Monday, November 13, 2006

Hector and the one

See this video. I never thought Hector can act.

AMD took the 2nd spot in super computer top 500. Intel has no machine in the top 5 and is being pushed to the ultra low end. Cray launched two super computer models equipped with AMD Opteron. These machines use HyperTransport as interconnect and are petaflop capable. The Cray XMT can run 1 million threads concurrently.

53 Comments:

Anonymous Anonymous said...

That's hillarious. I love it!

6:57 PM, November 13, 2006  
Blogger Unknown said...

If only it was 1999 this wouldn't have been so cheesy.

8:29 PM, November 13, 2006  
Anonymous Anonymous said...

About all Hector can do these days is act. His compnay can't seem to make CPUs worth a damm, nor make money long-term.. at least he has a career as a actor or gardner..

8:48 PM, November 13, 2006  
Anonymous Anonymous said...

Appro International, a systems maker based in Milpitas, Calif., is announcing Nov. 13 that it has delivered two supercomputer Linux clusters to the Lawrence Livermore National Laboratory. In building these four interconnected clusters, Appro used Advanced Micro Devices' Opteron 8000 series processor.
Another notch for AMD!!!!! Go AMD

8:56 PM, November 13, 2006  
Blogger Christian Jean said...

Great? And what purpose does it really have? About 90% of the people who will view this will already be AMD backers anyway.

I'm one of the biggest AMD fan, follower and supporter out there, but I'm really starting to loose patience.

Although both AMD and Intel are growing, AMD is not going for the kill... all these products coming late and/or after Intel.

Why so late with quad-cores (native or glued)?

Why after Intel's quads (regardless if it's two dies)?

Why so late with K8L (Revision H) processors (by mid 2007)? After 3 years, these revisions should already be available!!

Rule of thumb is "no news is good news". But right now its driving me F@#$% crazy!!!

Intel is pissed and they are definatly not sleeping right now! I'm sure that by early to mid 2008 when AMD will have their new products out for only 6 to 9 months, Intel will release their new architecture.

Jeach!

10:06 PM, November 13, 2006  
Anonymous Anonymous said...

This is soo stupid instead of making processors AMD is resorting to making cheap movies to sell their underperforming processors. Thank you sharikou for posting this to show what AMD has really been up to instead of designing K8l which should have been out by now.

12:02 AM, November 14, 2006  
Anonymous Anonymous said...

All this marketting is good for AMD. Link it to de.licio.us so that everyman and his dog will see it. This kind of marketting is exactly what AMD needs, yes it's cheesy but it's also the kind of thing that will get plenty of attention and open peoples eyes to the world of AMD. Showing them that its not just Intel inside!!

3:48 AM, November 14, 2006  
Blogger 180 Sharikou said...

Doctor - for once you're right. Hector cannot act.

3:51 AM, November 14, 2006  
Blogger enumae said...

Off topic, but an interesting watch.

-[link]-

4:59 AM, November 14, 2006  
Anonymous Anonymous said...

Why so late with quad-cores (native or glued)?
Because AMD doesn't like non-native solutions. But hey, they do have the 4x4 which is "2 native dual cores" combined.

Why so late with K8L?
Caught pants down with Core2Duo performance leap. The added delay when Core2 reviews came out earlier in the year suggest a re-design.

"no news is good news".
That's only true if you're ahead.

I don't think AMD will ever get the performance crown back. For AMD, it's just a matter of whether their top binning extreme CPU competes with Intels 3rd or 4th binning parts.
They need to catch up if they don't want to compete with Celerons. It must be hard to be an AMD fan right now.

5:42 AM, November 14, 2006  
Anonymous Anonymous said...

Hehe, it's nice that Hector can act as well as do a hell of an amazing job with a underdog company like AMD. He's got a good sense of humor, and I get the feeling he's a great person.

Too bad AMD processors now suck badly.

7:23 AM, November 14, 2006  
Anonymous Anonymous said...

As much as I hate and criticize the information you post, I must admit, I thoroughly enjoyed this video. Thanks!

7:26 AM, November 14, 2006  
Anonymous Anonymous said...

Haha, that was great. Hilarious, but very cool. Nice to know AMD still has somewhat of it's sense of humor/sarcasm, even if that is not thier intent.

