All AMD needs is some TLB

TLB – translation look-aside buffer. Sounds techie and yet doesn’t sound so disastrous. Bear in mind that TLB is a microcircuit within the CPU and what AMD wants you to know is where the problem occurs while not necessarily telling anyone what went wrong and what changes are required to fix it. Or maybe I’m just the type that wants more detail. As a consolation they did say what could happen and I suppose the worst is a “system hang” with a very low occurrence rate. It’s not really the worst problem but you can scratch servers that might be used for critical missions off your customer list. Above anything else, this should be the reason why no OEM is shipping Barcelona servers at the moment. The seriousness of this bug makes even the poor performance just an after thought.

The fact that the problem is across the K10 platform (Opteron and Phemon) makes it possible that this is a micro-architectural problem and should have been identified at the Functional Validation stage. There are several other validation screens in place to catch problems like this but because it managed to slip out makes this more of a symptom of how AMD operates these days. A rushed and under resourced design process can lead to disasters like this. But to be fair to AMD no amount of pre-silicon validation can successfully screen the complexity of today’s microprocessors. Running a full chip simulation on an almost infinite amount of combinations of dyadic instructions is impossible especially when time is crucial.

Validation Primer:
There are two levels of validation and they are done at pre-silicon and post-silicon levels. At pre-silicon, Functional Validation and Logic testing are done.
Functional Validation is a simulation done to test different micro-architecture features such as Barcelona’s TLB. This is time consuming and compute-intensive, typically running assembly language in low single digit Hz speeds completing billions of cycles per week. The test is feature-focused and localised.
The other pre-silicon validation is logic testing and the purpose is to validate circuit behaviour using logic sequence and combinations.

Post-silicon validation involves Performance Verification, Design Validation and ultimately Manufacturing Validation.
At Performance verification the chip is tested against physical specifications such as leakage, voltage, temperature and more importantly speed and timing. Being slow and leaky, anyone can imagine how AMD would have felt when K10 reached this stage. Timing analysis is done here and if Barcelona had critical path problems at the L3 TLB like GURU suggested, they should have detected the problem at this stage.
The next step is Design Validation where a complete system level check is done. At this stage the CPU is attached to normal peripherals (i.e., BIOS, chipsets, Operating System) and all features of the chip are tested.
The last step is Manufacturing Validation where yield becomes the major metric which is essentially driving overall cost to manufacture. Unfortunately for AMD their problem with design is compounded by 65nm process issues.

Due to lack of information, it’s hard to say if AMD is applying a circuit/architectural design change or a process design change on the B3 stepping. A circuit design change points to a poor pre-silicon validation process while a process design change (i.e., change in CD’s) points to post-silicon validation mistakes. To be honest, with the amount of problems K10 is having only God knows how many modifications AMD plans to include in their next stepping. We hope they get it right this time around because I can only feel sorry for the guy who buys a tri-core Phenom with a microcode patch disabling the TLB. Surely there is a line in the sand that says when a product in broken and cannot be sold.


Update: In relation to my blog post on the 27th Nov wondering where the AMD stock would settle, well today the stock closed at $8.91 as it continues to slide downward. It appears like institutional investors are finally bailing out so no bottoming out just yet.

One!

It is exactly a year ago when this blog started. Once intended as satire for an infamous blog site, I suppose today we can say that we have outperformed Scierikou in terms of accuracy and usefulness in predicting the future and providing analysis on current events. It is indeed facinating that what we have been discussing collectively for the past several months only ends up lately on someone else's blog. Throughout the year we've been accused of painting AMD in a bad light but we managed to prove that it's only because there's a thick dark cloud hanging over it. We've been brutal and honest but so is the truth.

I would just like to mention that this site never really kicked off until our friend Sharikou180 provided a link to this site. Site activity wouldn't have picked up if it wasn't for the people that come here and contribute. Many thanks everyone.

Looking ahead, I'm considering the option of opening up this blog for additional contributors. I can see that some put considerable effort in their post which are at times quite noteworthy and insightful enough of being an article by itself. The two options for contributing are adding authorised authors (requires blogger account), and as for our friends in the "industry" that requires anonymity, you can send your article through e-mail (I will post the article under your chosen pen-name).

