backtop


Print E-mail del.icio.us 22 comment(s) - last by iollmann.. on Dec 14 at 10:14 PM

AMD's TLB bug may go down as one of the largest technical blunders of 2007

Last week DailyTech mentioned the shortage of AMD Barcelona processors in the channel.  AMD's Phenom processor is also in shortage, but Barcelona's exit from the marketplace is more of a mysterious Houdini act than a drought. 

SPEC confirmed this shortage today when it invalidated all AMD K10 benchmarks from its website.  All AMD K10 benchmarks on the SPEC site have been replaced with the words "Not Available," and the description below:
SPEC has determined that this result was not in compliance with the SPEC CPU2006 run and reporting rules. Specifically, the submitter reported that the result would not meet the SPEC OSG requirements for continued availability.
SPEC's benchmarking scores are traditionally heralded as the "official" high-performance computing scores for industry analysts, universities and volume purchasers.

AMD's Scott Carroll elaborates. "As soon as we determined that we could only ship Quad-Core AMD Opteron processors earmarked for specific customer deals, we began the process of withdrawing our SPEC scores. As we get closer to the launch of the updated Quad-Core AMD Opteron processors in Q108, we will submit new scores."

DailyTech previously dubbed this refresh as the B3 Opteron stepping -- which was also confirmed by ZDNet's George Ou.

The specific deals Carroll describes are the result of a race condition in the K10 architecture.  Without a motherboard BIOS patch, Barcelona Opteron and Phenom processors are susceptible to lock-ups during nested memory operations; a scenario frequently found in virtualized workspaces.

DailyTech was able to confirm that all Barcelona chips have been removed from retail sources.  Any Barcelona sold in volume must pass AMD's screening in advance.


Comments     Threshold


This article is over a month old, voting and posting comments is disabled

Fair?
By NullSubroutine on 12/13/07, Rating: 0
RE: Fair?
By JumpingJack on 12/13/2007 9:25:28 PM , Rating: 4
You need to read the SPEC rules for running and reporting scores, they simply applied the rules. The reason it has not been done to Intel is because they have not violated the reporting rules.

SPEC was created specifically to remove ambiguity and be non-platform favoring both in running and reporting. It is their mantra. They even publish for you their policies, license agreements, and rules at www.spec.org.

http://www.spec.org/cpu2006/Docs/runrules.html
Section 4.1.4:

quote:
If a software or hardware date changes to more than 3 months after first publication, the result is considered Non-Compliant. For procedures regarding Non-Compliant results, see the SPEC Open Systems Group (OSG) Policy Document, http://www.spec.org/osg/policy.html.


From their policy:

quote:
If the subcommittee's review returns a finding that the result is not in compliance with the benchmark's run and reporting rules, the following should occur.
The results are modified:

To remove the metric values (summary & individual) and replace with "NC" (non-compliant) .

Add a header to the disclosure stating that the result did not comply to the rules and the reason(s) why.

The submitter, with the subcommittee's approval of the text, may add additional wording into a remedy section. The remedy section is expected to explain how the vendor has addressed the problem. For example, mentioning that a new result has been submitted on the same platform with the issue resolved is the expected typical usage of this section.


Every company that purcahses a licensing agreement reads and understands the risk of submitting pre-production material.


RE: Fair?
By NullSubroutine on 12/13/2007 10:40:03 PM , Rating: 2
So what you are saying they are saying is they invalidated the scores because they used engineering samples?

Didnt the engineering samples actually perform worse than the ones you could actually buy?


RE: Fair?
By JumpingJack on 12/14/2007 12:18:09 AM , Rating: 2
I am not saying anything... I am pointing out that spec has a general availability clause in their licensing, and if a vendor fails to make the HW generally available within a certain time period of the submission, SPEC will invalidate the results.

Barcelona is not generally available thus you cannot actually buy them and AMD has stated general availability will not occur until Q1 of 2008.


RE: Fair?
By JumpingJack on 12/14/2007 1:01:45 AM , Rating: 2
I should add ... as such, you have not seen them do this with Intel because Intel has not submitted scores for a non-shipping product in the past. What Spec did was within their guidelines, all vendors sign up via the licensing agreement, and hence it is fair what they did....

While Intel has had product problems before, they have yet to submit scores and fail the 90 day test.


RE: Fair?
By defter on 12/14/2007 3:54:15 AM , Rating: 2
quote:
Didnt the engineering samples actually perform worse than the ones you could actually buy?


