (AMD) was “Too Little, Too Late when compared with Penryn”, and that at best it was a catch up effort by AMD.
Complete idiocy. Barcelona is set to pummel the Core 2 by 40%, and Penryn will not be enough to save it. History has shown that a simple die shrink has a max benefit of a 10% speed increase, with diminishing returns as the size decreases. This means that Barcelona will crush Penryn by a devastating minimum of 30%.
Patty is hyping Intel so he could sell more options before he joins the army of unemployed.
Complete idiocy. Barcelona is set to pummel the Core 2 by 40%, and Penryn will not be enough to save it. History has shown that a simple die shrink has a max benefit of a 10% speed increase, with diminishing returns as the size decreases. This means that Barcelona will crush Penryn by a devastating minimum of 30%.
Do you have any proof of this? Intel leads with Clovertown in 2P servers by a whopping 79% already.
This comment has been removed by the author.
Link: http://tweakers.net/reviews/661/4
Woodcrest is 24% faster than Opteron. Clovertown is 79% faster. This is in 2P servers.
The day you can buy quad-core Barcelona. Intel will use theyr secret weapon, the FraggingCore. Produced at 22nm,a 16Core running at 5GHz.The 16Core FraggingCore cpu uses a ultra high speed radar frequency pumped FSB.
Intel to the bitter end!
Penix
AMD saw high profits during their "yard sale".
If the yard sale started with the introduction of Core 2 Duo, please show, with quarterly reports, when AMD was profitable.
If it started prior to the release of Core 2 Duo, please link to the yard sale price cuts your talking about.
If I remember correctly the price cuts happened after the release of Core 2 Duo, and since that time AMD has pretty much been losing money and market share to Intel's yard sale.
Please provide some links, thanks.
@enumae
My post is unclear and I apologize for that. I am referring to the infamous P4 "yard sale" which Sharikou posted about several times before.
The sentence should have read:
'AMD saw high profits during Intel's desperate "yard sale" of worthless P4 chips.'
Yo "Ph"ony "D"octorate is that the only thing you can find of note about AMD is a 65nm conversion.
Yawn, boring, is that all you can do. You are as limp as AMD is..
INTEL has 4 65nm factories and is building 2 brand spanking new ones and converting two others. AMD is boring old follower news.
Thos 65nm factories are nearly depreciated and will be making C2D for celeron and selling at less then 75 bucks and making profits. While AMD with their new expensive 65nm have to compete against 45nm factoires pumping twice the density of transistors. Can you say 2 x density times 2x factories = 8 x more transistors. AMD is going BK soon....
Those silly dicks who keep talking about Barcebalogna crushing C2D.. that is fully expected. The problem is that the current rumored but not seen pile "Phenom" balogna has to compete against Penrym then Nehalem.. So the fact is can beat C2D is NO news at all. Its too little too late as Penrym is going to crush it in the high volume space where all the profits will be made. Then in the high end there will be too little volume on the honking chip on 65nm to make enough money to pay for that expensive fab.
Game over AMD is BK in 2008.. Poor sharikou and his "Ph"ony "D"elusions
penix
"History has shown that a simple die shrink has a max benefit of a 10% speed increase, with diminishing returns as the size decreases"
So it is but you forgot that Penryn is no simple die shrink.
Hrm... Shiradouche is hitting the sauce or doing drugs again.
Integrated Circuit?
x86?
300mm technology?
MMX?
SSE?
65nm?
45nm?
PCI?
PCI-Express?
USB?
Bluetooth?
SMP?
All Intel firsts which AMD copied.
Once again Ph(ake)d gets slapped with the truth.
Integrated Circuit?
x86?
300mm technology?
MMX?
SSE?
65nm?
45nm?
PCI?
PCI-Express?
USB?
Bluetooth?
SMP?
Add to that list:
Integrated Audio
Integrated Graphics
Integrated Network Interface
Virtualization
Hyper Threading
x87
Memory reordering
DDR
DDR2
DDR3
FBDIMM
Integrated RAID
SpeedStep
Also, don't forget MCM (Multi Chip Module). Yes, you heard that right. AMD will be copying the dual-cheese-burger solution for its eight-core.
Also, AMD is copying Intel's unified last-level-cache architecture, after whining for 3 years about how unified LLC causes cache thrashing.
Also do not forget:
Ring Interconnect
Mesh Interconnect
On-die routing (demoed with Polaris)
x86-based graphics (Larrabee)
Penryn is only %15 faster than kentsfield on wide range of tests.Barcelona is %50 faster than kenstsfield barca is 2500mhz kentsfield is 2933mhz did you see.
Barcelona will be faster than penryn %30 you will see.Shangai will frag enormously.
Intel to continue copying AMD.
First x86-64 codes after IMC after native quad core.:) Continue intel your engineers are idiot because they aren't creative.
IMC is not AMD invention, so isn't native dual core.
64bit instruction set is over ten years old, AMD just extended x86 to work with it and in my opinion didn't do that well.
That's funny because if it wasn't for Intel, AMD wouldn't even exist.
Maybe you need a history lesson Sharikou.
Sharikou said Intel does not innovate, it steals AMD's ideas: x86_64, dual core, performance per watt, spinning off flash division. The following are items Intel has not yet copied: IMC, HT, true quad core...
normally, i don't comment on what Sharikou said ...
it is just funny that he has to quote spinning off the flash division to make the list looks long, which is clearly a business decision.
2ndly, performance per watt is a metric, if you really want to compare it, then by chronologically, Intel has it since Pentium M, which is much earlier. Intel lost it in desktop and server (still the king in mobile) during the Prescott era, and regaining it with the Core. While I would not say that AMD copy Intel in this, but per what Sharikou said, it implied that AMD copy Intel in this performance per watt thingy! :)
then come the quad core thingy. Like what AMD likes to say customer don't care about nanometer, I would say so do the quad core implementation as long as they have the performance gain. Obviously, FSB allow intel has much easier approach in the QC. A few AMD execs lately also said that they should have the MCM approach. Again, it is not that AMD do not what to copy, it is that current AMD chip is harder to implement the MCM approach, and that's the real reason they have to go native.
Yes, i'd agree that Intel copy the AMD's x86-64 implementation, due to market, business decision and compatibility reason, as Itanium has failed to fill in the original targeted market. but so what, if you really want to whine about this, there are much longer list that AMD copy from intel's.
in today's technology, there is no way that you can avoid copying, for marketing, compatibility, etc reasons. That's why there are cross licensing among companies. of course there are some money terms involved due to the IP portfolio differences. If you have intel or AMD friends, may be you can try to ask and see if Intel is paying AMD, or AMD is paying Intel for the cross licensing :)
So whan is AMD going to come out with memory running > 2Ghz? Aren't they still poking around at 400Mhz?
http://valid.x86-secret.com/show_oc.php?id=200226
Had you read the information correctly you'd seen that RAM was running at "only" 1040MHz. FSB != RAM != HT.
http://tinyurl.com/25zgeb
At the Microprocessor Forum, Intel demoed its V8 workstation running the POV-Ray benchmark. The machine is equipped with Intel's Workstation Board S5000XVN, 2 quad-core Xeons 5365, clocked at 3GHz and 16GB RAM. And the results are simply impressive: Intel scored over 4,900 pixels per seconds versus a little bit over 4,000 for AMD's 4 sockets quad-core (Barcelona) system. Again, this is an AMD 16 cores system versus Intel's 8 cores V8 machine.
"Why do you need 16 cores, when you can do better with 8. Our 8 core system is 30% faster than the 16 core machine AMD showed to the press yesterday. I just don't understand how they can claim to be 40% faster", said Francois Piednoel, an Intel engineer present at the show.
Well, now AMD has some explanation to do and sooner rather than later. Because if they can't figure out what happened with those POV-Ray results they showed us, that's the end of it. At the show, Intel also demoed a system with a 45nm Penryn quad-core processor (shipping by year end) that is 40% faster than the top of the line quad-core generation, the Core 2 Extreme processor QX6800. Wow!
AMD is finished. Their 4P quad core Barcelona is no match for Intel's 2P quad cores. Intel's 8 core system is 30% faster than AMD's 16 core system. Looks like Barcelona will be another flop just like R600.
AMD BK 2Q'08
For the record, PovRay scales rather well, almost linearly. The other popular ray tracing benchmark, cinebench, does not scale nearly as well.
I would expect Intel showing well over 5000 pixels per second results with 45n CPUs at same clock speed.
Of cource there is always the chanche that they did not use the same scenes and settings. Though that would be kind of odd since PovRay does have builtin standard benchmark.
^Referring to
http://tinyurl.com/3badr2
Hard Barcelona numbers appear
The Opteron machine on the left was started first and the Barcelona box finished in far less time. The raw results were about 2200 pixels per second on the Opteron while the Barcelona hit "just over" 4000 on the POV Bench.
AMD has NEVER done anything themselves, never, not a single thing.
The big issue is that PovRay is VERY easy to customize per CPU instruction set. Unless we can actually see PovRay running we don't know if they are doing a custom instruction set for the render.
Apples to Apples both renders on the systems need to be running the same binaries.
ho ho,
Intel uses a quad pumped fsb.
That memory was running 2080Mhz.
Further reporting is here:
http://techreport.com/onearticle.x/12526
ycon
"The big issue is that PovRay is VERY easy to customize per CPU instruction set."
Unless you claim that the makers of PovRay are as evil as the people that made ICC please tell us what instructions does Barcelona not have that Clowertown has. Also I'd like to know what kind of instructions do give it almost twice the speed over Barcelona.
Anyways, twice the cores and twice the performance would be really bad if those AMD machines on the URL were running at similar speeds. Anyone has got K8 to see how many pixels can it render in povray benchmark? With those numbers it would be trivial to calculate the speed of K8 based Opteron system details.