What do you think?

AMD's Stock Reaches New Lows

In an alarming trend, AMD's stock seems to be on a free fall since last week. Today the stock reached a new 52-week low dipping below $10 no thanks to the downgrade by analyst Doug Freeman. It's nice to see that the investment community is finally catching up with our own sentiments about AMD's future.

"We do not expect sentiment to improve anytime soon and believe even improved execution will not be enough to show true operating leverage in the near-term," he wrote in a research note...While we remain bullish on PC demand in general, we don't see a material catalyst for upside emerging in the next 3-4 months and suspect investors will view any announcements on the company's fab-lite strategy with skepticism, regardless of this view's merits."

What I find interesting about this downgrade is the suggestion that even if AMD improves on its execution, the outcome still won't be enough. Even if AMD proceed with its yet to be announced asset-lite strategy the outcome will still be immaterial. And even if the PC demand remains strong AMD will continue to lose money. So what that means is if AMD becomes lucky enough to get a "perfect weather" of good execution, strategy and excellent market conditions, still, there is no hope for the struggling company, not now or anytime soon. "Investor skepticism" has finally caught up with AMD and it will be interesting to see where the stock settles after hovering around the $13-14 region for a year.

Now if only some of the missing bloggers return and put up the usual face of hope, maybe the AMD stock can rally back up.

The QuadFX Disaster Part Two

AMD today released its much awaited quad core desktop, the Phenom. After reading the reviews I just felt a sense of deja vu. Almost a year ago when AMD released the QuadFX to compete against Intel's newly released quad-cores, here is what some had to say:

Ars Technica (2006):

"Today, AMD officially launched their much-anticipated 4x4 enthusiast platform under the name QuadFX. The results of the rash of QuadFX reviews that just came out are uniformly disappointing; Intel's quad-core offering, the QX6800, outperforms AMD's much hotter, more power-hungry offering in almost every benchmark".

Anandtech (2006):

"When only running one or two CPU intensive threads, Quad FX ends up being slower than an identically clocked dual core system, and when running more threads it's no faster than Intel's Core 2 Extreme QX6700. But it's more expensive than the alternatives and consumes as much power as both, combined... Until then, there's always Quad FX but you're better off with Kentsfield".

Then again a year later while trying to take another stab at the "enthusiast" market:

Ars Technica (2007)

"Current benchmark results from Anandtech, HardOCP, and Hexus indicate that the Phenom, while notably more efficient than Athlon 64 X2 in certain scenarios, still lags the Q6600 clock-for-clock. Phenom may have finally given AMD the ability to offer a quad-core processor to compete with Q6600, but Intel's chip still holds a better price/performance ratio".

Anandtech (2007)

Phenom is, clock for clock, slower than Core 2 and the chips aren't yet yielding well enough to boost clock speeds above what Intel is capable of... today's launch confirms that Intel is still the king of the quad-core market.

Hexus (2007)

We can debate all day whether the majority of consumer software is threaded enough to take advantage of four execution cores, but the immutable fact remains that AMD's fastest quad-core offering is slower than Intel's slowest. Compounding this depressing statement for AMD is the January 2008 launch of Penryn-based Core 2 Quads, furthering Intel's performance dominance.

In order to grasp the magnitude of how much AMD messed up on execution, it's important to realise that even after a year of humiliating itself with the QuadFX disaster it has yet to beat it's old nemesis the QX6700. Intel was kind enough to lower it's quad-core offering to QX6600 but it seems AMD's Phenom can't beat that either. Never mind taking back the performance crown or beating the opposition on some important benchmark, but failing to beat a year old processor is simply inexcusable. This is indeed a wake up call and should desensitise anyone who's heard far too many excuses from AMD.

