AMD's Technology & Closing Thoughts

Because of the nature AMD's poor financial situation, the key to the company's recovery must lie in its technology. In the processor market, AMD's position as a provider of low-cost, mass market CPUs is unsustainable given the costs of R&D for future processors, so they need to make the kind of technological advances that will launch them back to the top. Their graphics operation is under a similar premise, although their position there is not nearly as far behind the market leader.

We'll start with AMD's best strength right now: servers. 4 years ago AMD made a very, very good choice in creating and using HyperTransport, as this has scaled much better with 4P and higher systems than the traditional front side bus has. AMD's choice to use an integrated memory controller and NUMA has also worked out very well, as they can use large banks of traditional DDR2 DRAM instead of needing to use FB-DIMMs to achieve such large banks.

Both of these choices have enabled AMD to be very competitive with Intel in the server market, which is great news for the company as server processors (especially 4P and higher) are products with high margins that can be funneled back in to R&D and other projects. At this point AMD can't claim the best server performance scores across the board, as the Intel's Core 2-derrived Xeon chips can muster more overall performance, but the situation is such that the Xeon is being hobbled by the FSB and FB-DIMMs, which increases AMD's relative competitiveness. Furthermore, even with their older K8 processors AMD has been able to offer very good performance the performance-per-watt category, a metric of particular importance to datacenter operators who are using blade-style servers and other setups where there are practical limits on how much heat can be dissipated.

Q3 saw the launch of the Barcelona design, a particularly important change for AMD. They can now offer chips that have better overall performance than their old chips, and they have a quad core design to compete with the quad core chips Intel has been offering for some time now. This has closed some of the gap that has formed over the last year as Intel outpaced AMD, and as a result Q4 is going to be a big quarter for AMD as it will be the first full quarter with the Barcelona shipping. Unfortunately Barcelona isn't a new Intel killer, at current clockspeeds it's still underperforming Intel's best chips, and we don't know how much that is going to change.

Intel has the 45nm advantage, meanwhile AMD was unable to get a suitable amount of 65nm K8 chips to clock as high as the older 90nm K8s, so we're a bit concerned on just how good AMD's 65nm process is. This doesn't preclude the possibility that K8 scaling problems were due to design issues and not process issues, but AMD has never released enough information one way or another that we can put our worries to rest. For now AMD is in a respectable position, they can't offer the best server performance, but it's good enough for the current market. And while it's entirely plausible that their server division could do very well in the long term even if they were always behind Intel by a small amount, it's not an idea that is a good one to test. Simply put: AMD needs to raise its clockspeeds enough to completely compete with Intel if they want a guaranteed recovery in the server markets. For Q4 at least this isn't going to happen, and we're going to be looking at 2008.

Moving on to consumer processors, it's virtually the same story. The first numbers for Intel's latest high-end 45nm Penryn parts is very good, meanwhile we aren't expecting AMD to be able to match those parts with the Phenom line this year. Much like Barcelona, Phenom is still an important launch because it closes the gap, but AMD doesn't have other technological advantages here like they do in servers to offset this gap. Right now the Phenom launch won't push the company back in to profitable territory, but it will enable AMD to raise their average sales prices and cut some of the loss. Phenom's effect on AMD's situation will be much greater once it has shipped for a full quarter in 2008. Q4 however is seeing the first full quarter of AMD manufacturing at 65nm exclusively which is helpful for margins. With the shutdown of Fab 30 AMD no longer has any 90nm production facilities (although partner Charter may still be able to produce 90nm parts).

As for the graphics side of AMD's business, how Q4 will go may as well be anyone's guess. We know AMD will be launching a new line of GPUs this quarter, but we don't know how many, or with what kind of performance. In total performance the X2900XT was not enough to beat the GeForce 8800GTX, which deprived AMD the chance to sell extremely high-margin parts. The GPU industry, more so than even the CPU industry, is heavily biased in sales at the low end, so AMD was fortunate in that they missed out on comparatively few high end sales. This is a market where it's entirely possible to focus on volume sales and reach profitability, without the need to compete at the very high end of the performance spectrum. AMD's graphics division was after all almost profitable for the quarter, so no significant changes are required to reach at least some level of profitability.

But in the future AMD risks falling in to a trap similar to the R&D costs trap of the CPU industry if they aren't developing the high-end technology that will filter down in to lower products. With that said much of this rides on what their next generation of GPUs can do. We have no reason to expect AMD to be comparatively any worse off than they were with the Radeon HD2900XT, so what we're really looking as is if they can improve and by how much. A GeForce 8800GT killer is a must, so we're looking forward to seeing such a product hit the market.