Same goes for Intel systems. If anyone can give any results with the newest beta then it would be interesting to see and really simple to see if one of the two is cheating. Unfortunately you can't download it from their home page as they have some weird expired versions there.
bubba
"Intel uses a quad pumped fsb.
That memory was running 2080Mhz."
Seems as I misread the original link myself, sorry about that. I remembered wrong how cpuid presented RAM speeds.
That's rather interesting and is also making the whole dualchannel thing meaningless as two sticks of RAM at 2GHz will need FSB at 4GHz to deliver all of their bandwidth to CPU.
Since PovRAY is open, its very easy to customize per processor. There is already an AMD optimized (and Intel optimized) version. Both optimizations are geared and tuned specifically for the highs of the processor in question.
I'm NOT saying it was geared for Barcelona, just that its NOT apples to apples to use PovRAY on the two systems unless its compiled to take advantage of neither processor... it needs to be processor neutral.
Penix says: 'AMD saw high profits during Intel's desperate "yard sale" of worthless P4 chips.'
Hahahah kinda like the yard sale AMD has rite now with the release of C2D except Intel during the yard sale was still making money compared to AMD and its mediocre marketshare and lack of profits.
AMD reduced to selling FX-62 for about $300 how pathetic is that.....and from the looks of it pretty soon barcelona will be the next yard sale for AMD even before penryn comes out.
evil_merlin
Since PovRAY is open ..."
If you take a look at its home page you'll see that only the old non-SMP 3.6 is open. You can't just download and modify the new 3.7 betas.
"There is already an AMD optimized (and Intel optimized) version."
Where can I get one?
"I'm NOT saying it was geared for Barcelona, just that its NOT apples to apples to use PovRAY on the two systems unless its compiled to take advantage of neither processor... "
Only way that would be possible is for either company to get the unreleased code and heavily optimize it. Do you suggest this is possible?
I'd say that even though we know little about the details we can say that Barcelona has nowhere near the FP performance of Clowertown, at least not in PovRay.
Even if through some magick the application really is optimized towards one or the other there is no way you could optimize it to run twice as fast on half the cores and on CPUs with less special instructions.
There is also another possibility, perhaps Barcelona was running at super slow clock speed, something in the order of 1.3-1.8GHz. That would explain things quite well.
heat
"AMD reduced to selling FX-62 for about $300 how pathetic is that..."
Also starting from 22'th July you can get 2.33GHz 1.33GHz FSB dualcore for $163. That means AMD fastest K8 will be costing way under $200, down from current >$300. Not good for neither revenues nor profits.
So, AMD gets the credit for extending the 80386 instruction set to 64-bits the way Intel extended the 8086 instruction set to 32-bits 20 years ago?
AMD gets the credit for IBM's introduction of dual-core back in 2001 with the POWER4?
AMD gets the credit for the idea of CPUs with low power consumption?
AMD is finished. Their 4P quad core Barcelona is no match for Intel's 2P quad cores. Intel's 8 core system is 30% faster than AMD's 16 core system. Looks like Barcelona will be another flop just like R600.
Excellent. 2P Clovertown system is faster than a 4P Barcelona system! It's as Pat Gelsinger said, Barcelona is too little too late.
AMD BK Q2'08.
AMD gets the credit for the idea of CPUs with low power consumption?
No, because Intel invented that idea before AMD did.
About PovRAY, it's been a LONG time since I used it, and I didn't know they changed it that much. There USED to be custom versions for AMD and Intel chips. I was mistaken on any comments. The last time I used it (forget what version) there were indeed different binaries for Intel and for AMD
Poke said...
AMD is finished. Their 4P quad core Barcelona is no match for Intel's 2P quad cores. Intel's 8 core system is 30% faster than AMD's 16 core system.
Intelers are so predictible sometimes; claiming victory over a single erroneous benchmark. What completely absurd FUD.
We know Barcelona will be a minimum of 40% faster than Clovertown. As Abainstein demonstrated, the Intel FSB severely limits the scaling of the platform in real world situations. Finally, we know that Intel does not plan to fix the FSB problem until 2009.
What is Intel's answer to AMD's Barcelona 16 core? Nothing. Intel has a performance crippled 4 core, and cannot scale beyond that. Penryn will offer no performance gain over Clovertown, because it also uses the same worthless FSB.
All you ignorant Intel fanboys who want to buy Intel trash should get your wallets ready. Intel is about to have another yard sale.
Penryn will offer no performance gain over Clovertown, because it also uses the same worthless FSB.
This is proof you have no clue what you are talking about.
1.) Penryn is .45um, vs Covertowns .65um. Which isn't neccessarily a performance booster, but Intel has proven time and time again that it's die shrinks tend to also give a performance boost...
2.) SSE4 (and not AMD's half-assed implementation of them)
3.) hafnium-based dielectric
4.) Native DDR3 support
5.) Larger L2 cache.
I can go on and on, as all of these are performance enhancement features.
But you are the know-it-all Penis. Please enlighten us.
Evil_Merlin said...
I can go on and on, as all of these are performance enhancement features.
Correction, performance enhancement buzzwords.
penix
"Intelers are so predictible sometimes; claiming victory over a single erroneous benchmark"
As povray is part of SpecFP2006 does that make Spec benchmarks errorneous too?
"As Abainstein demonstrated, the Intel FSB severely limits the scaling of the platform in real world situations"
Wasn't that AMD who tried to show how much better Barcelona is in PovRay? Is AMD using synthetic benchmarks that are not from this world? Also as was seen, Clovertown still beats it by a wide margin even when having half the cores.
If Intels FSB is so bad then why it still beast 4P Barcelona with roughly 4x the memory bandwidth?
"What is Intel's answer to AMD's Barcelona 16 core?"
It will have 4P platform out soon, unfortunately I don't remember its name.
penix
"Correction, performance enhancement buzzwords"
Besides the buzz they do increase performance. Also two of the most important ones were left out:
6) double speed FP divide unit
7) double speed shuffling
Single benchmarks showing AMD on top are a revelation to behold, single benchmarks showing Intel on top are erroneous.
kevin
"Single benchmarks showing AMD on top are a revelation to behold"
There have been some real benchmarks of Barcelona? Where can I see them?
ho ho said...
Besides the buzz they do increase performance.
Here is some of Intel's previous buzz which also "increased" performance:
- Improved branch prediction unit
- Double-clocked ALUs
- Execution trace cache
- 2 cycle low-latency L1 data cache
- 256bit single cycle L2 cache
- 100MHz quad-pumped front-side bus
- SSE2 SIMD extensions
- 20 stage pipeline
Wow, the P4 sounds AMAZING!
penix
"Wow, the P4 sounds AMAZING!"
It is, you just need to compare it with its predecessors, some things need specially targeted code to show up, especially the long pipeline. It is similar to how K8 is good at FP and not that good in integer performance compared to C2D.
E.g SSE2 can give it twice the performance of P3, some integer tasks are nearly twice as fast, FSB bandwidth is considerably higher than on P3 etc. Even Core2 took parts of P4 to its architecture. There actually were quite a few really good things about P4, just that it didn't suite that well for lots of things.
Btw, what is "256bit single cycle L2 cache"?
As for Penryn I wouldn't be surprised to see it on sale before Barcelona, same for benchmark results.
To "randy allen" sorry to burst your bubble..Who cares? By the end of this year Intel will have four 65nm fabs and two 45nm fabs operational. In 2008 they're bringing another 45nm fab online and completing the conversion of the 90nm fab in New Mexico to 45nm.
The plant in Mexico is shutting down read Intels news....
Only amd said..."The plant in Mexico is shutting down read Intels news...."
Um, if you meant, New Mexico, only 1 of the fabs is closing. The other is being converted to 45nm.
One fab is 200mm and the other is 300mm. They are closing the 200mm fab this year.
Stop avoiding the subject Penis. You claim that there are no performance enhancements between Clovertown and Penryn, where there clearly is.
Never mind the fact that for right now, Clovertown is a server processor and Penryn is a notebook/workstation processor.
But your the fanboi, so I'm sure you see things in your own, quite warped way.
ho ho said...
It is [amazing], you just need to compare it [the P4] with its predecessors, some things need specially targeted code to show up, especially the long pipeline.
And I bet you're buying up as many Netburst CPUs as you can.
ho ho said...
Btw, what is "256bit single cycle L2 cache"?
The P4's L2 cache interface is 256 bits wide, and it sends data on every clock cycle.
evil_merlin said...
You claim that there are no performance enhancements between Clovertown and Penryn, where there clearly is.
The Penryn hype is just as big, and I'm sure will be just as much of an enhancement as the P4 was over the P3. Nevermind that the P4 was the laughing stock of the industry from start to finish.
It really doesn't matter if Penryn is an improvement or not. Intel's FSB is a pathetic joke and will not be replaced until 2009. This bottleneck will prevent Penryn from acheiving any performance gains. The FSB is maxed out. Clovertown is as fast as it's going to get for the next 2 years.
This comment has been removed by the author.
doorknob_dan, you are clearly a person of very low IQ. There are some obvious fanboys here, but I always make a conscious effort to remain neutral and unbiased at all times.
"Never mind the fact that for right now, Clovertown is a server processor and Penryn is a notebook/workstation processor."
Penryn is a notebook processor. The biggest improvement of Penryn over Clovertown is its improved power management. And you claim that Intel's going to release a mobile processor that outperforms its best server part. Are you not a fanboi?
Well, maybe it does, and people who run servers will all go buy Penryn instead. Lets just wait and see...
"AMD is finished. Their 4P quad core Barcelona is no match for Intel's 2P quad cores. Intel's 8 core system is 30% faster than AMD's 16 core system."
This would make Intel's C2D 140% faster than AMD's K8 (1.3x * 1.85x = 2.4x). Oddly this is not observed between my C2D E6600 and A64 X2 4600+ no matter how I bench them.
It takes not much more than a fool to know that such direct comparison of one benchmark result is false. I guess there are more fools than Intel's marketing (who are probably just foolishly shameless).