While I can understand your sentiment Jeach, first does not always mean best. During the K8 era AMD was well on top and making money as quickly and efficently as possible making updates to the K8 while working on the next gen.(K8L, aka K9/K10 at the time) Unfortunately good things take time and the development of the K8 was about near the same about of time, 3 years. Do it right and do it the first time will keep you from having to pump out revisions just to keep up performance wise with your competition.(*cough* P4)

Make no mistake, from all the data I've seen reguarding it's design, K8L is going to take the Core Duo 2 out back and give a whooping old school style. 2x performance over the current K8 in SIMD and FP with a healthy x86-64 performance lead vs Intel in equal mhz ratings will cement this fact. Now this isn't going to happen as fast as some of us would like, AMD is moving as fast as humanly possible to do so, without any mistakes I'm sure.(as they can't afford any)

For Intel, the early bird may get the worm, but the second mouse always gets the cheese. Intel has already blown it's collective wad with Core 2 and K8L hasn't even shown up to the fight yet. What is going on right now highly reminds me of right before the K8 vs P4 era all over again. Intel will be spending the next 2 years trying to regain the performance crown from the K8L once released. Why do I say this? It's because while Intel has promised a 2 year design cycle, AMD is moving to a modular design with continual improvements, a literal continuing work in progress, there will be no more "Kx" generations after the K8L. Major improvements will be made as needed and dynamically from one revision to the next.

8:13 AM, November 14, 2006  
Blogger Sharikou, Ph. D. said...

While I can understand your sentiment Jeach, first does not always mean best.

Yes. To punch hard, you have to draw back your fist. AMD has signed up all major OEMs, including DELL. All AMD needs now is ramping up capacity. Once K8L is launched, Intel fate will be sealed. Intel's BK is projected at 2Q08, but it can happen faster.

In the past, Intel has DELL to ensure 20% of market share, now even that insurance is gone.

8:19 AM, November 14, 2006  
Anonymous Anonymous said...

As I expected Intel cronies not able to understand what is really happing here. Without scaleable bandwidth new virtualization techniques won’t fit in Intel’s thin pipe. The collected brain trust of AMD’s partners and the hyper transport technology have already rendered Intel product lines useless in the virtualization arena until 2009.

Believe it or not Intel is still trying to perfect x86 64bit computing. Intel can still do well in the mom/pop business area until then.

The astronomical power savings in the data center with virtualization is the key. If Intel can’t scale then they can’t go virtual and without virtual you can’t save power. It’s as simple as that.

You still don’t believe me? Look at the accelerating market share gains in super computing arena. The reason is the same, HT. Super computers & Virtualization need more than just large cache.

Btw, I’m truly sorry you were all laid-off at Intel.

8:35 AM, November 14, 2006  
Anonymous Anonymous said...

"Why so late with quad-cores (native or glued)?
Because AMD doesn't like non-native solutions. But hey, they do have the 4x4 which is "2 native dual cores" combined."


No, the correct answer is because AMD could not do non-native solution economically. They have shortage on dual-core chips already, how are they going to make double dual-core chips?

I have to say AMD's executives were right about not releasing double dual-core CPUs.


"Why so late with K8L?
Caught pants down with Core2Duo performance leap. The added delay when Core2 reviews came out earlier in the year suggest a re-design."


Wrong again. Design changes in K8 rev.H (which I suppose is the "K8L") - native quad-core, L3 cache, improvements everywhere, does not finish in 3 months like your college term project. They most likely take >1 year to design implement, simulate, and test. By the time any Core 2's results was out in February, rev.H design was already fixed.

The true reason of rev.H "delay", IMO, is that AMD had only one major team to do CPU design.


"I don't think AMD will ever get the performance crown back. For AMD, it's just a matter of whether their top binning extreme CPU competes with Intels 3rd or 4th binning parts."

For SuperPi, you may be rigt. For wmv/ogg transcoding compiled by icc with SSE support, you may be right. But for some applications, such as memory-intensive, cryptographic, server and HPC, K8 still has the best performance right now.

9:02 AM, November 14, 2006  
Anonymous Anonymous said...

Btw, I’m truly sorry you were all laid-off at Intel.
Laid off? No...
Fired for systems abuse!!

9:23 AM, November 14, 2006  
Anonymous Anonymous said...

As previously posted, AMD has grown 100% year over year on the super computer list. This is just the beginning. HT is a new technology and with in a few years expect Intel to be eliminated from the list altogether. Why, you may ask? HT has dramatically changed the cost and the time it takes to build a super computer.

Have you noticed all the Intel server commercial lately. 99.9% of the people that view these commercial have never even seen a server room, except for on TV. Intel knows they have a big problem. Use your imagination to fill in the rest of the story. Be honest.

9:43 AM, November 14, 2006  
Anonymous Anonymous said...

While Intel is comfortable with its advantage in some aspects of modern computing in general, this topic of virtualization and super computers is a huge problem for Intel. A part of Intel’s DNA is a healthy competitive spirit and this must piss them off to the max, because they have no answer. Global enterprise is the holy grail of computing.

If they can’t even solve there graphic issues they’ll never be able to climb the virtualization mountain and forget about super computers.

10:32 AM, November 14, 2006  
Blogger Christian Jean said...

Although the 4x4 is technically similar to Intel's quad, it's not marketed and sold to compete agains Intel's quad product. But in the meantime, Intel's quad will sell like crazy while AMD will sit and watch for a while.

I also understand that the K8L will eventually arive and will be faster than the current C2D. But what prevents -- and this may be lack of in-depth understanding processor development on my part -- Intel from implementing those same exact features into the C2D right NOW (as we speak)!

Look at the expected K8L features:

Of all those, how long on average would it take to implement within an existing processor? To me, it doesn't seem like it could take Intel more than 1 year, which means that they could have a C2D revision H (for example) by the end of 2007, giving AMD a lead for only 6 months.

Jeach!

10:43 AM, November 14, 2006  
Anonymous Anonymous said...

Hmmm.. Is this another failure for AMD initiatives? Sorry dude, Intel is not the only one that makes mistakes!

"http://news.com.com/AMD+cans+low-cost+PC+project/2100-1003_3-6134891.html"

10:45 AM, November 14, 2006  
Anonymous Anonymous said...

When you run out of good products, what do you do?

Market nothing!

It is nice to see that AMD is following Intel Marketing..

Intel fired half of their marketeers though! Watch out Hector!

10:52 AM, November 14, 2006  
Anonymous Anonymous said...

Hmmm.. Is this another failure for AMD initiatives? Sorry dude, Intel is not the only one that makes mistakes!

"http://news.com.com/AMD+cans+low-cost+PC+project/2100-1003_3-6134891.html"


The initiative did not foresee cheap $100 notebooks. I don't think that means a "mistake" per se, but rather an initiative that was not bold enough.

11:28 AM, November 14, 2006  
Anonymous Anonymous said...

Intel has already blown it's collective wad with Core 2 and K8L hasn't even shown up to the fight yet.
They still haven't played the CSI/IMC/dual socket cards yet. And still on top where it counts. What's AMD got left to shoot its wad?

11:54 AM, November 14, 2006  
Blogger Sharikou, Ph. D. said...

Intel fired half of their marketeers though!

Intel's fate is sealed. BK is inevitable. All you see from Intel now is the painful struggle before the eventual death. Even if Intel produces a processor that is 4x faster, it doesn't matter -- because 90% of its production are old junk. AMD can comfortably sell CPUs at $90 a pop and ramp production along the way. Then Rev H comes on 65nm, all Intel processors are made obsolete. AMD's advantage is that it can switch to new design very quickly. By 3Q07, 85% of AMD processors are Rev H based. FAB30 will be ramped down and become FAB38 in 1Q08. Intel's BK is set at 2Q08.

12:06 PM, November 14, 2006  
Anonymous Anonymous said...

Intel’s ability to compete in the future is basically philosophical. They are a closed technology company. Intel likes it that way because mostly only they profit from there knowledge base.

Most of the fantastic future advancements will come from scientists that don’t even work for AMD. HT is an open technology (Philosophically different). As more and more scientist and programmers brain cells pile on for there collective company’s profits using HT and other technologies the knowledge base becomes insurmountable for Intel to over come.

Example; Apple and the PC in the 1980’s

If Intel doesn’t change its business model soon or at least before K8L, life for Intel and its employees could get ugly.

1:02 PM, November 14, 2006  
Anonymous Anonymous said...

Intel's fate is sealed. BK is inevitable. All you see from Intel now is the painful struggle before the eventual death. Even if Intel produces a processor that is 4x faster, it doesn't matter -- because 90% of its production are old junk. AMD can comfortably sell CPUs at $90 a pop and ramp production along the way. Then Rev H comes on 65nm, all Intel processors are made obsolete. AMD's advantage is that it can switch to new design very quickly. By 3Q07, 85% of AMD processors are Rev H based. FAB30 will be ramped down and become FAB38 in 1Q08. Intel's BK is set at 2Q08.

Fab this, fab that, design switching, blah blah, Intel sucks.

Bank all YOUR money on AMD then. You'd better hope that Hector's acting career really pans out and makes AMD some money. Or maybe that everyone will, uhm, eat the green breath mints.

As for the more sensible members of this audience that watched the Hectorix video, well, we don't buy into that little show, no matter how charming it is. And we have the intellect and rationality to understand that AMD's products are pipe dreams right now.

AMD 65nm is a dream. Sure K8L looks awesome, but where is it? None of it is here, it's not a reality. It's in the Matrix. Err, Hectorix.

Eat another green mint.

2:45 PM, November 14, 2006  
Anonymous Anonymous said...

“Then Rev H comes on 65nm, all Intel processors are made obsolete. AMD's advantage is that it can switch to new design very quickly.”

Dr Sharikou,
If I were an engineer at Intel and knowing AMD would soon have the ability to tweak processors runs on the fly, I couldn’t sleep at night. Think of the ramifications of this technology. Are you sure?

How come more isn’t published about this?

3:38 PM, November 14, 2006  
Blogger Sharikou, Ph. D. said...

Desktop
Q406 22%
Q107 36% (1% quad)
Q207 58% (3% quad, 8% Conroe-L)
Q307 70% (3% quad, 10% Conroe-L)


So, in 1Q07, 63% of Intel's desktop will be Netburst. AMD's X2 3800+ will frag 63% of Intel's production.

Server

Q406: 78% (13% quad)
Q107: 85% (30% quad!)
Q207: 90% (40% quad!!)


This look better, but by the time vendors validated Clovertown, K8L will be out. As we can see from Rackable's results, people won't change servers like they change underware. K8L will be out in 2Q07, and people won't waste their money.

By 2Q07, all FAb36 will be 65nm....

3:43 PM, November 14, 2006  
Blogger enumae said...

Sharikou said...

"This look better, but by the time vendors validated Clovertown, K8L will be out."

If it takes them 6-7 months to validate Cloverton, wouldn't it also take that much time to validate K8L?

3:51 PM, November 14, 2006  
Blogger Sharikou, Ph. D. said...

If it takes them 6-7 months to validate Cloverton, wouldn't it also take that much time to validate K8L?


No. AMD64 is the standard. Intel's EM64T is a clone and has to be validated with great caution to make sure it conforms to AMD64 standard.

4:51 PM, November 14, 2006  
Blogger Sharikou, Ph. D. said...

On the flipside, that means that 37% of Intel's production "frags" everything AMD has (remember we speaking about desktop apps Edward, not server now).

Ha ha, your brain lacks logic circuits. AMD's x2 3800 frags all Netburst (63%) doesn't lead to 37% of Intel frags all of AMD. Learn to compute. Keep in mind that Con X6800 has 10% lead over FX62. and FX62 frags the lower conroes on everything. You have to remember 60% of the Conroes are the E6300 with 2MB cache... So the picture is AMD frags 63% + 37%*0.6 = 80+% of Intel. Get it?

5:20 PM, November 14, 2006  
Blogger Sharikou, Ph. D. said...

Yes, but Cloverton will have been previously validated, as it merely repackaged Woodcrests. Even the thermals remain the same.