We are slightly concerned however that they haven't made a particularly strong effort to push the GPGPU aspects of their GPUs. Like server CPUs, GPUs in a high performance computing GPGPU environment can fetch high margins, and rival NVIDIA has made a very focused effort to capture this market. We'll have more on this issue specifically in a future article, but for now it's an oddity that AMD hasn't taken better advantage of this market.

Wrapping things up, this quarter will see the launch of the first of AMD's new 700 series chipsets to coincide with the Phenom launch. Chipsets are a great revenue source to go with CPU sales, so it's to AMD's advantage here to pair up more of their chipsets with their processors, instead of letting NVIDIA have such a big piece of that market. AMD's embedded division will be seeing a lot of fanfare with the first major Geode-using OLPC shipment going out this year, but as the OLPC is strictly a non-profit venture this won't immediately help their situation. However it could spur for-profit sales, which would be a good thing for the company.

Closing Thoughts

We won't attempt to sugarcoat AMD's Q3 and say it was any kind of recovery quarter; it was simply the latest in a string of bad quarters for the company. AMD is in a price war with Intel it is so far winning, but is costing the company too much right now. Furthermore the acquisition of ATI has proven to be very costly, and the direct costs of the acquisition and the borrowing for it are not treating AMD well. They can't have too many more of these quarters, and while things are going to get better on the business side as loans and other costs are repaid, Q4 will more than likely also be a losing quarter for AMD.

The best AMD can hope for in the rest of Q4 is cutting their losses, and this is something that is going to be executed through their technology. The Barcelona Opterons and the new Phenom processors can't reverse the company's fortunes in this quarter, but they are closing the gap with Intel, raising AMD's margins, and breaking the cycle of the painful price war AMD has been in. AMD will always be at some technological disadvantage compared to Intel due to Intel's better manufacturing process technology, which means AMD needs to be all the smarter with the advantages they do have, and Q4 will be about using those advantages. Meanwhile depending on what AMD releases for GPUs this quarter, their graphics division may finally start turning a profit for AMD, further absorbing AMD's other losses.

It's admittedly painful to see AMD in their currently situation. As consumers, things are best for us when both AMD and Intel (and AMD and NVIDIA) are at each others throats, and this isn't the case right now. AMD isn't going anywhere yet, but AMD needs to be at Intel's throat soon if they ever want to do it again.

AMD By The Numbers
POST A COMMENT

27 Comments

View All Comments

  • loa - Friday, November 09, 2007 - link

    What many people today fail to realize is that the single most important factor driving CPU performance increases is the manufacturing process. The manufacturing process is as most people know crucial for the clock-frequency and power consumption. The main function it has is that for every new process generation the transistor budget doubles, which gives opportunity for better performance. What I know AMD has always lagged behind Intel in process technology, typically 6 to 12 months, and thus beeing 6-12 months after in performance.

    Of course they can alleviate this if the microarchicture design is superior. Sometimes AMD:s microarchitecture is better, sometime Intels, but in the long run they are probably quite even. What decides who will win is then process technology. If AMD won't be able to out get new process technologies no later than Intel, they will in the long run, on average, also lag behind Intel in terms of performance.

    So all this talk about platform and microarchitecture is important, but the real issue is the process technology.
    Reply
  • yyrkoon - Saturday, November 03, 2007 - link

    quote:

    It's admittedly painful to see AMD in their currently situation.


    Last page, last paragraph. I think the word you're looking for is 'current'.
    Reply
  • magreen - Saturday, November 03, 2007 - link

    Wrong word? LOL. The whole article was so replete with spelling mistakes, grammatical errors and bad writing, I thout I was reading xbitlabs, not AT. Anand, please do something about this! It's the second article in a row! Reply
  • magreen - Saturday, November 03, 2007 - link

    Thought... Typing on a blackberry and can't see what I'm typing! Reply
  • strikeback03 - Tuesday, November 06, 2007 - link

    lol, so maybe they can make the same excuse. Reply
  • TA152H - Saturday, November 03, 2007 - link

    The 8086 hasn't even been out for 30 years, how can AMD be building it for over 30 years? They were building it before Intel invented it??? If they can do that, surely they can survive little annoyances like financial losses.

    AMD is in a price war they are winning? Are you crazy? They are losing massive amounts of money, while Intel is making enormous profits. Who do you think can sustain the price war, the one losing money on a huge scale, or the one making it? I'll give you a hint, the company losing massive amounts of cash is not the one winning a price war. They are the big losers. Let me guess, Germany and Japan won WW II, right?

    AMD has been competitive for a long, long time, not just recently. They did make clones, sort of, and were in fact behind by Intel by a generation most of the time. But fundamentally, the computer business was a lot different 20 years ago. You still have chips like the 8086 being released in new machines, even though it was already nearly 10 years old. That was a processor TWO generations behind the 386 (186 and 286 were designed at the same time, even though the 186 was released a little sooner). More than that, AMD's processors were considerably better than what Intel made, on a generational basis. For example, Intel stopped at 12.5 MHz for the 286, AMD went up to 16 (some say 20, but I have never seen one). The AMD 386 was much better; it ran at 40 MHz (Intel stopped at 33 MHz) and used much power than Intel offerings, and was an extremely successful processor. AMD went up to 133 MHz with the 486, Intel stopped at 100 MHz and didn't even want to sell them too much.

    They were not competitive in terms of performance, but they were extremely competitive in overall value (particularly the later 386 versions), and many companies used them.

    The K5 was a pure AMD design, but the initial version was something of a disaster for them, since they disappointed Compaq with the performance and time of release. Intel retaliated against Compaq too, so it seriously bruised AMD's reputation. The later K5s were not too bad, except in floating point, but they were usurped by the NexGen designed K6, which had somewhat less IPC but could run at considerably higher clock speeds. It was a competitive processor with the Pentium II and Pentium III, especially the K6-III which ran faster on integer apps clock normalized by a good bit. It was considerably smaller, and used MUCH less power than the Pentium III. So, it was clearly superior in some ways, although suffered from poor floating point, and lower clock speeds compared to the Pentium III. The problem was more the platform than the processor. There was simply no good chipset for the Super 7. The MVP3 sucked, and the Aladdin was miserable too. I still think they should scrap the Barcelona for the mobile, and update the K6 and use it. The K7 was a terribly inefficient chip, and the Barcelona is just two iteratives removed from that. The K6, by comparison, ran circles around the K7 in terms of performance per watt. It would seem a more natural starting spot for a design. But, AMD probably lacks the resources, so they keep putting out rubbish like the Turdion. An updated K6 would be so attractive in the mobile space.
    Reply
  • Mana211 - Monday, November 05, 2007 - link

    Would it be so hard to fact check before spewing idiocy?

    http://www.amd.com/us-en/Weblets/0,,7832_10554,00....">http://www.amd.com/us-en/Weblets/0,,7832_10554,00....

    1969
    AMD incorporates with $100,000; establishes headquarters in Sunnyvale, California

    1970
    AMD introduces its first proprietary device: the Am2501 logic counter

    1972
    AMD goes public

    1979
    AMD debuts on the New York Stock Exchange
    Production begins in new AMD Austin manufacturing facility

    1982
    At IBM's request, AMD signs an agreement to serve as a second source to Intel for IBM PC microprocessors

    1984
    AMD is listed in "The 100 Best Companies to Work for in America"

    Now if all you want to talk about is x86 then sure that has only been going on for 2007-1982=25 years, but ignoring the 13 years they were in business before x86 became important to them is just sticking your head in the sand.
    Reply
  • Calin - Monday, November 05, 2007 - link

    The K6-3 used to run with three levels of cache (two in the processor, just like the Pentium 3, AND one in the mainboard). The cache on mainboard was 512kB or 1MB (I have a Soyo mainboard with 1MB of cache, which would be second level cache with a K6-2 - as it is now - or third level cache on a K6-3).
    My mainboard uses some ETEQ chipset - which, strangely enough, is recognized as VIA if you installed the VIA 4-in-1 drivers.

    One more thing - the K6-3 might have had better performance per watt, and better IPC than the K7 - but K6-3 (which is the part with lvl1 and lvl2 cache AND is the part to compare K7 against) was available in 400 and 450MHz versions - and not any faster. I have no idea how much could you have overclocked one, but in the end K7 reached more than 800MHz. Hmmm, I wonder if the performance of K7 at 800Mhz would be higher or lower than the performance of K6-3 at 450MHz.

    And by the way, if Intel now is making 64-bit processors for laptops, AMD would be forced (feature to feature basis) to do the same, and only K8 was 64-bit.
    Reply
  • Spartan Niner - Saturday, November 03, 2007 - link

    "The rumors of my death have been greatly exaggerated"

    -AMD
    Reply
  • Regs - Saturday, November 03, 2007 - link

    Financial accountants, outside agents and firms, sales, marketing, etc...they all had this forecasted before AMD bought ATi years in advance with many of the "what if" variables calculated in every forecast.

    I hope everything is going as predicted over there at AMD, and hope they can swim in the squals for just a little longer.
    Reply

Log in

Don't have an account? Sign up now