"Intel also demoed a system with a 45nm Penryn quad-core processor (shipping by year end) that is 40% faster than the top of the line quad-core generation"
This "comment" is so f*cking biased and misleading. 40% faster on what? Some SSE4 instructions with 1%-usefulness?
And people are stupid enough to buy into such comments or claims. Imagine...
And people are stupid enough to buy into such comments or claims. Imagine...
You are stupid enough to believe slide shows and so is everyone here claiming barcelona superiority. The only benchmark not based on a slide show by AMD shows barcelona getting its ass handed to it by an Intel processor using half the number of cores.......what does that say about you??/
Intel's FSB is a pathetic joke and will not be replaced until 2009. This bottleneck will prevent Penryn from acheiving any performance gains. The FSB is maxed out. Clovertown is as fast as it's going to get for the next 2 years.
as much as a "pathetic joke" as it is, it is STILL beating AMD performance wise dollar for dollar and watt for watt...
Penryn is a notebook processor
No, its actually the replacement for the Merom/Conroe/Woodcrest line... its scheduled to first appear in the notebook lineup, but its geared to replace all of the above in the order above.
Don't let facts stand in the way of your fanboism now.
I highly recommend you read up again before shooting off your ignorance again:
http://arstechnica.com/news.ars/post/20070328-intel-spills-beans-on-core-2-successor-sse4-faster-virtualization-bigger-caches.html
the most important part of the above link:
All told, Intel will introduce six Penryn products this year, spanning the full range of segments from ultra-mobile to server. A full fifteen Penryn products are currently in development.
"What is Intel's answer to AMD's Barcelona 16 core?"
It will have 4P platform out soon, unfortunately I don't remember its name.
It is called Clovertown 2P (8 cores). As Intel demoed, Clovetown 2P is sufficient to frag Barcelona 4P by 30%.
Jokes aside, Intel's MP processor is called Tigerton, and the platform is called Caneland.
Abinstein
This would make Intel's C2D 140% faster than AMD's K8...
Actually in SPECfp - 453.povray, Intel is 140% faster.
Intel's 2P 5160 (108)
AMD's 2P 2222SE (76.9)
I am not trying to say anything relating to Barcelona, but simply pointing out that Intel is indeed faster clock for clock than AMD's K8 in 453.povray.
This comment has been removed by the author.
Intelers are so predictible sometimes; claiming victory over a single erroneous benchmark. What completely absurd FUD.
Oh that is so damn funny. Don't you remember when Sharikou started linking to AMD's benchmarks? Firstly he linked to two (2) benchmarks on AMD.com over the new Opteron 3Ghz that showed them beating the Xeon 5160 and claimed this was proof that AMD was faster.
Then AMD posted ONE benchmark that makes good use of an intergrated memory controller (so it's easy to expect Opteron will beat Clovertown in this) and then claimed this as proof that Opteron was faster than Clovertown! Then you agreed with him! Now we post results showing the opposite and you claim they are fraudent? Pffft.
Oh yes, here's your comment from that old article
Intel's best cannot compete with AMD's current generation. Barcelona will leave Intel in a grotesque pool of shame and embarrassment. To make things worse, Intel will surely try to twist the situation with their fabricated benchmarks. Only fools buy Intel.
Furthermore, here's the link to Sharikou's extremely entertaining post:-
http://sharikou.blogspot.com/2007/04/3ghz-opteron-frags-3ghz-woodcrest-in.html#links
The plant in Mexico is shutting down read Intels news....
If you did even the slightest hint of research before posting here you'd that as someone posted earlier, that Intel has TWO fabs in New Mexico. They are closing (selling maybe? Make some $$$) the older 200mm fab and keeping the 300mm/90nm fab while transitioning it over to 45nm next year.
Intel just doesn't need as many older fabs. Not with four state of the art 45nm fabs coming online. It wouldn't suprise me if they continue selling off and closing old pre-90nm fabs in the future.
So in recap:- Penix is a hypocritical fool, and only amd doesn't pay proper attention to the news.
Intel 8 core frags AMD 16 core by 30%. This is the most solid Barcelona performance data we have so far.
AMD BK Q2'08.
randy allen
Oh that is so damn funny. Don't you remember when Sharikou started linking to AMD's benchmarks? Firstly he linked to two (2) benchmarks on AMD.com over the new Opteron 3Ghz that showed them beating the Xeon 5160 and claimed this was proof that AMD was faster.
That wouldn't be the best representation of Sharikou's extraordinarily high intelligence.
Once he used a survey, which concluded that Intel users are less technology savvy than AMD users, to support his claim of "Intel engineers are also more stupid than AMD engineers".
While everyone, including Rahul Sood, an AMD fan, worried about AMD's future, Sharikou's the only one claiming AMD's superiority, and will "kill", "frag", Intel within the next two quarters. Please note that Intel has already released benchmarks as well as launch time frame for Penryn, while AMD hasn't shown anything credible but marketing BS and FUDs.
40% faster on what? Some SSE4 instructions with 1%-usefulness?
It should not be that hard to figure out. What does this look like to you?
40% faster in video encoding is truly impressive. It's clear that Penryn is an awesome upgrade of the current Conroe and Kentsfield products. As if all this wasn't enough, they have Nehalem coming next year as well.
enumae -
"Actually in SPECfp - 453.povray, Intel is 140% faster.
Intel's 2P 5160 (108)
AMD's 2P 2222SE (76.9)"
enumae... do you have to degrade your intelligence every time you try to respond to me???
108 over 76.9 is 40% faster, not 140%. Read my comment above. I clearly said 1.3x * 1.85x = 2.4x, which is 140% faster.
Now accept the fact that it's simply erroneous. You can't produce such difference no matter how you bench a C2D against a K8. If you (or anyone) can't get this simple fact, then you are a damned dumb fanboi, period.
"It should not be that hard to figure out. What does this look like to you?"
That encoding looks so slow to me. I remember a Barcelona system with 8 cores was demoed to encode HD 1080p video in near real-time. I guess Intel's 40% faster than its current fastest isn't going to match up with Barcelona after all?
AMD continues to be unspecific about Barcelona and now we can see why.
AMD Explains the Fragging,
http://tinyurl.com/2cywkv
Barcelona recent lack of performance and why AMD should start to worry
This morning I received a response from AMD on the 16 cores Barcelona system lackluster performance versus Intel's Clovertown V8 machine (8 cores) running the same POV-Ray benchmark. I'm still not convinced and until we actually can see/try out the "high-performance" quad-core chip AMD is referring to in their response, everything is still "up in the air"... and then, there's the 40% boost in performance over Clovertown that Intel is promising with its upcoming 45nm Penryn. I think it's now time for a reality check! Henri?
"The objective of our demo was to show performance scaling from our current dual-core processors to our upcoming quad-core processors within the same thermal envelope and drive home the point through a real-world demonstration that customers could expect to see 2x the performance without an increase in power consumption."
Whats wrong Henri, what happened to the 40% over Clovertown? Is this the best excuse Henri and Hector could come up with?
AMD BK Q2'08
"You are stupid enough to believe slide shows and so is everyone here claiming barcelona superiority."
Slide shows?
1. A system with 8 Barcelona cores was seen encoding Spiderman 3 to HD 1080p format in near real time. OTOH, the encoding video of dual-socket Penryn show speed so much slower than real-time.
2. 16 Barcelona cores double computing power of 8 K8 cores within the same power envelop. There is nothing from Intel that can do that, not in 8 cores nor 4 cores. Not only me but all datacenter managers are going to get impressed.
3. C2D is about 36% faster in PoVray than K8 at the same clock with the same number of cores. This fact is shown all over SPECfp submissions, proved and solid. Only fools or fanbois will believe Intel's BS marketing that C2Q is 30% faster than Barcelona with half the number of cores.
You get the conclusion, don't you? Now since you know at least how to use computers to post comments I don't suppose you are that stupid, so please remember to raise hands when people call Intel fanbois...
I guess Intel's 40% faster than its current fastest isn't going to match up with Barcelona after all?
At the rate AMD is going i wouldnt be surprised to see barcelona match up well and compete with the celeron. They mite want to go their GPU route and not bother with competing with the best anymore but aim for something lower and hope their blind fanbois will still pay money for trash
abinstein is wrong as usual.
http://www.intel.com/products/processor/xeon/competitive_guide.pdf
Woodcrest is up to 2.39x faster than Opteron. Clovertown is up to 4x faster than Opteron.
With SSE4 Intel's video encoding results will blow AMD away in this category. AMD's "SSE4A" is just two SSE3 grouped instructions and two new instructions. SSE4 is over forty new instructions.
AMD will continue to lose market share throughout the year and post bigger and bigger losses throughout the year before they finally run out of cash in Q2'08. They will be down to about 15% by the end of Q2'07. Remember folks, "Barcelona is too little too late compared to Penryn".
AMD BK Q2'08.
http://techreport.com/reviews/2007q2/core2-qx6800/index.x?pg=9
This link is povray benchmark 3000mhz k8 4 core equal 2933mhz QX6800
What do you thing about it?
"16 Barcelona cores double computing power of 8 K8 cores within the same power envelop. There is nothing from Intel that can do that, not in 8 cores nor 4 cores."
A few things I'd like to add on this matter -
1. The higher the number of cores go, the harder it is to scale up performance. Think of 16 cores that double computing power not just on Povray, but also a whole array of other applications. There is no Intel system that can do that.
2. The higher the number of cores, the more advantage K8 has over Core 2 Duo. This is not going to change with Penryn.
3. The absolute value of any single Povray benchmark is meaningless. You can enable/disable certain options in the benchmark, making cross comparisons erroneous.
I guess Intel is scared of K10 and is bullsh*tting false claims trying to make AMD respond and suffer the Osborne effect. AMD is just not as stupid as many here.