All Your Fabs Are Belong To Us - TSMC

There is a report that AMD is in talks with TSMC to sell Fab30. As AMD struggles to bring the company back to profitability this particular “rumour” doesn’t sound so outrageous, especially when we’ve already entertained the possibility of such a strategy. The “idle-race-car-in-a-pit-stop” description mentioned by Mr Ruiz on the current situation of Fab30 is probably not the best metaphor for an empty Fab. A Fab with a ready capacity is the real “idle race car”, but then again this cost $millions to sustain. An empty Fab without cash for the necessary upgrade sounds more like an empty pit stop - no race car! And if such is the case then its just a wasted asset. There is also the other problem of excess manpower that AMD isn’t allowed to release lest they lose their grant. Anyone familiar with wafer manufacturing knows that an idle Fab however less costly is just as bad as an underutilized one.

If there is one reason why anyone should believe the rumour it has to be based on this simple fact. Nobody builds or upgrades a Fab without expecting enough volume to sustain the added operating cost, let alone a return on investment. The initial assumptions made to support an AMD with two Fabs did not materialise and the result of which is just one idle Fab that it cannot afford to keep. Even if AMD did have the money to upgrade Fab30 to 300mm/(Fab38), there won’t be enough volume to support such a massive increase in operating cost. Clearly AMD finds no use for Fab38 at the moment and if AMD plans to keep Fab30 idle then that simply goes against any asset-lite approach. An unutilized asset which incurs cost doesn’t make any sense. What makes more sense is a foundry like TSMC with enough customers to fill a Fab jumping in and taking over. This even coincides with the other rumour that TSMC will ramp AMD’s 45nm in 2009.

Barcelona Benchmarks Are Non-Compliant - SPEC

As if the initial Barcelona benchmarks submitted eons ago wasn't bad enough, SPEC.org appears to have labelled them as non-compliant. IBM having problems securing enough Barcelonas failed to meet the requirements of shipping systems 90 days after submitting results.

What I find really surprising is the fact that even a Tier1 vendor like IBM can't seem to get hold of AMD's latest quad-core. If an important customer can't even ship a lower bin 1.9Ghz Barcelona months after product launch one has to wonder how many has AMD really managed get out of the door. I was joking around when I said AMD most likely shipped 20,003 Barcelona's because it's the lowest end of the statement "shipped tens of thousands". With the news spreading and the gathering noise about K10's lack of availability, it appears my tongue in cheek comment may be a bit more accurate than first thought.

IBM fails to ship even the most meager four-core Opteron box - The Reg
"In the meantime, IBM has to look the sucker with NON COMPLIANT stamped all over its benchmarks."

If you're one of those that ever wondered what is the worst kind of evidence when caught doing a paper launch,... well, now you know!

What's In A Price?

In basic economic terms, 'price' is the monetary equivalent of 'value' at a particular point in time. But 'value' is such an abstract concept that it varies from person to person, from time to time and depends very much on circumstance. One such circumstance is of course availability. We know the more scarce an object is the more valuable it becomes. The second is the benefit it provides. If Phenom is the best processor at any meaningful measure, naturally it should also be the most valuable processor in the market.

Then why is Phenom being rumoured to launch at a priced only equivalent to Intel's mainstream quad processor, the QX6600 at $288? For a consumer, the bottom line is value for money and usually that means the most performance per dollar spent. Processors are typically priced based on how they will perform and AMD and Intel follows this accordingly. So at the end of the day, it doesn't matter what the benchmarks says or what some crazy blogger thinks. If AMD has priced its next generation processor right around the bottom of Intel's old generation products, there isn't much anyone can say about how badly K10 turned out to be.

How much is Phenom valued? (as per Newegg 5th Nov 2007):
Intel Core2 Extreme QX6850 3.0GHz at $1029.99
Intel Core2 Extreme QX6800 2.93GHz at $1015.00
Intel Core2 Extreme QX6700 2.66GHz at $949.99
Intel Core2 Quad QX6700 2.66GHz at $534.99
Intel Core2 Duo EX6700 2.66GHz at $319.99
AMD Athlon64 FX-74 at 3.0GHz at $299.99
<-- Phenom launches here -->
Intel Core2 Quad QX6600 2.4GHz at $284.99
Intel Core2 Duo E6850 at 3.0GHz at $279.99

Phemon performance sits where it is priced. End of discussion? Not really because we can still discuss how launch prices are typically overvalued due to the anticipated initial demand. But that's just beating a dead horse.