Clovertown is only suitable for low end workstation load. Intel's problem is when you add more cores, instead of getting more bandwidth, you get less. Clovertown is unsuitable for mission critical enterprise computing because the FSB bottleneck will kill performance when performance is really needed. Clovertown, like all other Core 2 CPUs are only super pi champions. Try some server load on Clovertown, you will see it crap out everywhere, 4 cores fighting for a 1066 MHZ FSB is a bloody scene. It's like hooking a supercompter on a 2400 baud modem, you are limited by bandwidth.

5:24 PM, November 14, 2006  
Anonymous Anonymous said...

"On the flipside, that means that 37% of Intel's production "frags" everything AMD has (remember we speaking about desktop apps Edward, not server now)."

so your desktop never runs cryptography? actually i believe your desktop runs crypto much more than super-pi or sse.

Core 2 does not frag nothing in general programs. Core 2 performs well on selected applications, but it is in the same ballpark as K8 in terms of (non-sse) performance per clock.

5:35 PM, November 14, 2006  
Blogger Sharikou, Ph. D. said...

That wasn't your argument. Your argument was that Cloverton would not have market acceptance due to validation delays pushing it out past K8L rollout.

My argument on Clovertown validation is based on Rackble's results. People who have standardized on Opteron will not switch to Intel, and the Opteron base is growing as HP, DELL, IBM, SUN are pumping Opterons. People know their socket F can be upgraded to Rev H. People don't buy servers for 6 months, they buy them for a couple of years. In 6 months, clovertown will be just a piece of crap. On the other hand, you buy a socket F opteron server today, 6 months later, you get a upgrade to native quad. Now, that's compelling.

6:26 PM, November 14, 2006  
Anonymous Anonymous said...

OK. If AMD had made an MCM quad core by sticking two dual core processors together, would you be booing and hissing at them? Would you declare that everyone should wait for a native quad core part before upgrading? Nah.. I didn't think so!

6:54 PM, November 14, 2006  
Blogger 180 Sharikou said...

Doctor - this argument sounds strangely like your position prior to the Woodcrest launch that it was going to be a dud - a dud that has become 40% of the UP/DP market in 3 months from launch. Also...in spite of Netburst being crap...Intel continues to sell millions of desktop parts. Well, certainly more than AMD. Because my dear chap - performance is not the only thing that drives the markets. There are some simple issues like price, supply and brand. Pentiums at 90 bucks do have a market whether you choose to accept it or not.

http://sharikou180.blogspot.com
(A more balanced POV)

7:12 PM, November 14, 2006  
Blogger enumae said...

Sharikou said...

"...On the other hand, you buy a socket F opteron server today, 6 months later, you get a upgrade to native quad."

How do you know that Intel's native quad core won't be socket 771, and that people who bought Woodcrest won't be able to upgrade?

Why do you continue to negate the fact that 80%+/- of the server market is 2P and continue to call that low end?

7:15 PM, November 14, 2006  
Blogger Sharikou, Ph. D. said...

this argument sounds strangely like your position prior to the Woodcrest launch that it was going to be a dud - a dud that has become 40% of the UP/DP market in 3 months from launch.

That's just because Intel simply sold Woodcrest instead of Netburst Xeons -- so called Dempsey. Clovertown will kill off Intel's 4P business for sure.

7:27 PM, November 14, 2006  
Anonymous Anonymous said...

Hey doc how about an update on the exploding chips. I get a laugh every time you say that. I sure enjoy your sense of humor. How about this one...you ready??? Intel has already gone bankrupt and was bought out by them Iranians which actually buy all them dumb conroes to make it look like Intel is actually selling them when in fact they are NOT!!!! As a matter of fact, conroes do not EXIST!!! They were made up by them Intel marketers to make us believe that Intel could actually make a chip. WOW!! am i good or what? I'm sure you can make this one work doc. Hey let me know if you need more material? Your #1 fan, Minikou G.E.D

9:39 PM, November 14, 2006  
Anonymous Anonymous said...

"Sharikou, Ph. D said...

That's just because Intel simply sold Woodcrest instead of Netburst Xeons -- so called Dempsey. Clovertown will kill off Intel's 4P business for sure.

7:27 PM, November 14, 2006"

Wait a minute, I thought Intel had to get rid of the 90% Netburst inventory before they could actually sell Woodcrest. How can a company who still takes home over a billion dollars a quarter go bankrupt? You sir need some help. I thing that you should do the right thing and acknowledge that you don't have a clue, if you were a real man you would tell the real amd fans to go to Scientia's blog for a real educated discussion not this garbage. Poser!