"At the rate AMD is going i wouldnt be surprised to see barcelona match up well and compete with the celeron."
So your celeron can encode HD 1080p in real time?
Get lost.
"http://www.intel.com/products/processor/xeon/competitive_guide.pdf
Woodcrest is up to 2.39x faster than Opteron. Clovertown is up to 4x faster than Opteron."
I show you evidences from SPECfp submissions (link here again, as some have difficult to accept the truth), which are proved, repeatable, and acknowledged/certified, and you show what, Intel PR article? How pathetic is that?
"With SSE4 Intel's video encoding results will blow AMD away in this category."
You comments like this above convinced me that you have zero knowledge of computing, microarchitecture, or instruction set.
"AMD BK Q2'08."
Funny. I would have believed that a bit more if it weren't you who was speaking of it.
Stop being a moron... well, actually keep being one, who cares.
"This link is povray benchmark 3000mhz k8 4 core equal 2933mhz QX6800"
Well, if you go to povray's official website, then you see results that are all over the places. Some slower C2D beating faster, some with less core beating more.
In my opinion, the compiler optimization and the scene under test may have great effect on the rendering speeds. For this matter I'd rather trust SPECfp results (which indeed shows Povray favoring Core 2 Duo over K8).
The results speak for themselves. AMD is finished. After Q4'07 they will have 1% market share left.
AMD BK Q2'08.
penix
"And I bet you're buying up as many Netburst CPUs as you can."
I'm sorry but you would loose that bet. I did buy P4 when Prescott and K8 were not yet availiable and when x2 3800 was costing twice as much as P4 920. Those times Intel had considerably better bang for buck. Situation is different today. I'd rather get a cheap dualcore e2000 series CPU than P4, they are about as expensive as x2 3600 and will be availiable from next month.
"The Penryn hype is just as big, and I'm sure will be just as much of an enhancement as the P4 was over the P3."
It cannot be since P4 was a whole new uarch whereas Penryn is just upgraded Conroe.
"Nevermind that the P4 was the laughing stock of the industry from start to finish."
... and when K8 was not yet availiable Northwood was kicking AMD hard.
"This bottleneck will prevent Penryn from acheiving any performance gains."
If it is so then why is Intel beating AMD again and again?
"The FSB is maxed out."
You should know that they will be upgrading the speed up to 1.6GHz.
"I always make a conscious effort to remain neutral and unbiased at all times."
That just made my day.
abinstein
"The biggest improvement of Penryn over Clovertown is its improved power management."
What about those two things I listed? What about being twice as fast as Conroe in video encoding?
"Oddly this is not observed between my C2D E6600 and A64 X2 4600+ no matter how I bench them."
There are two possibilities how that is possible:
1) Barcelona scales awfully
2) Barcelona in their tests were running at awfully slow clocks, in the order of 1.3-1.6Ghz.
Take your pick.
"This "comment" is so f*cking biased and misleading."
... and AMD has been throwing out comments like that for months.
"40% faster on what?"
Perhaps wide variety of workloads? Also please see the benchmark with alpha version of divX. You can't say that video encoding is a meaningless thing that nobody needs.
"I remember a Barcelona system with 8 cores was demoed to encode HD 1080p video in near real-time."
Define "near real-time" and describe what kind of algorithms were they using to encode that video. What? You say you can't? Well, then I guess "It takes not much more than a fool to know that such direct comparison of one benchmark result is false."
"1. A system with 8 Barcelona cores was seen encoding Spiderman 3 to HD 1080p format in near real time."
Where? All I remember is seeing a picture of a box with lots of casefans and green LEDs and one sentence saying what you just said.
"2. 16 Barcelona cores double computing power of 8 K8 cores within the same power envelop."
It is the same (or better) with Conroe. Just compare their 2.66Ghz dualcore and quadcore, quad takes 10W less power than two dualcores. Better yet, compare their 50W quads with whatever other dualcore Conroe you like.
"Only fools or fanbois will believe Intel's BS marketing that C2Q is 30% faster than Barcelona with half the number of cores."
Then please explain why Barcelona showed so awful performance in PovRay? Perhaps because they had to lower the clock speed to awfully low to match the K8 thermals?
"So your celeron can encode HD 1080p in real time?"
The almost nonexistent information about Barcelona hints that it takes at minimum 8 Barcelona cores to do it.
"In my opinion, the compiler optimization and the scene under test may have great effect on the rendering speeds"
They both used the standard povray benchmark. That means same scene, same settings.
I still wait response from penix to those questions I asked before. Seems as he simply chooses to ignore them since they don't fit into his view of the world.
Intel delivering the goods now. Quad core up to 4x faster than the competition. Penryn coming this year that will increase the speed of Intel's fastest quad core by 40%. Nehalem coming next year. This will be the biggest change in Intel's architecture since the Pentium Pro in 1995.
AMD, OTOH, just keeps releasing vaporware and awful products. R600 is slower than the competition and more expensive. The same is true with Barcelona.
AMD BK Q2'08.
ho ho
don't argue with the lackies. All they can do is spew out FUDs like their master Sharikou does.
Sharikou's credibility BK Q208
yomamafor2
"don't argue with the lackies."
Why not? It is fun seeing them trying to make a point and mostly failing miserably :)
At least Scientia tried to have an intelligent conversation most of the time. Too bad same can't be said to most people here.
This is just awesome:- http://www.theinquirer.net/default.aspx?article=39829
I can't wait to upgrade to a Geforce 9 at the end of the year. Penryn will be a nice upgrade too, but since I have the Q6600 already I think I'll just wait for Nehalem.
Ho Ho said...
At least Scientia tried to have an intelligent conversation most of the time.
I disagree with this statement and here is my version :)
At least Scientia tried to have an intelligent conversation sometime, intelligently-twisted conversation more often, and ban you from conversation sometime :)
Another funny thing with scientia is that he himself and others repeatedly told much worser things about others than I did that one time. Still that one time was enough to earn eternal removing of my posts.
That only makes me think that he didn't like how I was sometimes proving his claims being wrong and he just saw that as a good way to get rid of me without totally undermining his credibility.
Abinstein
enumae... do you have to degrade your intelligence every time you try to respond to me?
LMAO!!!
...108 over 76.9 is 40% faster, not 140%.
I guess I am not as smart as you, and its to bad, but I do understand that...
76.9 * 1.4 = 108
That 1.4 would equal 140%.
So do you want to try this again?
--------------------------
...then you are a damned dumb fanboi, period.
The other day, I provided a link to an HKEPC article showing the HD2900XT in a positive light, and you said nice find.
Today, I post a comment that sheds a favorable light on Core 2 Duo and I am fanboi... Your maturity, once again is raining through.
Thanks for the laugh.
What no comments abinstein? Can't admit the fanboi in ya' was simply wrong? Or are you going to try to move the goal post now in the typical fanboi way?
All told, Intel will introduce six Penryn products this year, spanning the full range of segments from ultra-mobile to server. A full fifteen Penryn products are currently in development.
The other day, I provided a link to an HKEPC article showing the HD2900XT in a positive light, and you said nice find.
Today, I post a comment that sheds a favorable light on Core 2 Duo and I am fanboi... Your maturity, once again is raining through.
Thanks for the laugh.
Don't you understand the system here yet? You can post all the pro-AMD material you want. That's fine. But post even just a modicum of pro-Intel or pro-Nvidia material and you are clearly a fanboy. There can be no arguments on this issue!
AMD BK Q2'08.
PENIX said...
doorknob_dan, you are clearly a person of very low IQ. There are some obvious fanboys here, but I always make a conscious effort to remain neutral and unbiased at all times.
I may posess a low IQ, but you're an idiot.
Here's a nice example:
We know Barcelona will be a minimum of 40% faster than Clovertown.
So, you have your own benchmarks of Barcelona on hand? Wow, let's see them!
As Abainstein demonstrated, the Intel FSB severely limits the scaling of the platform in real world situations. Finally, we know that Intel does not plan to fix the FSB problem until 2009.
Wow! I had no idea a front side bus was so terrible! But ... wait ... the the Intel FSB CPU is clearly outperforming ALL of AMD's current CPU's! They could use a FSB until 2020 for all I care, as long as it outperforms everything else out there, there IS no "problem"! (Your term.)
What is Intel's answer to AMD's Barcelona 16 core? Nothing. Intel has a performance crippled 4 core, and cannot scale beyond that. Penryn will offer no performance gain over Clovertown, because it also uses the same worthless FSB.
So how did your testing of a Penryn go? Got some data for us? You continually make ridiculous presumptions in your posts. Or compare current technology to older technology. Or compare unreleased technology to unannounced technology. Go ahead and read a few of your posts over the last few months, bearing in mind this whole paragraph.
All you ignorant Intel fanboys who want to buy Intel trash should get your wallets ready. Intel is about to have another yard sale.
Great! I'm looking forward to the cheap prices! And I'm sure YOU certainly won't whine too much when you can get your beloved AMD cpu's for pennies.
You argue how future things justify your favour of AMD today. You argue how past Intel processors justify your position on AMD/Intel today.
But you neglect the present. The present is all that there is. And presently AMD is in pretty dire straits.
What is Intel's answer to AMD's Barcelona 16 core? Nothing. Intel has a performance crippled 4 core, and cannot scale beyond that. Penryn will offer no performance gain over Clovertown, because it also uses the same worthless FSB.
Tigerton is coming out in Q3'07, quad core CPU for MP servers ready to frag Barcelona. It will scale all the way up to 32P in servers from the likes of IBM. Barcelona is for a ultra low end 8P only.
enumae
I guess I am not as smart as you, and its to bad, but I do understand that...
76.9 * 1.4 = 108
That 1.4 would equal 140%.
So do you want to try this again?
i guess it is a language problem, as u said 140% faster. anyway, he will try to attack this statement, as this is the only thing he can do, beside claim others as fanboys, FUD, despite the fact that he is one.
