Original Link: http://www.anandtech.com/show/1532
Industry Update - Q4-2004: AMD adds SSE3 Support, Intel's 925/915 not selling and moreby Anand Lal Shimpi on November 3, 2004 1:00 AM EST
- Posted in
Every year I make at least one trip to Taipei, Taiwan, usually for the annual Computex show. The flight itself is usually grueling, traveling from the East Coast you're generally in the air for around 20 hours. Then there's getting used to the time difference, which is a full 12 hours from EST. But it's all worth it, because a trip to Taipei is like a hardware-guy's dreamland. Tons of manufacturers spread out all over the northern tip of the island all working on bringing the latest technology and performance to your PCs. It's through these manufacturers that you can get a very interesting perspective on the industry as well as get a good idea for the truth behind a lot of the issues we see.
The Taiwanese motherboard manufacturers are the first hand recipients of roadmaps and future product information from companies like AMD, Intel, ATI and NVIDIA. The manufacturers are also privy to information that usually doesn't flow through a PR group before reaching them, so tapping our sources in Taiwan often gives us a much more honest (and bleak) view of the PC industry as a whole.
The other type of information we get from Taiwan is good updates on what types of products are actually selling. It's one thing to hear AMD and Intel talk about market share, but when the motherboard manufacturers tell us that a product isn't shipping, we usually know the truth.
I met with manufacturers for three days straight, usually from 8AM until as late as 11PM every night. And while I'm not able to share all of the information discussed in the meetings, I'll do my best to put forth a summary of some of the hot topics we talked about. But before I get to what the motherboard manufacturers told me, I'd like to touch on some of the questions they had for me and thus, for all of you. Just as we are at the mercy of the PR teams at AMD, Intel, ATI and NVIDIA, the motherboard manufacturers are at the mercy of the same folks when it comes to understanding what you all, the end users want.
The biggest question I was asked in Taiwan was about why I felt the 915 chipset wasn't selling well. I'll touch on this more in the chipset section of this article, but with Taiwan coming to us for answers you get an impression of the current situation.
The next question, or worry, on the minds of the manufacturers in Taiwan is the future of dual core technology on the desktop. This is another issue that I'll discuss later in the article, but you can understand the sense of caution if dual core is the number two question on their list.
A surprisingly popular question also revolved around ATI's upcoming chipsets. Next week we'll see the launch of ATI's latest AMD and Intel chipsets, but for the first time we're seeing an unusually large amount of interest from the motherboard manufacturers. This is yet another area I'll be touching on later in the article.
There are many other interesting tidbits of information I picked up while in Taiwan, ranging from Intel's 1066MHz FSB plans to AMD's first Athlon 64 chips with SSE3 support, so without further ado, let's talk about what's going on today.
AMD vs. Intel
The heated AMD vs. Intel debates of the web weren't present in Taiwan. Most motherboard manufacturers were actually surprised to hear that more users appear to be buying AMD today, simply because their shipments are still predominantly Intel based. As far as percentages go, the motherboard manufacturers unanimously agree that the number of AMD motherboard shipments today are higher than the overall 80/20 market split between AMD and Intel.
The advantage is still in Intel's corner, with the highest percentage we were quoted being that only 30% of all motherboard shipments were for AMD platforms. More than anything this is a testament to Intel's strength. While AMD has been performing very well and chipping away at Intel since the release of the first Athlon, it's going to take a lot more than AMD doing well to dethrone Intel.
The server end of things is also looking very strong for AMD, arguably even stronger than on the desktop side. This month will see the first public demonstrations of 8-way Opteron servers, which will be a first for AMD. The Opteron has not only given AMD the ability to compete in the 4P (soon to be 4 Socket) space, but now stepping into the 8P realm, Opteron is the biggest threat to Intel it has ever been. With the Opteron available in 8P configurations, AMD will have brought 8P processing power down to the affordability of a 4P system. Looking at CPU costs alone, a single Opteron 850 currently sells for around $1500, a single 3GHz Xeon MP (4MB L3) sells for $4000 on average. Multiply that by 8 and you've got a $20K difference in CPU costs alone. We've compared the performance of 4-way versions of the Opteron 8xx vs. the 3GHz Xeon MP and already seen that the Opteron is the better performer of the two thanks to the better scalability of its architecture. While we're still waiting for a review sample of an 8-way server, we'd venture to guess that the superior scalability of the Opteron architecture will only be amplified in an 8-way scenario.