9:45 PM, November 14, 2006  
Blogger Fujiyama said...

I really don't understand why you all argue about Intel vs. AMD where all information until K8L introduction is available. It is easy to observe that the market is going into situation where 50% belongs to AMD and 50% to Intel. A lot of companies are trying to implement this model right now: HP, even DELL.
AMD is late with 65nm because they are not able to deliver enough CPUs. Why don't you comment that Doctor?

I also see the mistake in stopping faster single-core CPU like 2.6 or 2.8GHz. It is cheaper to produce and most of the people don't need second core. So the single core can give good margin and the advantage in the midrange PC offerings.
It can be 90nm - why not? IT can be 65nm.

10:28 PM, November 14, 2006  
Anonymous Anonymous said...

Edwards said: Core 2 does not frag nothing in general programs. Core 2 performs well on selected applications, but it is in the same ballpark as K8 in terms of (non-sse) performance per clock

You need to take the comment in context to what sharikou was talking about – XPs fragging Intel’s CPUs.
Maybe you need to open your eyes and see the benchmarks on Media encoding, Gaming and other computational heavy tasks. Core2 performs better in all applications, and increases the performance gap in selected applications. K8 is nowhere near Core2’s ballpark. Maybe P4.

12:51 AM, November 15, 2006  
Anonymous Anonymous said...

Ha ha, your brain lacks logic circuits. AMD's x2 3800 frags all Netburst (63%) doesn't lead to 37% of Intel frags all of AMD. Learn to compute. Keep in mind that Con X6800 has 10% lead over FX62. and FX62 frags the lower conroes on everything. You have to remember 60% of the Conroes are the E6300 with 2MB cache... So the picture is AMD frags 63% + 37%*0.6 = 80+% of Intel. Get it?

This is the dumbest argument ever since apes started to communicate.
The argument is false because it takes away the fact that the market is segmented even for desktop. There is no point in saying that a CPU intended for the mid range is better than the budget CPU. Just like it would not make any sense to say that Core2Duo extreme edition ABSOLUTELY at 100% frags ALL of AMD’s desktop.
The market is segmented and the bottom line is that you corner the segments and dictate pricing with the best price/performance CPU.
Sure the FX62 beats Intel’s low end CPUs, but isn’t that what Intel wanted? That AMD is competing only with its lower binned, low margin parts? That is a sure long way from $1000 FXs controlling the price points and milking the market. Now that is just sad.

1:15 AM, November 15, 2006  
Anonymous Anonymous said...

AMD Promises Imminent Supply Relief
REVIEW DATE: 10-NOV-2006
By Michael Vizard

In an effort to quell unrest among system builders, Advanced Micro Devices said Nov. 9 it expects all the issues related to availability of its processors to be resolved within a month.
http://www.channelinsider.com/article/AMD+Promises+Imminent+Supply+Relief/193765_1.aspx

7:07 AM, November 15, 2006  
Anonymous Anonymous said...

"They still haven't played the CSI/IMC/dual socket cards yet. And still on top where it counts. What's AMD got left to shoot its wad?"

Lets see... CSI will be eating dust out of the gate, HT 4.0 will be here by the time Intel finally decides to get CSI up and working in silicon.(2009 at this rate) Judging by HT 3.0's specs 4.0 will most likely double that in bandwidth, let alone any nifty new features it will boast.

Torrenza should be well established and developed by then, co-processors of all kinds will be out on the market as well, not just using the HTX standard, but being a direct drop-in compatible replacement on socket F as well and even eventually on-die with AMD's processors. Possibly even thier new CPU/GPU hybrid by that time frame. Most likely it will be a laptop chip, a successor to the "Bulldozer"? Most likely it will be in place on the Torrenza standard as well on desktop/servers.

Intergrated memory controllers are supposedly already been done by Intel in the past, however, getting it in Core 2 (or Core 3 by this time) is a different story. Hard to say, but I will say early revisions of it will not be without issues or problems as AMD has had to deal with at the K8's inception. AMD's crossbar also plays a big part in it's communications interface on-chip,(between cores) Intel will have to come up with something simular I assume for core to core communication as shared cache is prone to cache thrashing(and not the best solution for on-chip communications, much less using the FSB) and the IMC isn't the only thing that provides can provide a latency advantage.(I won't even bother going into HT's advantages here either) I keep hearing AMD is rumored to be going to dual memory controllers(DDR3 and beyond?) by this time, however I'll wait to see what materializes in thier roadmaps pertaining to this first.


Bottom line: AMD's got plenty to shoot. Does Intel have any ammo left? We'll see, so far I'm not very optimistic with thier current plans.

7:13 AM, November 15, 2006  
Anonymous Anonymous said...

Intel's newest chips were developed in Israel, and now laptops are exploding.

8:17 AM, November 15, 2006  
Anonymous Anonymous said...

"AMD took the 2nd spot in super computer top 500"

It has ~26000 cores at 2.4GHz. Second spot has ~41000 cores and ~90% of the AMD's performance. Also those cores are ~3.4x slower than AMD's. Normalizing to AMD CPU speeds that machine could theoretically deliver around 3x more processing power than AMD.

I'd say it scales way better than AMD cluster since it has much more cores to distribute data to. Also I wouldn't be surprised if the power bill would be smaller.

Also that AMD system takes ~2MW of power whereas top1 with ~2.8x more performance only takes ~1.2MW.

8:33 AM, November 15, 2006  
Blogger Sharikou, Ph. D. said...

Just like it would not make any sense to say that Core2Duo extreme edition ABSOLUTELY at 100% frags ALL of AMD’s desktop.


This is absolutely untrue. Core 2 may be faster than K8 on integer, but Core 2 is at least 40% slower than K8 on FP performance. In that case, K8 simply frags Con XE like crap. That's also why in HPC environment, people are not buying Core 2. As you can see from http://www.spec.org/cpu2000/results/res2006q4/, a 2P Rev F get a specfp score of 119.

11:08 AM, November 15, 2006  
Anonymous Anonymous said...

"This look better, but by the time vendors validated Clovertown, K8L will be out. As we can see from Rackable's results, people won't change servers like they change underware. K8L will be out in 2Q07, and people won't waste their money."

So 6-7 months to validate cloverton and 0 months to validate K8L? Oh wait this is just a beefed up K8 anyway so it should validate quickly? (Q2'07 release = Q4'07 release by your validation timelines).

On the other hand if this is an "innovative, new" microarchitecture shouldn't it also take some validation time too?

Also will AMD ramp K8L instantly to 100% server mix and EOL K8 in Q2'07? Or do you think they may do a phased approach like Intel?

For it to be ready in Q2'07 in volume (let's say June for arguments sake), that Si has to be started in Feb to allow wafer processing and packaging.... Q2 (and probably Q3'07) will be low quantities...

12:38 PM, November 15, 2006  
Anonymous Anonymous said...

"The Cray XMT can run 1 million threads concurrently."

Around 99.9% of those threads are being run on the special Cray chips, not in the x86 CPU's. Each of those chips are running up to 128 threads, for 8000 chips it makes a bit >1M threads.

4:09 AM, November 16, 2006  
Anonymous Anonymous said...

"Around 99.9% of those threads are being run on the special Cray chips, not in the x86 CPU's."

Wrong. The Threadstorm is a 4 million gate ASIC around 500MHz; it is not a general processor and it runs only userspace codes (up to 128 sequences at once). Since these ASICs do not perform process/thread or IO/memory management, any thread they run would be managed by the Opterons at some point of their lives.

I cannot imagine such tasks could ever be done by memory-starved processors such as Core 2. Cray chooses Opteron for good reasons (even to support the massively multithreaded programs).

Besides, massively parallel scalar codes are run by the Opterons, since being a general purpose processor it is much more capable of exploting ILP than the Threadstorm. Again, Core 2 with mere large cache and FSB simply cannot perform such tasks well.

1:14 PM, November 17, 2006  
Anonymous Anonymous said...

eat this!
http://www.extremetech.com/article2/0,1697,2068252,00.asp

12:21 AM, December 08, 2006  

Post a Comment

<< Home