AMD 8-Core to slaughter Intel
AMD is using two K10 quad core and HT3 to produce a 8-core monster.
I suggested that AMD should bundle two Opteron dual core with HT long ago.
Some Opteron boards work just like that, but they are using two sockets.
Now AMD is finally doing the cheap trick -- connecting a K10 quad to another with HT3, the other quad has direct access to memory.
What can Intel do? Packing 4 dual cores on the FSB? Yeah, you get 4 Conroes hanging on the bus, each core getting about 100MHZ.
posted by Sharikou, Ph. D. @ 4/02/2007 09:51:00 AM
35 Comments:
As shown by AMD's stock skyrocketing today, all of the Analysts love this idea and think it is very compelling.
Scott
Thats funny.
Sharikou
Couldn't that be an April Fools?
Why wait for HT3?
It was just a matter of time before is was proven scientifically...
Intel Users Less Intelligent Than AMD
I knew it all along, but that's only because I'm smart :)
Jeach!
Not to burst your bubble, but there is nothing definitive in your link that could substantiate the articles claims (source, number of Intel/AMD users of the 5,000, ages...).
Still feel smart?... Just kidding :)
"Intel Users Less Intelligent Than AMD "
I didnt need a survey to tell me that ....only fools fall for the marketing. The "Intel inside" program is always cited as a brilliant marketing move, But I think it was kick start to a monopoly business, any AMD lawyer reading this ?? :-))
sharikou
"Yeah, you get 4 Conroes hanging on the bus, each core getting about 100MHZ."
It would be actually 1600/8=200MHz for the new 45nm CPUs :)
Also with dualchannel DDR2 800 AMD "octacore" would have 800x2/8=200MHz of bandwidth per core, though it would be a bit more effective than Intel, the difference isn't too big. Interesting, isn't it?
scott
"As shown by AMD's stock skyrocketing today"
It skyrocketed? Sky is upwards, you know. Going from 13.1 to 12.8 is not skyrocketing.
jeach
"Intel Users Less Intelligent Than AMD"
Who buy AMD? Mostly people who know a bit about computers and CPUs. It is only logical that they can do things like move pictures from camera to HDD better than Intel users.
"Yeah, you get 4 Conroes hanging on the bus, each core getting about 100MHZ."
It would be actually 1600/8=200MHz for the new 45nm CPUs :)
But you have to consider the effect of bus collision. Eight cores fighting, the effective bus speed will be about 1000 even at 45nm--which no one has seen.
I do not know if it will kill or not intel, but just remmbr, for AMD there is a tradeoff in usng MCM. If they use MCM the pcocessor either half of the cores lose the advantage of the on-die memory controler, or the socket has to be different, and therefore they lose the drop-in-replacement market. Due to their use of a FSB (granted, a technicaly inferior solution) this is not a problem for intel....
oh, tradeoffs, tradeoffs.....
sharikou
"But you have to consider the effect of bus collision."
It is exactly the same for AMD. As a "bonus" one CPU will also load the IMC of the one that is directly connected to the RAM. Thus when your program runs on one CPU it will be faster than running on the other one. With Intel it would be at equall (slow?) speed.
"Eight cores fighting, the effective bus speed will be about 1000 even at 45nm"
And how good might AMD's memory efficiency be? For single socket single CPU that was around 80% IIRC.
Btw, didn't you once say that for CPU core to work efficiently it would need at least 266 MHz bus to RAM?
"which no one has seen"
Intel has publically demonstrated its 45nm quads running at 3.2GHz. Where are the images from Barcelona? Shouldn't it become availiable very soon? Kind of makes me wonder if they are afraid to show them ...
"But you have to consider the effect of bus collision."
It is exactly the same for AMD. As a "bonus" one CPU will also load the IMC of the one that is directly connected to the RAM.
You are missing the main item--cache coherence traffic. The 8 cores must use bus the send cache coherence traffic.
ho ho:
I guess you did not pick up on the sarcasm. At least enumae did...
sharikou
"You are missing the main item--cache coherence traffic"
Exactly how much of it there is? I can't remember 4x4 beating QX6700, even though it basically had a whole bus dedicated for synchronizing caches
Hey, now I know why OCZ has come out with 1000 watt power supplies!
Eight AMD cores with an AMD graphics chip will need that much power.
Hey, now I know why OCZ has come out with 1000 watt power supplies!
Eight AMD cores with an AMD graphics chip will need that much power.
AMD uses an MCM? How could they EVER justify that after continually whining that Clovertown "is not native quad core"?!