No, the Barcelona chips that you can buy today perform 5-50% worse than previously reported scores because of the TLB bug. AMD doesn't want to embarrass themselves by resubmitting a new very low SPEC scores using current stepping + TLB BIOS fix.

In addition, SPEC has a rule: hardware and software used in submission should be generally available within three months of the submission. Barcelona SPEC scores were submitted in August-September timeframe, while the B3 stepping will be available in mid/late Q1 2008, that's more than 3 months, thus current scores are invalidated.

That's why AMD is waiting for B3 stepping availability before submitting new SPEC scores.


AMD needs to get it together
By 13Gigatons on 12/13/2007 7:37:01 PM , Rating: 2
They are really struggling to make K10 cook. I wish they would fire the top guys and bring in some new guys.




RE: AMD needs to get it together
By InternetGeek on 12/13/2007 8:23:27 PM , Rating: 2
I don't think this is a matter of firing execs. At this point I don't think they should be doing it anyways. This sounds more like something they realized [i]after[/i] sending out the CPUs, or they realized the BIOS patch would take longer or is not really a workable solution once they got their software engineers to work on this problem.


RE: AMD needs to get it together
By bobbronco on 12/13/2007 11:09:49 PM , Rating: 5
Ergo it's a problem that needs to be addressed by a change in attitude and perspective within AMD's own management. Changing this attitude should not only be done as a result of this recent 'cock-up' but also for the sake of the company's own reputation and that of all future products AMD releases to the marketplace.

The whole reason this bug made it into the end K10 products was mostly due to AMD managment's own hubris stemming from the robustness of its previous generation processor design. In fact, the first Opteron products did so well that they [management] decided at some point to not adequately support or sufficiently fund the internal validation organizations in charge of proving out future products.

This attitude and apathy towards concurrent engineering practice is not uncommon amongst most business level VPs and executives at tech companies. These execs, unless they have experience in product engineering, don't like their own internal product validation groups because they are cash hogs and they require significant recurring investment in the form of engineering overhead. When processor designs are executed well and the number of critical bugs in a product is low or non-existent, these validation groups are perceived to add little value to the company and the overall product development chain. They appear to be a drag on profits. Executive management reacts by scaling back funding and head count for these groups in order to improve the bottom line.

This same pattern of initially funding and then scaling back the mission critical function of product validation has been done time and again in virtually all engineering centric companies. Intel has done it in the past as well, but they learned from their mistakes. Now, it's AMD's turn to learn this painful lesson. If I were a shareholder, I would be demanding a change in the guard at AMD, and for a shift in strategy where engineering drives the business and key decisions therein -not the other way around.


RE: AMD needs to get it together
By RamarC on 12/13/2007 8:39:54 PM , Rating: 2
"top tech guys" are not the problem. i'm sure the root of the problems is senior management. amd won't right the ship until there's a major shakup at the boardroom level and they divest and refocus their efforts.


Subtitle
By AstroCreep on 12/13/2007 9:00:45 PM , Rating: 2
quote:
AMD's TLB bug may go down as one of the largest technical blunders of 2007

Am I the only one who finds the subtitle funny? ONE of the biggest technical blunders? What technical blunders did I miss? lol, I can't think of any others.




RE: Subtitle
By ZJammon on 12/13/2007 9:19:14 PM , Rating: 2
AMD had both in my opinion seeing as the R600 was late and a fire breathing monster. Nor did it surpass Nvidia's year old 8 series. They need to do something unless everyone here loves being extorted...


RE: Subtitle
By JumpingJack on 12/13/2007 9:19:25 PM , Rating: 1
Well, funny yes in the sense that it 'may' ... considering the magnitude of the importance of this product to AMD and how badly it was handled. There are roughly 2 weeks left in the year, something really major in the industry would need to go down between now and then to top it.

On the flipside, it is a bit sad because the repercussions are tremendous.... what this has forced AMD to do, if you gleen the info from the web, is push out bulldozer and scale back R&D ... this is not a good way to compete against Intel who is picking up the pace. It appears AMD may be revising their business strategy to pull away from the high-performance market in the long term and focus on becoming the mainstream second source supplier.

This would not be fun to the performance demanding enthusiast, nothing to challenge Intel at the top end just hurts my wallet :)


RE: Subtitle
By PitViper007 on 12/14/2007 10:05:58 AM , Rating: 2
quote:
It appears AMD may be revising their business strategy to pull away from the high-performance market in the long term and focus on becoming the mainstream second source supplier.