The funny thing is that in the other post he claimed that people talk bad on ATI only after the it's purchase, despite the fact that those ppl provided solid links, despite the fact that ppl like him, host and some other fanbois talk bad about NVIDIA (the host would priase NVIDIA before this!! :))
some time i wonder, if at least one of them is actually AMD employee, i wonder what is the AMD and NVIDIA relationship after the purchase ... NVIDIA sort of saved AMD few years back with its chipset support:)
"ultra low end 8P"
I smile every time when I read yet another definition of "ultra low end" :)
Indeed, exactly what is an Ultra Low End 8P system?
I hate to see what that makes AMD's current dual core only processors...
I do that just to spite Sharikou. He used to say "Bensley is for super low end 2P servers only".
I guess Intel's 40% faster than its current fastest isn't going to match up with Barcelona after all?
I guess a system with 8 cores and 2 GPU encoding monsters encoding in "almost real time" is fragging a 4 core system intended to demonstrate CPU power.
enumae -
"76.9 * 1.4 = 108
That 1.4 would equal 140%.
So do you want to try this again?"
With every response of yours, I am one step more surer that my comments toward you were correct.
You are apparently not smart enough to understand the difference between 1.3*1.85 = 2.4x thus 140% faster, and 108/76.9 = 1.4, thus only 40% faster (or 140% as fast as the base score).
If 108 was 140% faster than 76.9, then what is 40% faster than 76.9? 31.1?
By assuming 8-core C2D 30% faster (1.3x) than 16-core K10, which in turn is 85% faster (1.85x) than 8-core K8, it is required that the 8-core C2D be 2.4x as fast as an 8-core K8, or 140% faster. This is simply erroneous unless you think all those SPECfp2006 submissions were false.
"i guess it is a language problem, as u said 140% faster. anyway, he will try to attack this statement, as this is the only thing he can do, beside claim others as fanboys, FUD, despite the fact that he is one."
I guess I was right, that these fanbois don't have high IQ. This is not just language problem. He is (by purpose or by stupidity) mixing up 2.4x, thus 140% faster, with 1.4x, thus 40% faster.
I am a "fanboi" of truths and facts. I show you repeatable, verified scores and analysis, nothing more. I set those ridiculous and erroneous claims straight. It's never my concern that fanbois like you wouldn't like it.
This comment has been removed by the author.
This comment has been removed by the author.
enumae -
"The other day, I provided a link to an HKEPC article showing the HD2900XT in a positive light, and you said nice find."
The article you found is good because its tests were done with the latest drivers, something none other "reviews" have done.
"Today, I post a comment that sheds a favorable light on Core 2 Duo and I am fanboi..."
There is a huge difference between 2.4x (140% faster) and 1.4x (40% faster). You found a case of the latter, and claim it is the former. Should I have say what, good boy, nice find, but poor logic?
"Your maturity, once again is raining through."
Just because I had to burst your bubble? Man, you're getting real old, and don't use it as your bragging right for wrong logic.
Intel FTW!
"Wow! I had no idea a front side bus was so terrible! But ... wait ... the the Intel FSB CPU is clearly outperforming ALL of AMD's current CPU's!"
Yes, FSB is adequate for single-socket Core 2 Duos, as shown here. It is actually also fine for dual-socket ones, on which C2D still outperforms K8 on integer codes by quite a margin.
hoho
"It will have 4P platform out soon, unfortunately I don't remember its name."
You mean Tigerton?
So abinstein, still no comment on how wrong you were about Penryn?
you claim to be a fanboi of truth and facts...
Are you going to claim like Penis: "Penryn will offer no performance gain over Clovertown, because it also uses the same worthless FSB.
or are you going to tap dance around this one too?
I mean after all is said and done you are a fanboi of truth and facts...
ho ho -
""This "comment" is so f*cking biased and misleading."
... and AMD has been throwing out comments like that for months."
AMD officially said Barcelona is going to be 40% faster than then current Clovertown on floating point workloads.
Penryn, OTOH, is claimed by Intel to be 40% faster than Clovertown on media encoding.
It is plain f*cking biased and misleading to pull the 40% out of context from either of the two claims above.
"Define "near real-time" and describe what kind of algorithms were they using to encode that video."
Well, so you finally get it, such direct comparison is invalid. But Intel and your fanboistic comrades have been making claims based on such direct comparisons over and over lately. They are just f***ed, aren't they?
"It is the same (or better) with Conroe. Just compare their 2.66Ghz dualcore and quadcore, quad takes 10W less power than two dualcores."
A C2Q taking 10W less than two C2D is taking 55W more than one C2D.
The quad-core Barcelona takes no more power than the dual-core Opteron.
Which you call is better?
"Better yet, compare their 50W quads with whatever other dualcore Conroe you like."
For the same power/thermal requirement, a QC Xeon E5320 can replace a DC 5150, or a QC Xeon E5310 can replace a DC 5140.
The problem is you are replacing 2.66GHz DC with 1.86GHz QC here (or 2.33GHz DC with 1.6GHz QC for the latter case). You can easily calculate performance increase as follow -
1. At the same clock rate, Xeon QC has about 1.66x throughput as much as Xeon DC.
2. With 70% clock rate (1.86/2.66 or 1.6/2.33), performance reduces roughly to 80%.
=> You get 1.66x * 80% = 1.33x, or just 33% more processing power under the same power/thermal envelop.
Remember the Barcelona demo shows QC K10 offers 80-90% more processing power than DC K8 under the same power/thermal envelop.
Intel is clearly losing, and it is (rightfully) scared.
abinstein, why does every post of yours have to start with some semblance of intelligence only to degenerate into the flames of a desperate AMD fanboy? BTW, please tell me how 8 cores with 2 GPU encoding omnsters encoding in "almost real time" without numbers used in a PR stunt to whet the appetite fanboys as yourselves is comparable to a 4 core system used to emphasize CPU power?
AMD officially said Barcelona is going to be 40% faster than then current Clovertown on floating point workloads.
Wrong, I remember "40% across a wide variety of workloads", "50% on FP", "20% on integer", the latter two at the same clock.
The quad-core Barcelona takes no more power than the dual-core Opteron.
First, you use Intel top bin to compare TDP to C2D. Then you compare quad HE to dual HE. ??? Second, TDP and actual power being consumed is a very different thing. I don't think you have measured Barcelona watts.
"you claim to be a fanboi of truth and facts...
Are you going to claim like Penis: "Penryn will offer no performance gain over Clovertown, because it also uses the same worthless FSB.
"
I'd agree more with a comment made by dakaha (search on this page) that Penryn - actually Yorkfield - will offer minor performance boost over Clovertown at the same clock rate.
Note that the ability to have 12MB on-die cache is very powerful by itself. In comparison, a quad-core K10 only has a bit over 4MB total cache size.
Penryn, however, is a mobile chip, and its focus is on low power usage and better power management. You don't meaningfully compare a Merom with a Xeon, do you? (If you did, then I'd feel sorry for you...)
Penryn is NOT just a mobile chip.
If you bothered to read when Intel first talked about Penryn they very specifically said:
Penryn will be the first core to benefit from the 45nm High-K and metal gate transistor technology and will be the foundation of future processors that span each product segment (mobile, desktop, and server) and power envelope.
and
All told, Intel will introduce six Penryn products this year, spanning the full range of segments from ultra-mobile to server. A full fifteen Penryn products are currently in development.
never mind the fact that there are massive differences between the Clovertown and Penryn designs as I listed before, including but not limited to:
1.) Penryn is .45um, vs Covertowns .65um. Which in itself with Intel's past has been proven for a 10% speed increase out the door.
2.) SSE4 (and not AMD's half-assed implementation of them). Which includes a large change and update to Intel's x64 implementation.
3.) hafnium-based dielectric. (High-K and metal gate transistor)
4.) Native DDR3 support
5.) Larger L2 cache.
6.) Microarchitecture optimizations
7.)Full-width, single-pass shuffle unit that is 128-bits wide.
8.) Xeons based on Penryn get the bus updated to 1600 Mhz buses.
Truth and facts indeed.
Care to go another round?
abinstein said...
Penryn, however, is a mobile chip, and its focus is on low power usage and better power management. You don't meaningfully compare a Merom with a Xeon, do you? (If you did, then I'd feel sorry for you...)
you clearly have not been reading tech news ... yes, Penryn refers to the intel's coming 45nm mobile processor, and yet, it also refers to the 45nm CPU family itself. Intel has said, it would release the 'server' CPU from its Penryn processor family.I feel sorry for you ...
"abinstein, why does every post of yours have to start with some semblance of intelligence only to degenerate into the flames of a desperate AMD fanboy?"
It's just fun to burst fanboistic bubbles. But I probably overdone it. ;-p
I have verified and repeatable scores and analysis by my side. I see no reason to be desperate.
"BTW, please tell me how 8 cores with 2 GPU encoding omnsters encoding in "almost real time" without numbers used in a PR stunt to whet the appetite fanboys as yourselves is comparable to a 4 core system used to emphasize CPU power?"
So one more person finally understands that direct comparison like such is meaningless.
"Penryn is NOT just a mobile chip."
Penryn is as much a mobile core design as Conroe is a desktop core design. Enough said on the nomenclature.
The point is with Penryn the most "improvement" is 45nm processing and power efficiency/management (together both would allow the larger cache size).
The fast divider is okay - it would make the Core 2 a bit more competitive on floating point workloads. Note that division is still slow and not recommended by any processor programming manual. My guess is Intel is going to show a few poorly optimized codes where excessive divisions are used, to show how (uselessly) good the Penryn's fast divider is.
The SSE4 is targeted to accelerate specific media encoding algorithms. Its use for the general computing is very limited.
Intel is in the lead here simply because it is controlling the SSE standard, and has the monopolistic authority on which algorithms are to be accelerated/optimized for.
"never mind the fact that there are massive differences between the Clovertown and Penryn designs as I listed before"
You just convinced me that you know nothing about computer microarchitecture. The changes in Penryn from Clovertown that you described are so localized that they don't even worth a section in a modern uarch book or paper. Probably just a paragraph. There could well be advancements in circuit design, though.
Anyway, the notion of "Penryn and Clovertown designs have massive differences" is just plain silly.
Abinstein
I could/should have worded my statement differently, but what I said is still true.
----------------------------------
2.4x (140% faster) and 1.4x (40% faster)
2.4x = 240% times x
1.4x = 140% times x
0.4x = 40% times x
x = 76.9
x * 1.4 (140%) = 108
Could you say it is only 40% faster... Yes.
Could you also say it is 140% faster... Yes, because the 100% would be represented by the base score, and then the additional 40% of the base score would be added.
This was reflected in one of my post.
It's just fun to burst fanboistic bubbles. But I probably overdone it. ;-p
You look more frustrated than threatening if anything ;) Maybe you would get more serious responses if you behaved correspondingly?
You look more frustrated than threatening if anything ;) Maybe you would get more serious responses if you behaved correspondingly?
Thats what happens when you put your faith in slideshows and make claims based upon them.......
Barcelona competing with the celeron Q108. AMD BK Q208........
Its like speaking to a wall, or a rock...
"You look more frustrated than threatening if anything ;) "
Not really frustrated, but kind of sorry that it takes a Snow White to make animals understand human singing...
I'll just pass on this task, and keep making funs of those. ;D
Maybe abinstein isn't getting frustrated, but I am. You Intel fanboys just don't fucking get it. I've tried over and over to reason with you idiots, but you're so blindly infatuated with Intel that you are incapable of seeing the truth. All the proof you need is laid out right in front of you yet you still option to cling desperately to every single piece of ridiculous FUD that Intel throws at you. You people are pathetic.
Did everyone hear about the latest Intel sponsored benchmark? Intel Clovertown beats powered off AMD Barcelona in everything, FOREVER!
Penix says: All the proof you need is laid out right in front of you yet you still option to cling desperately to every single piece of ridiculous FUD that Intel throws at you. You people are pathetic
Yep all the proof is in those AMD supplied powerpoint presentations with "estimated performance numbers".
Kinda like the proof on AMD's website where they cheat customers into thinking K8 is beating C2D when the whole tech industry points to otherwise.
Intel has spread no fud the only FUD is being spread by the penises and apesteins of the world.....
Did everyone hear about the latest Intel sponsored benchmark? Intel Clovertown beats powered off AMD Barcelona in everything, FOREVER!
I like the intel sponsored demo where the Intel Celeron beats vaporware barcelona better......
Hahahah crap AMD a few months away from their announcement of paper launch and still nothing to report
Hey Penis...
Pot, Kettle... Black.
You bitch and moan about Intel's benchmarks but are perfectly willing to jump for orgasmic joy when AMD releases benchmarks.
Or to put it one way:
You AMD fanboys just don't fucking get it. I've tried over and over to reason with you idiots, but you're so blindly infatuated with AMD that you are incapable of seeing the truth. All the proof you need is laid out right in front of you yet you still option to cling desperately to every single piece of ridiculous FUD that AMD throws at you. You people are pathetic.
abinstein
"AMD officially said Barcelona is going to be 40% faster than then current Clovertown on floating point workloads."
Yes, it said that but failed to prove it. Have you seen any benchmark data that would confirm AMD claims?
"Penryn, OTOH, is claimed by Intel to be 40% faster than Clovertown on media encoding."
Not only claimed but proved. Just see that video encoding link I posted before.
"Well, so you finally get it, such direct comparison is invalid."
So why you keep on repeating that Barcelona is so much better in video encoding?
"They are just f***ed, aren't they?"
There are f***ed up people everywhere, it is not a miracle that some of them are also fans of either Intel or AMD.
"1. At the same clock rate, Xeon QC has about 1.66x throughput as much as Xeon DC."
That is highly dependent on specific application and goes for both, AMD and Intel. Just compare the scaling of povray and cinebench to see it.
"=> You get 1.66x * 80% = 1.33x, or just 33% more processing power under the same power/thermal envelop."
You forgot to factor in the fact that DC5150 is 65W, not 50W.
Btw, did AMD use its low voltage (68W) dualcores in that benchmark against Barcelona? Or perhaps even its old 90nm 55W ones?
"Note that division is still slow and not recommended by any processor programming manual"
Yes, it is much slower than multiplication but you see no sane people doing such low level optimizations in most of their code. A single L1 cache miss is considerably more expensive than a division, not to mention what happens when you have to read stuff from RAM. If you don't believe me just download ODE and see for yourself what kind of code is used in physics engine that is used in Stalker and many other games.
"The SSE4 is targeted to accelerate specific media encoding algorithms."
Besides that tt also gives huge boost to ray tracing and other places where you have to do lots of vector math. Fast dot product is one of the nicest instructions in SSE4. Btw, what kind of other performance increases do you expect from a CPU? Even AMD is slowly moving towards the specialized cores, just the same as Intel.
It is rather expensive and pointless to create a CPU that can run scalar FP code rather fast when you could vectorize the same thing and run it even faster on considerably less optimized CPU. P4 vs K7 is a great example of that.
"Its use for the general computing is very limited."
Yet again that depends on your definition of general. If SpecFP is general enough for you then it does have quite a bit of benefit in real world too.
"Anyway, the notion of "Penryn and Clovertown designs have massive differences" is just plain silly."
Penryn is not revolutional, just evolutional and it seems there has been a considerable jump, err ... leap, ahead. Claiming anything different wouldn't be wise. SSE4 can give a huge performance boost to many algorithms and is quite significant. Only problem is that those applications would need to have special codepaths for SSE4. Though I don't think this would be a too big problem for the bigger players.
penix
"You Intel fanboys just don't fucking get it."
Get what? The fact that many people have little idea of CPU architecture? If so then I personally understood that months ago.
Btw, I'm still waiting for a reply to many of the questions I've asked so far.
"enumae said...
2.4x = 240% times x
1.4x = 140% times x
0.4x = 40% times x
x = 76.9
x * 1.4 (140%) = 108"
You are absolutely retarded. 2.4x is 140% MORE... not 140% OF. 1.4x is 40% MORE... not 40% OF.
Just multiply 1 x 1.4 and what do you get? You get 1.4. That's 40% MORE than 1. Multiply 1 x .4 and you get .4 which is 40% OF 1... not MORE. Did you skip every single day of 5th grade and up or are you naturally stupid?
abinstein said... "Penryn is NOT just a mobile chip."
Penryn is as much a mobile core design as Conroe is a desktop core design. Enough said on the nomenclature.
Well, another typical fanboys claim knowing everything, can't admit that he is wrong. Conroe is actually a Merom-based desktop design ...
abinstein said...
It's just fun to burst fanboistic bubbles. But I probably overdone it. ;-p
yes, especially fun when
1) you, penis, host, etc say things like "i'm giving my unbiased view ...blablablah",
2) or say "you guys hate ATI, spread FUD just because AMD bought it ... NVIDIA is a piece of crap, AMD fragged NVIDIA" ...
3) this benchmark link is good, nope, that link is not, oh just that page is correct, or maybe just a particular line of the paragraph, AMD slides is the most believable number than any other 3rd party bench number
and the list goes on. :) I'm just having real real fun visiting this blog.
Abinstein
I would like to apologize for my poor choice of words, and inability to clearly explain my interpretation of the 140% which has lead to this, as it should have said 40% faster, or 1.4 times faster.
"I would like to apologize for my poor choice of words, and inability to clearly explain my interpretation of the 140% which has lead to this ..."
Thanks for the clarification. An apology is too much, indeed. However, the acknowledgment from you that C2D is only 1.4x, not 2.4x, as fast as K8, thus debunking the silly claim of "Intel's 8 cores 30% faster than AMD's 16 cores," is quite a move of courage of intelligence.
abinstein said...
"I would like to apologize for my poor choice of words, and inability to clearly explain my interpretation of the 140% which has lead to this ..."
Thanks for the clarification. An apology is too much, indeed. However, the acknowledgment from you that C2D is only 1.4x, not 2.4x, as fast as K8, thus debunking the silly claim of "Intel's 8 cores 30% faster than AMD's 16 cores," is quite a move of courage of intelligence.
wow, what a ignorant. Just because AMD lose a bench, then you want to disqualify the bench? AMD itself didn't claim the bench is false but trying to come out with a reason, although it is a funny one.
http://www.uberpulse.com/us/2007/05/amds_barcelona_lack_of_performance_still_unknown.php
The bench is quite standard one and you can find some similar result for Intel in other bench site: http://www.bootdaily.com/index.php?option=com_content&task=view&id=460&Itemid=56&limit=1&limitstart=5
in this bench, Intel system is having 4GB that's why i said AMD's hint of having 6GB is a funnny one
it was ran 1 month ago, if the AMD marketing pay some attention to this, they would ran into this embarrassment. Calm down, this is just a single bench which favors Intel's current CPU against AMD's coming CPU, there are other benches too that we yet to know.
randy
"By the end of this year Intel will have four 65nm fabs and two 45nm fabs operational."
Let me show you some math:
In 2005 Intel had three 300mm FABs.
AMD had one 200mm FAB.
Since a 300mm FAB is 2X the capacity of a 200mm FAB this put Intel at 6X AMD's capacity.
Two 45nm x 2 = 4
Four 65nm + 4 = 8.
8 / 2 = 4
This would suggest that Intel will have a smaller production advantage than they had in 2005. How is that an improvment?
ho ho
"As for Penryn I wouldn't be surprised to see it on sale before Barcelona, same for benchmark results."
Would you bet money on that? I would indeed be surprised to see any volume of Penryn ahead of Barcelona.
evil merlin
"Integrated Circuit?
x86?
300mm technology?
MMX?
SSE?
65nm?
45nm?
PCI?
PCI-Express?
USB?
Bluetooth?
SMP?
All Intel firsts which AMD copied."
X86 is just an ISA; this is nothing to count as a first. Why didn't you mention though that Intel invented the microprocessor? Of course, then you have to mention that AMD invented the Bit Slice.
You can't count integrated circuits since both AMD and Intel came from Fairchild, and SSE came after 3DNow! PCI took awhile to become better than VLB so being first wasn't much of a gain, same with PCI-e, USB and Bluetooth. MMX, 300mm, and SMP are good examples though and, yes, Intel is typically ahead on process technology.
core2dude
"Add to that list:
Integrated Audio
Integrated Graphics
Integrated Network Interface
Virtualization
Hyper Threading
x87
Memory reordering
DDR
DDR2
DDR3
FBDIMM
Integrated RAID
SpeedStep"
Integrated audio, graphics, and networking are a joke. Even the Atari 400 had integrated graphics and audio. If you are going to mention memory types then why did you leave out the mistake of RDRAM? Also, it is not yet clear that AMD will ever adopt FBDIMM so this one doesn't count for anything. Why are you mentioning HyperThreading after Intel dropped it with C2D? Virtualization, instruction reordering, and Speedstep are good examples.
ho ho
I have no idea why you suddenly brought in the idea of being invented by AMD.
MCM was done by IBM well ahead of Intel. Obviously Intel didn't invent 300mm FABs or build their own FAB equipment; they bought them from third parties just like AMD and IBM did.
I would tend to give Intel credit for MCM with Pentium Pro which is when they first got a real server chip. Again, well ahead of AMD.
So, yes, you have to give AMD credit for X86-64, IOMMU, and designing for dual core. We'll have to give Intel credit though for being first to quad core, 80386 segmentation, and for the ATX standard.
Scientia from AMDZone said...
X86 is just an ISA; this is nothing to count as a first.
so do the x86-64 that AMD claims? or we can compare it to how intel extend the 8 to 16, then 16 to 32 too.
and SSE came after 3DNow!
you mention SSE instead of MMX just because it is later and the MMX is earlier? what an easy choice. You simply choose and pick (on the wrong one) to make up your statement, as usual. read this for yourself:
http://en.wikipedia.org/wiki/3DNow!
PCI took awhile to become better than VLB so being first wasn't much of a gain, same with PCI-e, USB and Bluetooth.
another fanboish attempt to demote whatever Intel has achieved. New technology is alway chicken and egg problem. Intel managed to drive them to become standard, it takes a lot of effort. And now we all enjoy the benefit of compatibility, including AMD, because if Intel. And i believe you would say and tone it differently for AMD's X86-64 right?
Scientia said
Integrated audio, graphics, and networking are a joke. Even the Atari 400 had integrated graphics and audio. If you are going to mention memory types then why did you leave out the mistake of RDRAM? Also, it is not yet clear that AMD will ever adopt FBDIMM so this one doesn't count for anything. Why are you mentioning HyperThreading after Intel dropped it with C2D? Virtualization, instruction reordering, and Speedstep are good examples.
the reason of them mentioning it because of those ppl hyping the fusion, which is an integration effort as well, at least at the 2 earlier stage. Some comment on the integrated audio though, it is again becoming another industry standard, for cheaper and compatible audio implementation.
intel didn't help develop the rdram as much as it drive the ddr and fbdimm statndards. you got the point now? please do not mix up the use of a technology and the creation/development of an technology. You can say good thing about AMD's ability to integrate the Immersion into its process technology, but not the creation part of it. Got it?
oh, btw Hyperthreading is not dropped at all. it will makes its come back in the form of SMT in Nehalem.
pointer
I can't believe you are trying to claim that the Pov-ray demo indicates that Penryn is faster than K10. In this demo, K10 is twice as fast with twice as many cores. Contrary to what you are claiming, this demo does not make use of K10's SSE capabilities. With SSE added in, the quad core K10 would be considerably faster. In fact, this is stated in the same presentation.
randy allen
"This is what Pat Gelsinger had to say about Barcelona:"
Why didn't you include an AMD rep's comments about Penryn? According to AMD, K10 will beat Penryn.
Scientia said: MCM was done by IBM well ahead of Intel. Obviously Intel didn't invent 300mm FABs or build their own FAB equipment; they bought them from third parties just like AMD and IBM did.
this is not of claiming the creadit of those tools, but the ability to integrate them into the manufacturing and have reasonable/good benefit of out it. And like what i said before, AMD can claims the ability to integrate the Immersion into its future process too (whether better or worse than the alternative way in terms of technology and economical return is another question, which we have no answer yet)
Scientia from AMDZone said...
pointer
I can't believe you are trying to claim that the Pov-ray demo indicates that Penryn is faster than K10. In this demo, K10 is twice as fast with twice as many cores. Contrary to what you are claiming, this demo does not make use of K10's SSE capabilities. With SSE added in, the quad core K10 would be considerably faster. In fact, this is stated in the same presentation.
when and where did i claim that? please do not put word in my mouth. What i said is that in this particular bench, Barcelona lose out to the current Core (not even Penryn). Did i ever generalize it as saying Barcelona lose out in all benches?
And it is funny that you try to point out that it didn't uses the SSE4a (as much as what AMD hint its system only has 6GB ram). The current core doesn't has SSE4 too. What is your point here?
Scientia quoted me, but failed to quote the part of my message that states that Intel has an additional two 300mm/45nm fabs coming online next year.
Scientia also only counts Intel's 65nm and 45nm fabs when making this flawed comparison. Intel has two 300mm/90nm fabs as well.
Of course, one of these is the New Mexico fab that will be upgraded to a 45nm facility. So by the end of 2008 Intel will have Four 45nm fabs, Four 65nm fabs and 1 90nm fab all with 300mm wafers.
Why didn't you include an AMD rep's comments about Penryn? According to AMD, K10 will beat Penryn.
When did anyone from AMD say that? AMD has said "Barcelona will blow Clovertown away." Sharikou quoted that on one of his postings.
scientia said ... PCI took awhile to become better than VLB so being first wasn't much of a gain, same with PCI-e, USB and Bluetooth. ...
i said ...New technology is alway chicken and egg problem. Intel managed to drive them to become standard, it takes a lot of effort. And now we all enjoy the benefit of compatibility, including AMD, because if Intel. ...
just wanna mention some of the innovations of those technologies here:
1) first and foremore, PNP, computing life become so much eaisier for the masses.
2) the technology lifespan is long ... they are designed with future needs (eg. bandwidth)in mind and sustain a reasonable life span usage, (this in turn promote the standardization and partly solving the chick and egg problem)
3) extensibility and backward compatibility (another thing that actually helps out the standardization beside the technology gain itself)
scientia
"I would indeed be surprised to see any volume of Penryn ahead of Barcelona."
I wasn't saying I'm sure of it, just that I wouldn't be surprised if Barcelona is yet again delayed.
"and SSE came after 3DNow!"
And Intel made MMX before both.
"same with PCI-e, USB and Bluetooth
Before USB and BT there were nothing that would be even remotely like those two. Against what are you comparing them? LPT, serial or something else?
"If you are going to mention memory types then why did you leave out the mistake of RDRAM?"
Why was that a mistake? Did it not provide much more memory bandwidth than anything else at the time of release? Was it Intels fault that the manufacturer didn't want to sell it cheaper?
Also I wish that PCs would start using XDR(2) again some day as it is capable of providing several times more bandwidth and even better latency than DDR.
"Also, it is not yet clear that AMD will ever adopt FBDIMM so this one doesn't count for anything"
It might not be FBDIMM but something similar will surely be used in the future. Two 64bit channels are not nearly enough to feed CPUs with ten times more computing power and those CPUs will be here rather soon. Dualchannel DDR3 can go up to around 35-40GiB/s but at least four times as much is needed, unless CPUs resort to using huge caches.
"Why are you mentioning HyperThreading after Intel dropped it with C2D?"
Why is that important that it is not there in Conroe? Didn't HT improve performance and GUI responsiveness on PC's? Is it not going to be in future Intel CPUs?
"I have no idea why you suddenly brought in the idea of being invented by AMD."
It would help if you'd actually had had my text there as I have no idea what you are talking about. Currently your text sounds as if I myself was invented by AMD :)
"So, yes, you have to give AMD credit for X86-64, IOMMU, and designing for dual core"
Why for the dualcore? It wasn't the first true dualcore.
"pointer
"or we can compare it to how intel extend the 8 to 16, then 16 to 32 too."
You missed the 4->8 bit ;)
scientia
"I can't believe you are trying to claim that the Pov-ray demo indicates that Penryn is faster than K10. In this demo, K10 is twice as fast with twice as many cores"
Unless Penryin is slower than Conroe in that benchmark then it should be quite safe to say it shows at least as good results as Conroe does and so far it seems as the results are considerably higher than Barcelonas.
"In fact, this is stated in the same presentation"
What presentation? Where can I read it?
"According to AMD, K10 will beat Penryn."
Isn't it so that according to Intel, Penryn will beat Barcelona? I wonder who to believe ...
Intel never "dropped" Hyper Threading. Tulsa has Hyper Threading. Hyper Threading will also feature in the Nehalem architecture.
Going with RDRAM was a sensible choice at the time. Nothing else could provide the bandwidth the Pentium 4 need to perform at it's best. When the Pentium 4 was first introduced DDR wasn't available, and the Pentium 4 performed awfully with PC133 SDRAM.
Lets also take a look at Sharikou's humorous list of thing that AMD apparently innovated and AMD copied. This should be good.
Intel does not innovate, it steals AMD's ideas: x86_64, dual core, performance per watt, spinning off flash division. The following are items Intel has not yet copied: IMC, HT, true quad core...
Dual core was first done by IBM back in 2001.
64bit is not new either, but we'll credit AMD with bringing it to x86. Yes, Intel did copy and add it to it's CPUs and renamed it EMT64.
Performance per watt is nothing new, that's been a focus of everyone for years. Think of the Pentium M back in '03.
I have no idea how one can consider spinning off part of your company and 'innovation', sufficed to say that AMD is not the first to spin off a part of a company.
Intel first did an IMC in Timna, though the product was canceled. The first shipping processor with an IMC was the Alpha processor, not any of AMD's CPUs.
HyperTransport was created by AMD. It's a replacement for the FSB. Intel has CSI which is also a replacement based on PCI-E I believe. Until Intel reveals more about CSI it's hard to make a direct comparison. AMD gets credit for creating the first replacement for the FSB.
Intel clearly beat AMD to the market with quad core CPUs. They had them shipping in November '06. Yes they are MCMs. Both Intel and AMD have "native" quad core CPUs under development so this is a silly statement from Sharikou.
So out of Sharikou's list, the two things I can find that have some merit are 64bit on x86 and Hyper Transport. The rest is utter stupidity.
giant
"HyperTransport was created by AMD"
Wasn't it just Alpha bus evolved? And what about HT Consortium?
Good point.
http://en.wikipedia.org/wiki/Hyper_Transport
It wasn't developed just by AMD but a whole group of companies.
Wikipedia is wrong. HyperTransport was not developed by a consortium.
DEC had the basics in the Alpha bus which you can clearly see in Athlon MP. However, both DEC and AMD wanted to extend this further. And, this can clearly be seen in the K8 and the EV-8. There is no doubt that AMD was working closely with DEC.
None of the technology for HT 1.0 came from the consortium partners. The consortium was merely a group that agreed on HT standards. However, it is possible that they have contributed to the technology since HT 1.0, particularly since DEC dropped out of development.
pointer
When properly compiled, the K10 Pov-ray score would be twice what was shown in the demo. The demo was not to demonstrate K10's SSE capabilities.
ho ho
Which is actually faster between Penryn and Barcelona? Barcelona is faster at the same clock. However, if Penryn clocks high enough it will be faster. There's no way to tell until we see what clocks will actually be released.
giant
"Scientia quoted me, but failed to quote the part of my message that states that Intel has an additional two 300mm/45nm fabs coming online next year."
List each 300mm FAB.
scientia
"When properly compiled, the K10 Pov-ray score would be twice what was shown in the demo"
How do you know this? Are you suggesting they used the x87 version of the benchmark instead of SSE?
"Which is actually faster between Penryn and Barcelona?"
Correct ansver would be "nobody* knows".
*) Besides AMD, Intel and selected few others who have done benchmarks.
"Barcelona is faster at the same clock"
In what kind of workloads?
"There's no way to tell until we see what clocks will actually be released."
You mean what would be the clock speeds at release or at, say, by H2 08?
List each 300mm FAB.
I was half way through a lengthy list when I realised Intel already has such a list:
http://www.intel.com/pressroom/kits/manufacturing/manufacturing_at_a_glance.pdf
It seems fairly up-to-date. The only information it's missing is that FAB11X in Rio Rancho, New Mexico is being upgraded to a 45nm fab by the second half of 2008 some time.
Scientia from AMDZone said...
pointer
When properly compiled, the K10 Pov-ray score would be twice what was shown in the demo. The demo was not to demonstrate K10's SSE capabilities.
i wonder where you get the idea that SSE4a would allow 2X improvement. Anyway, even if this is true, as i pointed out that the Intel used in comparison is actually current C2D. Thus, if Penryn type is used and with the 'proper' compilation that you said, it would have the similar improvement too as it has SSE4. The current and coming AMD CPU would lose out to Intel's CPU in this particular bench, base on the data that we have now.
remember, this data point is provided by AMD, and we have data points from intel and 3rd party bench that shows similar result in intel 2P platform.
Okay, these 300mm Intel FABs were online in 2005.
FAB 12
FAB 24-2
FAB 11X
FAB D1C
These were twice the capacity of a 200mm FAB, so:
4 X 2 = 8
AMD had only FAB 30.
So, this ratio was 8:1
Intel gets two more 45nm FABs this year which are double the 65nm capacity.
FAB 28
FAB 32
4 x 1 = 4 for the 65nm FABs.
2 X 2 = 4 for the 45nm FABs.
4 + 4 = 8
However, AMD will have two 300mm FABs in 2008.
So, the ratio is 8:2
This isn't an exact ratio but, again, there is no doubt that AMD has increased its production ratio against Intel.
pointer
"i wonder where you get the idea that SSE4a would allow 2X improvement."
You keep talking about SSE4a but I haven't mentioned it. And, obviously, since I haven't mentioned it, I have not stated that SSE4a is twice as fast as SSE3.
K10 is twice as fast at SSE as K8 but it isn't due to SSE4a. K10's SSE speed is due to having wider SSE execution units, wider buses, twice the prefetch width, and the fact that nearly all SSE instructions decode in a single clock whereas they take 2 clocks to decode on K8. There is no doubt that K10 is roughly twice as fast at SSE code as K8.
Your mistake is that you assume that the AMD Pov-ray demo used SSE when it did not. Since it didn't use SSE it does not show K10's SSE ability. If the Pov-ray demo had used SSE the K10 quad scores would have been 4X K8's dual scores rather than merely double.
Scientia from AMDZone said...
pointer
"i wonder where you get the idea that SSE4a would allow 2X improvement."
You keep talking about SSE4a but I haven't mentioned it. And, obviously, since I haven't mentioned it, I have not stated that SSE4a is twice as fast as SSE3.
K10 is twice as fast at SSE as K8 but it isn't due to SSE4a. K10's SSE speed is due to having wider SSE execution units, wider buses, twice the prefetch width, and the fact that nearly all SSE instructions decode in a single clock whereas they take 2 clocks to decode on K8. There is no doubt that K10 is roughly twice as fast at SSE code as K8.
Your mistake is that you assume that the AMD Pov-ray demo used SSE when it did not. Since it didn't use SSE it does not show K10's SSE ability. If the Pov-ray demo had used SSE the K10 quad scores would have been 4X K8's dual scores rather than merely double.
I fully understand that they used the same binary for K8 and Barcelona in that test, and thus, if the POV test is the standard one, it should have make use of SSE3. I do not aware that there is any indication saying that they do not use any SSE instruction at all.
care to share the link that indicates that AMD's POV bench test didn't use SSE at all? The only reason (hint) that AMD gave out so far is that they have a 6GB system (which is a lame one, as there is another test on intel 2P with 4GB scoring 47xx)
pointer
So, your argument is that in the absence of information you interpret the results in the worst possible way even when it contradicts previous information? That doesn't seem very logical.
I've seen examples of Pov-Ray compiled to use SSE or without SSE. Therefore, I don't of anything preventing the Pov-Ray code from being non-SSE.
You can read the K10 Optimization Guide yourself; it clearly shows that SSE instructions decode in 1 clock versus 2 on K8. AMD has also previously stated that SSE is 360% as fast for quad K10 as dual core K8. This would be 180% for dual core K10 or 80% faster on K10 than K8. Again, this matches the changes in the K10 architecture.
The logical assumption is that the Pov-ray demo did not use SSE and therefore shows no per core improvement over K8. Instead you assume that Pov-Ray is using SSE and AMD's previous 80% figure is a lie and the K10 Optimization Guide is a fabrication and the large architectural changes have given no improvement. Which is the simpler explanation?
Scientia from AMDZone said...
pointer
So, your argument is that in the absence of information you interpret the results in the worst possible way even when it contradicts previous information? That doesn't seem very logical.
and ......
The logical assumption is that the Pov-ray demo did not use SSE and therefore shows no per core improvement over K8.
and some earlier post: In fact, this is stated in the same presentation.
wow, a sharp turn from having the information in the presentation to now your own logical assumption.
if this is the case, AMD would explain it out loud, instead of giving the excuse that their system only have 6GB instead. Is this logical enough for you? Don't be too desperate defending on this bench that AMD lose out, it is only one bench that we know of, there are others that we are yet to know.
If the Pov-ray demo had used SSE the K10 quad scores would have been 4X K8's dual scores rather than merely double.
Stupid, stupid, stupid. If a program were to include SSE, and a certain CPU has 2x the SSE performance, it will not exhibit 2x overall performance.
I did some simple benchmarks with 3.6 series of Povray yesterday and found out that -mfpmath=387 vs -mfpmath=sse have less than 1% effect on overall speed on my Core2. I'm not saying those are the final results and they are quite definitely not comparable with 3.7 series but it still makes me wonder if only SSE vs x87 has such a big difference in speed.
Since yesterday 3.7 series Povray beta became acailiable on Linux also but I haven't done any real benchmarking with it yet. Though I doubt I can make any meaningful benchmarks as there are only precompiled executables and I'm not sure what kind of optimization flags they used.
I'll try to make some further benchmarks when I get time.
scientia
"So, your argument is that in the absence of information you interpret the results in the worst possible way even when it contradicts previous information?"
So you on the other hand assume best possible way? Btw, what information does this contradict with? I've still not seen any official explanation of the thing.
he
". If a program were to include SSE, and a certain CPU has 2x the SSE performance, it will not exhibit 2x overall performance."
You missed the additional two cores
ho ho said
he
". If a program were to include SSE, and a certain CPU has 2x the SSE performance, it will not exhibit 2x overall performance."
You missed the additional two cores
no, he didn't miss it. He is argue along the line of per core basis, which is Scientia's arguement.
Scientia said 2x performance improvement per core and 4x for the entire system.
I was pulling 2x SSE and 2x performance out of thin air just for kicks. Scientia insists that not only will double the cores double the performance, but the SSE enhancements in Barcelona would also double the scores for overall 4x. Of course when we compare Yonah and Merom scores, we see that benchmark scores are not dictated by SSE performance of CPUs.
Post a Comment
<< Home