Henri Richard now:- "Barcelona is much better than clovertown! Clovertown isn't even a native quad core!"
Henri Richard when this comes out:- "Forget that crap I said earlier. Using an MCM is a good way to do things. Sorry Intel, I take all that back!"
Crazy.
I don't think they would take back what they said about Intel's MCM, they'll just say their way of doing things is less evil because in their case the signal doesn't have to leave the chip.
How could they EVER justify that after continually whining that Clovertown "is not native quad core"?!
Maybe by having one eight core processor that your main competitor doesn’t have?
And can’t do it, leaving it behind in year(s)?
Non native quad core processor = not being able to do one non native 8 core.
2x2 OK
2x4 OK
4X2 not possible. Intel V8 to be announced soon?
I’m posting this but I really don’t care, I still prefer cheap single core processors.
aguia
"Non native quad core processor = not being able to do one non native 8 core."
What technical reasons stop Intel from putting together four dualcores instead of only two? Certainly not power, it has 50W 65nm quadcores, putting two of those together will be at around 100W or less.
Also there shouldn't be connection problems as first dualcores worked just fine with older boards and there were no extra pins added.
Sure, FSB will get crowded but it won't be much better for AMD either. Also I wouldn't be surprised to see >2GHz FSB with those theoretical octacores.
You can say that there will be too much cache synchronizing traffic there and with some programs it might be true. Though those programs won't scale well to so many cores anyway, with Intel they would simply scale less well.
"Intel V8 to be announced soon?"
It is called "MacPro" ;)
Though I do wonder if we would have 4x16 or something. That would be a monster ray tracing platform for sure :)
I think that video AMD Interview Sharikou pointed us to a while ago, talking about the Native Quad Core hoopla, the AMD guy mentioned that they wish they had just glued two Dual-Cores together to get a Quad core out there quickly, just like Intel.
So, sure they took the high-horse by claiming TRUE Quad-core is the real shizzel, but they wish they did the fake one for the cashola...
Quad Cores...
Octa Cores...
Either way... Santa is bringing me one hell-of-a gaming rig, just in time for Crysis.
Intel will not use the FSB for an Oct-Core CPU. Intel's first processor with eight cores will be based on Nehalem, using CSI.
I guess you did not pick up on the sarcasm. At least enumae did...
He seems to be an Inteler. Go figure.
symbiansn
He seems to be an Inteler. Go figure.
Am I an Inteler because I like facts and sources?
they took the high-horse by claiming TRUE Quad-core is the real shizzel, but they wish they did the fake one for the cashola...
Yep, I would agree that they've made a mistake. I doubt many will admit it though. AMD can't afford this kind of mistake in the future.
Who buy AMD? Mostly people who know a bit about computers and CPUs.
Most of the top-end open source developers use AMD processors (such as Torvalds, the linux kernel inventor/maintainer).
For the last 15 years, some of the best engineers I've worked with use mostly AMD. Techs and system admins tended to use mostly Intel. But then again, maybe that's why they never made it as engineers?
Isn't Intel going the 4-core, 2-thread-per-core way? This won't even be able to touch an 8-core processor. Actually it's probably the only reason they are bringing back HT, not to look too distant from AMD.
I hope AMD does HT also, as to market it as 16-thread processor.
But then Intel would have done 4-HT/core... so I guess there is just no wining in their "I'm the best" marketing.
Too bad it stops at the marketing and it doesn't make it to reality!
jeach said For the last 15 years, some of the best engineers I've worked with use mostly AMD. Techs and system admins tended to use mostly Intel. But then again, maybe that's why they never made it as engineers?
oic, that's the reason you are using Intel CPU.
jeach
"Most of the top-end open source developers use AMD processors (such as Torvalds, the linux kernel inventor/maintainer)."
Have you got any ideas what kind of CPU is he using? If it isn't the newest one then I think it might be because he bought it before C2D, Netburst wasn't a good competition back then. If it is one of the first K8's then it is possible he just wanted to get a cheap 64bit CPU to play with.
"For the last 15 years, some of the best engineers I've worked with use mostly AMD."
Interesting. IIRC AMD became competitive around K6/K7, a lot later than 15 years ago. Before that it just had cheaper and mostly slower CPUs.
"Isn't Intel going the 4-core, 2-thread-per-core way?"
It is but not before the end of 2008 with its new 1-8 core (native?) CPU.
"This won't even be able to touch an 8-core processor"
It will have up to 8 cores on new second generation 45nm microarchitecture against AMD's third gen 65/first gen 45nm it not only touches but pretty much crushes it.
"Actually it's probably the only reason they are bringing back HT, not to look too distant from AMD."
Remember when I talked about RHT I said how long are the memory latencies and how difficult it is to feed all the execution units? Well, HT can feed those idleing execution units and use the full core while the other thread is waiting for recources. It was quite effective even on P4 giving it around 10-30% performance boost. I wouldn't be surprised to see much bigger gains on the new core.
That new core was designed long before AMD even started talking about its MCM octacore, it has nothing to do with "me too".
"I hope AMD does HT also, as to market it as 16-thread processor."
Some said it will probably do reverse HT, I wonder what people think now ...
If anything then I'd be worried about AMD and their octacore. To me it is pretty much a clear sign that having only four cores won't cut it against Intel.
enumae
I wasn't bitching on you but off the other guy who didn't understand scott's sarcasm...
Guess who he is.
It will have up to 8 cores on new second generation 45nm microarchitecture against AMD's third gen 65/first gen 45nm it not only touches but pretty much crushes it.
K10 will set a new architectural standard. Intel talks about radix-16, power states, HT and another goodies while AMD hints at "various little tweaks that will dramatically improve processor performance as a whole". Amd will bring into its table every improvement made into either Alpha 21264, P6 and Core 2 and in a better shape thanks to its geniuses.
K10 will be GOD of all processors. AMD's ready to bitchslap Intel even though they're one generation behind in terms of process.
Notice also that Vista was built with K8 in mind. Have you seen the new 64-bit benchmarks where Core 2's tiny 10% advantage turns into vapour?
Because Core 2 is P6-based, it's wonky at 64-bit and thus its performance doesn't improve well with Vista enhancements. Intel needs to fix that urgently. They can't survive forever without an IMC too.
Intel fixes, stopgaps, AMD improves (and sues too)
Intel BK 2Q'08
symbiansn
"K10 will set a new architectural standard. Intel talks about radix-16, power states, HT and another goodies while AMD hints at "various little tweaks that will dramatically improve processor performance as a whole"."
So basically Intel sais exactly what it does whereas AMD only hints?
"K10 will be GOD of all processors."
Then let us hope that God won't have smaller performance than >3GHz Intel does.
"Have you seen the new 64-bit benchmarks where Core 2's tiny 10% advantage turns into vapour? "
I haven't seen anything revolutionaly. Can you link to those benchmarks you talk about?
ho ho
Intel's now setting its AMD64 powerhouses for redeeming from the Netbust mistake, however, unfortunately doesn't mean that it'll really show off against AMD.
Intel's pretty much like an engine manufacturer who has close-to-infinite resources but drains'em off pitily, thanks to uncommitment and series of "blah's"
I really trust AMD. AMD's like a makers with a committed team who really knows what they're doing and thus have the hand to set their own impeccable high standards to new heights.
It's AMD against itself here, in terms of engineering skills. Intel's minds are bright, contrary to popular belief but just can't work together as well as AMD's.
AMD will bring to K10 everything K8 lacked: lower instruction latencies, Penryn-like improvements, efficient OOO scheduler algorithms, Itanium-like floating-point, extra memory buffers, etc. They can't say it all because the list would be too much for most websites' limited format and for the AMD criticizers' limited minds to absorb at once. K10 will need part1-2-&-3 to get it explained. It's no secret for we computer enthusiasts who're used to AMD's long run over Intel year after year.
Intel is a worthy opponent and a threat to AMD in terms of marketing, financial resources and legal battles only. In a few months I'll prove myself right.
Their low-k 45nm process is pretty cool, though
Vista benches:
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/30579_3_74.pdf
http://www.tomshardware.com/2007/01/29/xp-vs-vista/page4.html
Intel BK 2Q'08
AMD is running out of cash. No one wants to give them more. They are in serious strife.
Have you seen the new 64-bit benchmarks where Core 2's tiny 10% advantage turns into vapour
Where is that? First 4x4 comes out, people claim wait for 64bit and NUMA support in Vista. We waited, it still gets soundly fragged by the QX6700. In fact, the results show that one 2.4Ghz CPU from Intel is enough to frag TWO 3GHZ CPUs from AMD. Whilst using less than half the power. That's real value.
http://techreport.com/reviews/2007q1/quad-core/index.x?pg=1
AMD is losing marketshare in all segments. No one wants Crapteron processors. It's dual core CPUs use more power than Intel's quad cores! AMD Turdion processors are nice and slow as well, far inferior to Core 2 Duo.
K10 will be too little too late. K10 won't even make up the massive performance lead Intel has now. Penryn will come and spoil the K10 launch, devaluing AMD's stock further. Nehalem will be the final nail in the coffin for AMD.
AMD BK Q2'08.
Yeah, Intel always get the upside, and that's way the research says Intellers are dumber than AMDers, cause Intellers get "royalty" and says fanboyish talk all over the Internet, without making any sense as "Intel rulz, and pwned AMD"!!
BTW, HT3 operates over 2.0 GHz mark, for that number, it provides well over 32.0 GB/s bandwidth (based on the 2.6 GHz HT3 for 41.8 GB/s bandwidth), four cores share interconnected using ccHT links inside the cores, and the two dies share one HT3 link, and one to the motherboard, another one to the other processors (as in Quad FX platform) or co-processors, a total of 3 HT3 links! Is that enough compared to that of "1333 MHz FSB" with bandwidth slightly passing through the 10 GB/s mark and share between two dual-die quad-core processors!? Sheesh!
Yeah, right, the memory is not enough, but guess what, each die has its own memory controller to feed with, okay, four dies and four memory controller and a total of eight slots (4 slots for each memory controller, maximum of 8 due to motherboard spacing problem - you cannot put 16 DIMM slots on a desktop board, can you?), enough!? Oh, I'm sorry that I forgot the NUMA problem, and slows down a lot, in a few tens or hundred of cycles......
And do somebody actually compares a thing to another thing, okay, it's the same to compare K8 to Conroe, while K8 was out, Intel fanboys do tell that Conroe beats everything K8 has, okay exactly how can you compare an old thing to a new thing when the technology keeps improving!? Do tell me to compare the MCM octal-core Montreal in Quad FX to V8 with Quad-core Clovertown. Hotter, louder but faster? Yeah right! Just gimme 2 Montreal+nForce 680a MB+4 GB DDR2-667+3 R600 crossfire bundle for a cheap price and I'm sure I'll grab one soon, as "People buy holes, not drills". :)
And exactly how much of the proportions of Core 2 Duos are being manufactured compared to the old crap of Pentium inventories? AMD cut prices to prevent those K8/K9 (dual-core K8 are K9) from stacking up in the inventory, so that an 3600+ matches PD 820 in performance and in price. But what about Intel, now everybody wants a Core 2 Duo/Quad, and nobody wants PD, oh well, dump all of them to the third world!! What a shame! And with AMD producing 100% 65 nm parts in one of the fab plants, that means the yields are satisfactory, and we'll hopefully see Q2-Q3 for the K10/K8L quad-cores later. So it is a good move for AMD to cut price like crazy in Q1 and Q2.
Do not forget that, most (more or less 80%) of the revenue came from niche markets (around 20% of all products) with higher margins, such as high-end servers for a semiconductor company, 4-way or more is very profitable and is less available to general public, and with AMD's Opteron beating up Intel's NetBurst Xeon inventories, and not enough capacity for Clovertowns, that's okay enough for AMD to cut prices like crazy in the deaktop market, who cares if AMD is posting a lost in the desktop market? even Hector doesn't care about that... :D
And oh, it seems that it's high time to buy AMD shares, and dump Intel's. With AMD has profits with K10/K8L and Radeon R600 (nice profit from gamers :p), and "Bacelona" and "Shanghai" later in 07.
Okay now, server market loyalty to AMD, and some more gaming systems using R600, home theatre PCs using AMD Live! and notebooks such as HP IQ770 with Turion X2 and HP tx1000, AMD has a bright future.
It seems Intel has nothing to do except touting the "Penryn" with HyperThreading (Pseudo-multithreading) and the no-technical-information-available CSI (okay, the "Common System Interface" thingy, not the drama) giving a 6400 MT/s or 4800 GT/s numbers and nothing else hoping to kick HyperTransport out of the market... and oh, the "Centrino Pro" with iAMT and fatures for management, and the "Centrino Duo", both having "Turbo Cache" support, as well as the copycat move Geneseo interconnect plan. So what exactly Intel wants to copy further? Maybe they're discussing behind the doors now. *smile*
What does AMD have, Torrenza for HTX and socket co-processors, Fusion (as also copied by Intel) as GCPU/CGPU (i don't know), AMD Live! makes a complete HTPC ecosystem after CeBIT and CES, Apple iMAC-ish products ideas support from HP (IQ770 TouchSmart PC - do not forget the home integration thing in mind - and tx1000 notebooks), and complete GPU development team for discrete graphics...
You see the differences between the two companies? It is the smaller companies get the concepts first than the so called "giants" of the industry in the recent years, so where's the innovation of Intel's came from? COPIED, granted. Examples? Torrenza to Geneseo, QuadFX to (non-Multi GPU) V8 for Mac Pro only (yeah, the only customer adopting the multi-CPU workstations :p), Radeon line to "Larrabee" GPU (featuring DirectX 11.0 support with ray-tracing), AMD Live! Ready to Intel Viiv ("ICHxDH" digital home southbridge and some difficult technical stuff not known to the general public), CSI to HyperTransport.
What left? NAND and NOR businesses?
AMD has spun that off a long time ago, it's named "Spansion" and is world leading NOR memory supplier for handheld devices and read-intensive devices, oops surpassed Intel, too. And the UMPC CPU? Oh, that was fresh, follwing the steps of C7-M, innovation? Nah, shame again. P6 architecture, the only hope for Intel, still copied, also known as the Integrated Menory Controller and point-to-point interconnect (aka CSI).
The big picture is that AMD and other firms has the innovation recent years, and Intel saw the potential, and moved the ideas into its own products using a different way (patents!), and become a copycat.
And you guys praise that copycat with no innovation at all!? Shame on all of you!
You say that it's unfair to compare Conroe to K8? But then you go and compare products that aren't even released yet, and won't be coming until next year, to products out now? LOL!
Lets look at AMD's product lineup shall we? We'll start with CPUs.
Desktop: AMD has dual core only using more power while being slower than Core 2 Duo. Quad core? Not a chance.
Servers: It's the same thing again! Dual cores using more power than Intel's quad cores. Wow! No one wonder AMD's marketshare is eroding so quickly! Who wants an AMD 4P server when an Intel 2P quad-core server has similar performance for a fraction of the cost?!
Mobile: Turdion 64 gets eaten alive by the ancient Core Duo. Core 2 Duo is just the icing on the cake.
And we all know in GPUs AMD is so far behind they'll never catch up. R600 is delayed time and time again. Nvidia has the 8800 Ultra ready to frag AMD all over again.
Oh wait... I forgot! AMD has the 4x4! FX-74 can't even match the Q6600 at a mere 2.4Ghz, even with a 600mhz clockspeed advantage!
AMD is losing money, losing marketshare and is about to crumble. Look at that stockprice! Headed down towards $12! AMD is valued at only $7bn now. Barcelona won't even have time to ramp up before AMD BKs in Q2'08. Intel makes billions of dollars, AMD is losing billions!
symbiansn
"http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/30579_3_74.pdf"
Interesting.
Overall performance, 32bit Vista, compared to FX62:
FX62, FX70, FX72, FX74
100%, 141.7%, 150.7%, 160%.
"Overall performance, 32bit XP Pro
100%, 155.4%, 166.8%, 176.1%.
So does that mean XP Pro has considerably better NUMA support? As it comes straight from AMD I guess 4x4 indeed scales better under XP than in NUMA aware Vista. Thanks for clearing that up.
ck
"Pseudo-multithreading"
You, sir, seem to have no idea what is multithreading and what is hyper threading.
"Fusion (as also copied by Intel)"
Intel was researching it in 2000, if anything then AMD copied them.
The new updated EIGHT CORE Mac Pro! That's eight 3Ghz processing cores all in one workstation!
Ladies and gentleman, we can officially consider the 4x4 concept totally trashed! This thing would just eat an AMD 2P workstaton for breakfast.
CK, thanks for trotting out the old falsehood about Intel "copying" AMD on Fusion (and IMC for that matter). It is patently false. Please do your homework before stating this as fact.
Intel saw it first, productized it first, and then had to kill it because it was foolishly wedded to Rambus, which made the platform too expensive for the targeted value market after the memory translator hub approach prevented it from working with standard DRAM. Did it ever hot the market- no. Was it taped out, debugged, and ready to ship when it was killed? Yes.
What was it? Timna- circa 2000, which meant design likely started in the 1998 (or earlier) timeframe. It was a Pentium III, with onboard graphics and northbridge, built on the .18um process. Sound a bit like Fusion? No, the graphics were not "leading edge", even at the time- but high performance users will always opt for discrete graphics.
Face it- it doesn't matter who thought of it first- it's who gets the customers what they want/need AND manage to make enough money doing so that they can fund the next round of development. Right now, AMD is failing miserably in that department- they are slipping schedules, lagging in performance, and not making money. I'm not happy about it- I want to see AMD healthy. But management seems bound and determined to keep shooting themselves in the foot until and blaming Intel for selling them the ammo.
Post a Comment
<< Home