Again


RE: Subtitle
By iollmann on 12/14/2007 10:14:02 PM , Rating: 2
Well, c'mon, to be fair, they only ever got into the performance market because Intel's Netburst core hit the thermal wall and couldn't ramp up the frequency to 10 GHz and beyond as planned. Without the frequency ramp, the design inefficiencies of Netburst's narrow architecture were left to wallow in the mud. AMD found itself sitting pretty with a wider, more efficient design that ran faster and used less power. There was also the Itanic distraction. Intel justifiably lost a lot of revenue, but has since altered course and has the ship pointed in mostly the right direction again. Intel's ISA design is and always has been three-sheets into the wind, but the average consumer doesn't care about that.

Intel's biggest enemy is Intel. Their tombstone, when it comes, will have one word written across it: "Groupthink" They execute well, but as far as plotting a course into the future, they are often just lost in the fog of minutia.

So yeah, I expect AMD to be opportunistic at times and seize some big chunk of the market whenever Intel runs aground. That will force Intel to change course again. Since Intel generally executes better, AMD will probably cede that market share back to Intel, until Intel runs aground again.

I like to think of smaller competitors as pilot boats for the big guys.

I still don't get the ATI merger, though.


Misleading title
By KernD on 12/14/2007 9:20:06 AM , Rating: 2
The title reads
quote:
SPEC Invalidates All "Barcelona" Benchmarks due to Availability

It's misleading since it's AMD themselves that had the scores removed.
quote:
AMD's Scott Carroll elaborates. "As soon as we determined that we could only ship Quad-Core AMD Opteron processors earmarked for specific customer deals, we began the process of withdrawing our SPEC scores. As we get closer to the launch of the updated Quad-Core AMD Opteron processors in Q108, we will submit new scores."


The title should read:
"AMD withdraws all "Barcelona" SPEC benchmarks score due to availability"




RE: Misleading title
By TomZ on 12/14/2007 9:49:25 AM , Rating: 2
From what I have read, AMD has notified SPEC they would not be able to meet the availability requirements (probably in response to a private notice from SPEC to AMD), and SPEC therefore invalidated the benchmarks.

AMD is just putting a marketing spin on it saying they "withdrew" the benchmarks. The reality is that SPEC would have been required to invalidate the benchmark regardless of whatever action AMD took short of actually shipping working silicon.


RE: Misleading title
By defter on 12/14/2007 3:09:15 PM , Rating: 2
Results are still there, they had not been withdranw. They are just marked as "non-compliant" or "not available"

Look here:
http://www.spec.org/cpu2006/results/res2007q3/cpu2...
http://www.spec.org/cpu2006/results/res2007q4/cpu2...


RE: Misleading title
By defter on 12/14/2007 3:09:15 PM , Rating: 2
Results are still there, they had not been withdranw. They are just marked as "non-compliant" or "not available"

Look here:
http://www.spec.org/cpu2006/results/res2007q3/cpu2...
http://www.spec.org/cpu2006/results/res2007q4/cpu2...


Good news for AMD fanbois !
By DallasTexas on 12/14/2007 9:57:52 AM , Rating: 3
It's not all bad for AMD fans. Hector is the highest paid executive among his semiconductors peers. That's right, AMD again winning against Intel in CEO pay!!

Posted by Tom Krazit/New.com
After spending most of Thursday apologizing to financial analysts for AMD's performance in 2007, the company revealed in a filing with the Securities and Exchange Commission that Ruiz's compensation agreement was amended on Wednesday with his new salary: $1,124,000....a total compensation package for the year of $12,848,435.The move should allow Ruiz to remain the highest paid CEO among his peers in the semiconductor industry on Forbes' annual list of CEO pay next year.




RE: Good news for AMD fanbois !
By Proteusza on 12/14/2007 11:28:58 AM , Rating: 2
Oh thats great. At least Hector is being well compensated for running AMD into the ground, Heaven knows it was a difficult job considering how good the Athlon 64 was doing.


Gasp
By Crazyeyeskillah on 12/13/2007 7:25:37 PM , Rating: 3
How does a blunder affecting your target consumer make it to the marketplace in the firstplace? Was this a case of the Ford Tire/Death Lawsuit collateral risk that simply didn't play out as projected by AMD research??




"Vista runs on Atom ... It's just no one uses it". -- Intel CEO Paul Otellini














botimage
Copyright 2009 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki