9.08.2007

The Leaked Barcelona Numbers

Leaked Barcelona launch slides have reached George Ou. Obviously the numbers doesn't look very good for AMD. But if you've been in this blog before, then you shouldn't be surprised.

Instead, I am quite intrigued by the nature of the leak. It looks to me that if Barcelona's numbers are weak, AMD has much to gain by leaking the numbers and slowly deflating everyone expectations before the launch. These sort of tactics never really amount to anything in the long run as we've already seen how Core2 was benchmarked to death when it was launched. Thanks to the internet and the growing numbers of benchmarking sites, AMD's PR Department, specifically the Damage Control Team will be very busy for the next few years.

15 comments:

Anonymous said...

Yeah but how did the "platformance" benchmark look? Was AMD leading in the "powerpoint slides per clock per Sales and Marketing VP per dollar per simulation" metric?

It really doesn't matter what the benchmarks are like early on - the people who are buying these at the onset would likely buy them short of them causing a brownout.

The question as alway will be can AMD manufacture the higher bin parts at decent enough yield to survive. If they need to price these fairly comparable to Intel's offering (which it now looks like they may have to) and Intel has both a clockspeed, MCM and process node advantage, AMD will be in REAL TROUBLE.

If of course they somehow end up clock for clock better then they can afford to price a bit higher which would offset their manufacturing cost and process technology disadvantages.

So once again it all comes down to AMD manufacturing... AMD flourished when their design was superior, but if their new designs are simply equal, Intel's manufacturing muscle will likely be felt for some time to come.

I expect to see some totally new benchmark metrics at the launch (meaning look for AMD to create some new benchmark to spin Barcelona). At least Henri's probably going to get a good night's sleep!

Anonymous said...

there once was a man named brent rehmel

he prophesized amd's great 2007

alas it wasn't so

poor sceintia

Anonymous said...

I think the early benchmarks will matter, because they will give us an idea of how competitive Barcelona will be. I'm sure everyone is pretty tired already of all the marketing and "geek-speek" numbers being thrown around telling us how fast or slow it is. I want to see the general benchmarks that the various sites will run and get a clue as to how it compares to anything else.

After all the hype and cross-hype and the multitude of layers of BS being thrown around by all sides, I'm interesting in seeing just what Barcelona brings to the table.

Anonymous said...

See this this comment by who. Fifth from the top. Real world performance of Core2 and Barcelona may be closer with regard to SPEC.

Anonymous said...

Roborat there another more important chart (at least for us gamers) where it show one 2.0Ghz AMD Quad fraging one 2.0 quad core intel by 68%.

Axel said...

In an act of blatant fascist censorship, Scientia has once again stooped to deleting posts that he's unable to refute. I made the mistake of forgetting to copy my latest post over here (never again!), so I will spend a few precious moments to recreate it:

Scientia

Or are you agreeing with roborat's silly theory that AMD stuffed the channel?

No, I'm agreeing with Mercury Research. "McCarron explained that the Sunnyvale, Calif., company appears to have overestimated demand in the fourth quarter of 2006 and shipped more processors to OEMs and channel partners than were needed."

Now will you try to claim that Scientia Research is more credible than Mercury Research? Time to come down from the clouds.

Well, Gary has been wrong so far with everything he has said about K10 but I can see that you like playing the long shots, so, why not?

If you hadn't noticed, Gary based his comments on K10 chips that he's currently testing. Whereas your rosy outlook is based on, umm, the AMD-produced K10 Software Optimization Guide rather than empirical evidence. I don't think you're going to be a happy camper come Monday.

It doesn't need to. Barcelona will only raise the ASP in servers.

Not much it won't. You would know this if you'd seen the price list.

But more importantly Barcelona will reverse AMD's losses in server volume which are subtantial.

No, AMD will continue to lose server share through the rest of 2007 due to low clocks. 2008 will be a pitched battle between Harpertown & Barcelona. But AMD are unlikely to return to 2006 server volume share levels because Harpertown will be far more competitive with K10 than Netburst was with K8.

However over Q4 and Q1 this [clock advantage] drops to 1.3X and then 1.1X.

Can you link to an official AMD roadmap showing this? I didn't think so. I bet you're basing this conjecture on the 3.0 GHz cherry-picked K10 demo, that no one was even allowed to benchmark to verify that it was indeed a K10 at that speed. Tsk tsk.

Yes, but unfortunately a 5% increase in IPC doesn't match a 25% increase in clock

And unfortunately this claim is based on a shaky foundation, the 3.0 GHz no-benchmark so-called demo.

They could if AMD didn't release higher clocks but they will in Q4 and higher again in Q1...In the same time it takes Intel to get a 10% increase in clock, AMD will get a 50% increase in clock.

As you're correctly inferring, AMD's ability to compete with K10 depends entirely on how quickly they can ramp clocks. However, without any official roadmap or other communication from AMD, you're just going on blind optimism.

If you look at the benchmarks it is clear that dual core C2D hits a memory bandwidth wall at 2.66 Ghz and quad core hits the same wall earlier at 2.4 Ghz.

No. If you look at this review, you'll see some comparisons between the QX6850 (3000/1333) and the QX6800 (2933/1066). The performance increase in most cases is small and atributable to the 2.3% higher clock. This shows that the FSB & memory subsystems are not a bottleneck for most common multithreaded tasks, even for high clocked Kentsfield.

This means that Penryn quad will be scaling badly above 3.2Ghz.

Incorrect. As explained above, even Kentsfield doesn't hit the FSB wall at 3.0 GHz in most tasks. I don't know how you can keep making these categorical statements without supporting evidence. All you do is stick your foot in your mouth (like you did repeatedly over the past 12 months on multiple predictions). After all this time you still haven't learned to never listen to AMD's marketing propaganda but to listen to third parties instead.

What could you possibly be talking about? You know that AMD will release both desktop and mobile processors in volume in 2008.

Yes, K10 for desktop and power-optimized K8 for mobile. Unfortunately, both are "too little, too late" for these markets. K10 was designed for the server space and memory-intensive apps. It will not compete well against Penryn on the desktop, as Penryn has a great die size advantage, MCM benefits, higher clocks, and SSE4. And it's laughable to expect K8-based Griffin to contend with Penryn mobile. Penryn is returning to the 25-watt TDP level and offers far higher performance than Griffin to boot. Griffin is dead in the water.

This won't happen. AMD should have its losses under control by Q4.

Keep dreaming. You have failed to provide any compelling links or logic to support this position. AMD will continue to hemorrhage for several quarters, though perhaps abated temporarily by the sale of old equipment and German bailout funding.

SPARKS said...

"After all the hype and cross-hype and the multitude of layers of BS being thrown around by all sides, I'm interesting in seeing just what Barcelona brings to the table."

A truer word were never spoken. This bullshit has been going on for over a year. The outcome, however, will reveal what was going on during the past 14 months. Frankly, I can't wait, regardless.

SPARKS

Anonymous said...

Georgi,
You made some mistakes in your comments,
The Intel Quad core implementation is two dual core processors glued. What does that mean?
Well it easy... This means that you have two 4MB L2 cache processors. So?
So you have only 4MB L2 max not 8MB like you said. I'm not seeing one application running in one Core and borrowing the 4MB L2 cache from the other core. Do you see that happening with two physical detached processors?

Anonymous said...

damn axel, you beat that bitch with a bat lol

SPARKS said...

Check this out.

http://www.bangkokpost.com/
Database/05Sep2007_data006.php

SPARKS

Anonymous said...

George has updated his blog. This time, with a link to an IBM release with SPEC scores.

ftp://ftp.software.ibm.com/eserver/benchmarks/news/newsblurb_x3455_speccpu_091007.pdf

The other interesting piece is that the IBM system won't be available until 2nd half of November.

If Intel's Penryn servers are launching in early November as reported, this could really blunt any AMD Barcelona momentum.

Axel said...

Occam's Razor rings true: The most logical explanation is usually the correct one. Why was AMD so quiet all year? As suspected, K10 appears to be too little too late. It's a server chip that will compete well with Intel in that space when the clocks come up sometime next year. But in the desktop space, K10 is on average only 10-15% faster than K8. It will be no match for Penryn. Based on those benchmarks, Penryn will generally beat K10 in the enterprise space as well. And in the mobile space, K8-based Griffin has no chance against Penryn in terms of performance per watt.

Many people are set to eat a healthy portion of humble pie today. AMD, in their current incarnation with their current business model, are finished. They will restructure dramatically in order to survive 2008, it is inevitable.

Poke said...

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3092&p=5

K10 gives about ~15% IPC improvement over K8 clock for clock, per core.

What we know so far,
Conroe has about a ~25% performance advantage clock for clock over K8. K10 IPC improvement still not enough to take down Intel's 65nm Kentsfield.

Since Penryn has about a 5% IPC improvement clock for clock over Kentsfield, AMD is going to have a hard time. Pricing is key for AMD.

Giant said...

AMD is lying and deceiving people with this "ACP" BS.

Intel's definition is straight forward:

ftp://download.intel.com/design/processor/designex/31559405.pdf

Thermal Design Power: A power dissipation target based on worst-case applications. Thermal solutions should be designed to dissipate the thermal design power.

There is nothing in there that say's 'average', WORST-CASE applications, generating the MOST HEAT.... this is just BS. It is dissapointing AMD resorts to this kind of deception.

Intel is consistently under TDP with Core 2: http://www.xbitlabs.com/images/cpu/core2duo-shootout/power-2.png

AMD goes over TDP. FX-62 with 125W TDP uses 130W of power. All Core 2 are under 65W. Core 2 Extreme X6800 is well under 75W.

AMD is a deceptive unethical company.

Giant said...

Here is a full suite of tests run by TechReport, an unbiased neutral third party.

http://techreport.com/articles.x/13176/1

Remember when looking at the results, as pezal states, the 8360SE won't be available until the end of the year.

For one, it is not going to capture the overall performance lead from Intel soon, not even in "Q4," which is when higher-clocked parts like the Opteron 2360 SE are expected to arrive. Given what we've seen, AMD will probably have to achieve something close to clock speed parity with Intel in order to compete for the performance crown. On top of that, Intel is preparing new 45nm "Harpertown" Xeons for launch some time soon, complete with a 6MB L2 cache, 1.6GHz front-side bus, clock speeds over 3GHz, and expected improvements in per-clock performance and power efficiency.