It takes time to develop CPUs and platforms, especially for the enterprise market, so Intel can't respond with product to AMD's threat - at least not today. Instead, Intel is offering motherboard makers much more support to develop some of the next-generation Intel server platforms. From what we heard in Taiwan, Intel is being much more cooperative with the manufacturers in making sure that their next-generation (2005 and 2006) server and workstation platforms come along quite well. Intel has also expressed a great deal of concern about Opteron to the motherboard manufacturers we talked to, although you'd never get them to publicly admit that.
AMD has also put dual core Opterons in the hands of the motherboard vendors. While we couldn't get a confirmation of clock speeds, the motherboard manufacturers we talked to could not say the same for Intel. It seems that AMD is truly ahead of Intel when it comes to dual core.
On the notebook side of things, as we've reported in a recent Insider Story Intel wants to kill the "Transportable" (aka "Desktop Replacement") market segment. The goal being a Pentium-M and Celeron-M dominated notebook world, and also conveniently forcing AMD into a tight spot. The motherboard manufacturers we talked to who also happen to do OEM notebooks asked us more than once whether we thought the DTR segment was truly dead. Maybe Yonah in 2006 will change things, but today the DTR market is definitely alive and well. In fact, many of the Taiwanese manufacturers were confused as to why DTRs sold so well in the U.S., with ultra portable designs selling much better in Japan; after all, who wants to lug around a 10+ lbs laptop? Many of the DTRs today are so heavy that they are more like PC-iMacs instead of notebooks. Needless to say, Intel pulling out of the Transportable/DTR market opens that avenue up to AMD, which is what most of the manufacturers we talked to seem to think as well.
There is also a lot of talk about dual core among the motherboard makers; most of them are wondering whether dual core will be interesting enough to force an upgrade cycle on users. For years, Taiwan has depended on higher clocked CPUs and new cores to sell the latest motherboards, but for the first time in recent history, they aren't being given a higher clocked CPU to count on. The fear is that there won't be any tangible performance increase from dual core CPUs to desktop users and thus the next generation of Glenwood/Lakeport based motherboards (the first boards with dual core support) will see dismal sales much like today's 915. Given Intel's recent track record to these folks, with promises of a quick transition to PCI Express platforms, the caution and concern in the eyes of the motherboard manufacturers is easily understandable.
AMD Athlon 64 Revision E adds SSE3 Support
As we discovered most recently, AMD's 90nm CPUs are of a brand new revision that featured a number of bug fixes and performance improvements. Internally AMD refers to this CPU revision as Revision D. However we noticed something very interesting when looking at the latest AMD roadmaps while in Taiwan: the 90nm chips listed as San Diego and Venice (Athlon 64 FX and Athlon 64 respectively) claimed SSE3 support as a feature, but the current 90nm chips do not have SSE3 support.
We went around to quite a few manufacturers asking what the difference was between the 90nm chips shipping today and the 90nm chips that supposedly feature SSE3 support, and unfortunately we were left with very little information - until we learned to ask for the right thing.
Internally, AMD's San Diego and Venice CPUs are referred to as nothing more than Athlon 64 Revision E chips. Revision E includes even more bug fixes and performance improvements over those we found in Revision D, including support for the 13 new instructions that were added with Prescott, more commonly known as SSE3.
The performance enhancements that go along with Revision E chips include some optimizations to the Athlon 64's memory controller. The more optimized memory controller improves bandwidth efficiency with regards to unified graphics memory accesses; given that the only type of graphics that uses system memory (and thus the on-die memory controller) is integrated graphics, it's safe to say that the Rev E chips will offer better integrated graphics performance.
The first Revision E CPUs will begin shipping in early 2005. AMD also has plans to introduce an Athlon 64 4200+ towards the middle of 2005; they are not listing whether the part will feature a 512KB or 1MB L2 cache, but it will most likely run at 2.6GHz. The Athlon 64 FX-57 is also listed on the roadmap as a 2H-05 part, it's specifications are also unclear but we'd expect it to be a 2.8GHz part with a 1MB L2 cache. Both the Athlon 64 FX-57 and Athlon 64 4200+ appear to be 90nm parts built on the San Diego and Venice cores, respectively.
AMD's 2005 roadmap did not specifically list anything faster than the 4200+, although the classic "> 4200+" was present on the roadmap to indicate potentially faster parts.
On the Intel side of things we heard the name Cedar Mill thrown around a bit more. In the past some have referred to Intel's dual core Pentium 4 as Cedar Mill, but we now know that is Smithfield. This time we heard Cedar Mill referred to as the first 1066MHz FSB Pentium 4s that aren't Extreme Edition chips. Cedar Mill is supposed to be out in the 2nd half of 2005, which supports a claim we made in our recent review of the Pentium 4 3.46EE: "the 3.46EE will be followed up by the 3.73EE as the only two chips to support the faster FSB for almost a year."
So what does having no mainstream 1066MHz FSB chips until the end of next year mean? It means that Intel is holding out for Glenwood and Lakeport and that Sunday's launch of the 1066MHz FSB was merely a gimmick. The 925XE will be buried as the chipset that offered support for two CPUs that no one bought, while Glenwood and Lakeport will be the knights in shining armor that brought 1066MHz FSB support to all.
Intel seems to have learned from their 925X and 915 chipset launches - multiple fundamental technology changes without performance gains don't go over well. Glenwood and Lakeport will be able to support faster DDR2 (DDR2-667), 1066MHz FSB and dual core support upon their release, not to mention that lower latency DDR2 should be available by then (mid next year) and PCI Express graphics should be much more prevalent as well.
When Intel talks about aggressive ramps of new products there's usually little attention paid to exactly how they're going to achieve such an aggressive ramp in product shipments. For example, it turns out that the market for the 915 chipsets is very soft, so motherboard manufacturers are finding that selling LGA-775 motherboards equipped with the 865PE chipset is much more in tune with their customers' desires. The result of this is that shortages of 865PE chipsets will occur (such as the one going on right now) and the only way to get more 865PE stock is to not only purchase more 865PE chipsets, but also more 915 chipsets. It has to work this way because otherwise the motherboard makers would take much longer to transition to new platforms and that wouldn't do so well for chipset or even CPU sales, and since Intel holds the power in the business relationship things like this can happen.
The problem is that right now, no one wants 915 motherboards - they simply aren't selling well at all (925X boards aren't doing any better; I leave them out of this discussion because they are generally much lower volume boards, 915 is the mainstream product so that's what matters). The same isn't true for 865PE based motherboards according to the manufacturers, but in order to get more 865PE chipsets they must buy more 915 chipsets, thus it makes more sense for them to just try to do whatever it takes to sell more 915 boards - rather than buy twice as many chipsets and still have poor sales.
Why is 915 selling so poorly? There are a few reasons for the current phenomenon:
1) The US and worldwide economies are still soft.
2) In its "optimal" configuration, the 915 chipset requires DDR2 memory and PCI Express graphics - both of which are currently more expensive than the technologies they replace.
3) PCI Express graphics cards are still relatively rare on the market. The highest end cards are all being bought up by the major OEMs, and there aren't enough entry level and midrange cards to meet demand. Graphics cards also aren't cheap, convincing users to upgrade their motherboard, CPU, graphics card and memory all at the same time is an expensive proposition.
4) Intel platforms aren't as attractive today as they were when the 865 was announced - AMD is much more competitive in price and performance.
5) Take all of the reasons above and keep in mind that even if you can make it through all of those issues, you still end up with the fact that Intel's 915 chipset doesn't really outperform the 865PE - the 915 quickly becomes one tough pill to swallow.
Help is on the way, but boosting 915 sales by the end of 2004 appears to be a lost cause. Soon there will be no more Socket-478 Intel CPUs left on the market, with the only remaining chips being LGA-775 based. Here's one of those aggressive ramping situations. By getting rid of Socket-478 Pentium 4 CPUs, Intel ensures that the only Pentium 4s you'll be able to buy are LGA-775 chips. When you buy a LGA-775 CPU you'll most definitely need a new motherboard, and with most LGA-775 motherboards being 915 based, there's a high likelihood that you'll find yourself buying a 915 motherboard as well. Then of course you'll need DDR2 and a PCI Express graphics card, so the memory and the graphics card makers benefit as well. But the chain reaction will take place tomorrow, as far as a solution today goes, unfortunately for the motherboard manufacturers - it's going to be a tough few months.
The next problem is that once 915 sales begin picking up next year, 915's replacement will be just around the corner - adding 1066MHz FSB and DDR2-667 support. I wouldn't be too surprised if Glenwood and Lakeport get pushed back to late in Q3 of 2005 rather than towards the middle of the year to at least somewhat better accommodate the motherboard manufacturers. Now you can begin to see why introducing a 1066MHz FSB variant of the 915 wasn't an option for Intel; with so much unsold 915 inventory, the motherboard manufacturers would be in a very difficult situation if they were given a 915E to sell as well.
On the AMD side things are much simpler; just about every single motherboard manufacturer has a nForce4 solution for AMD as their high end Athlon 64 platform. In fact, NVIDIA is quickly turning into the Intel of AMD chipset manufacturers, which is something we've been asking for ever since the introduction of the Athlon.
Although there is a lot of support for ATI's upcoming chipsets (you'll read about them here next week), almost all the manufacturers were saying that their ATI products will be Intel-only. The worry is that with such a strong competitor in the Athlon 64 realm that their ATI products won't sell; there's also a lack of confidence about ATI's ability to supply their South Bridges. Whether or not the fears are well founded, none of the motherboard manufacturers expressed much interest in an ATI Athlon 64 chipset just yet. We'll see what happens next week, there may just be a few changed minds.
VIA is still quite present on the Athlon 64 motherboard roadmaps, however the chipsets are only being used as entry-level or mid-range solutions, the high-end appears to be completely dominated by the nForce4. What's even more interesting is that this is without even talking about the nForce4 SLI chipset; the motherboard manufacturers appear to be quite happy with NVIDIA's latest chipset in any incarnation possible.
When we look at nForce4 SLI, as we mentioned in our most recent article, the chipset will only be shipping in two motherboards this year from ASUS and MSI. NVIDIA has been going around the US demonstrating the ASUS nForce4 SLI board, and our most recent article was done exclusively on MSI's board.
The rest of the motherboard manufacturers will have to wait until December at the earliest to begin talking about their SLI motherboards. And that's just talk, from what I've seen don't expect to see any nForce4 SLI boards from anyone other than ASUS and MSI until next year. NVIDIA would very much like for the nForce4 SLI to become their mid-range chipset, found in motherboards priced at the $150 mark instead of closer to $200, with more manufacturers making boards next year it may be that by the middle of 2005 you'll be able to pick up some bargain nForce4 SLI boards at $150 or less. Until then you can expect prices to be in the $180 - $200 range at best.
Desktop Pentium-M Motherboards
A few months ago I put Dothan (90nm Pentium M) to the test and compared it to an equivalently clocked Athlon 64 and a high-end Pentium 4. In general application performance, a 2GHz Pentium M actually outperformed the desktop chips and even in gaming and workstation applications the Pentium M was competitive, all while running at significantly lower temperatures with much lower power requirements.
The problem is that the Pentium M, although electrically uses the same bus as the Pentium 4, has a completely different pin-out, preventing it from being used in desktop Pentium 4 motherboards. There are also other voltage requirements that most desktop motherboards (and chipsets) cannot meet that prevent the Pentium M from being used as well.
It didn't take long for motherboard manufacturers to put a mobile chipset and a compatible socket on a motherboard and thus while in Taiwan I saw two of the first shipping desktop Pentium M motherboards with AGP support.
AOpen and DFI both have motherboards ready, and are both targeting the Japanese market first. DFI built their board for a particular customer and is planning an enthusiast level board based on the desktop 915 chipset with some overclocking features in the near future. We know that Shuttle has been working on a SFF based on the Pentium M for quite some time now but have yet to see anything from them.
While Pentium M processors are still priced significantly higher than desktop CPUs, the value is in the lower power consumption and cooler operation - so in '05 there may be another, quite attractive option for cool and quiet PCs.
ATI vs. NVIDIA
Just a couple of years back and a motherboard manufacturer producing both ATI and NVIDIA graphics cards would find themselves kicked off of the nForce2 partner list. These days things are very different, most manufacturers are producing both ATI and NVIDIA cards thanks to around 12 months of NV3x slip-ups. That being said, there are still incentives for a manufacturer to only deal with one of the two companies. For example, ATI is only willing to share a certain amount of information with their partners if the partners in question happen to produce NVIDIA cards as well. If you are ATI-exclusive, then you get more information. Makes sense, right?
Despite having access to more information, there's little I could find out about what's coming next for ATI and NVIDIA. Despite close relations with their partners, ATI and NVIDIA can keep their board vendors in the dark for much longer than Intel can, for example. The reason being that most of the time ATI and NVIDIA simply hand their vendors a reference design, and the vendors do nothing more than duplicate the relatively simple design. In some cases with NVIDIA, the board vendors must purchase both the GPU and the memory from NVIDIA, leaving very little work other than assembly for the board manufacturers to do. Combine that with very short product cycles and you can see why there's not much information floating around.
The big discussion in Taiwan about ATI and NVIDIA is with the current state of PCI Express. Without a doubt, Intel did a wonderful job of convincing all of their partners that the transition and ramp to their 915 and 925X platforms would be extremely aggressive. The reality obviously was completely different, as 2004 has almost come to a close and the only demand that we've seen for the latest PCI Express platforms comes from the OEMs themselves. Intel's efforts are very evident from the most recent product releases we've seen from ATI and NVIDIA; the GeForce 6600GT and X700 XT were both PCI Express-only parts initially, with AGP versions due out in the coming weeks.
The supplies of PCI Express graphics solutions from both ATI and NVIDIA are limited as you've heard over the past several weeks, and unfortunately there were no signs of a fix from the manufacturers in Taiwan. We received varying explanations for the shortages - with everything from low packaging yields to demand miscalculations. There is light at the end of the tunnel and most manufacturers were optimistic that PCI Express graphics supplies should pick up in the new year, until then we can expect sparse availability of the more desirable PCI Express GPUs.
The memory market is pretty unchanged from what you'd expect, the strengths are definitely in DDR sales right now. With poor 915 and 925X sales and no AMD support for DDR2, there are simply not enough platforms out there to drive up DDR2 demand. The only real demand for DDR2 comes from, as you can guess, the OEM platforms.
We would expect better adoption of DDR2 in 2005 as more 915 and 925X platforms make their way into the hands of more and more end users. Towards the end of 2005 Intel will have DDR2-667 support on their platforms, and without AMD here to force an early adoption of DDR2 standards we can expect that Intel's upcoming Glenwood/Lakeport platforms will be the only driving force behind DDR2-667 adoption.
I hope you enjoyed this little update on the industry, if you'd like to see more of these types of articles just let us know - the flight to Taiwan isn't too bad and the information is usually top notch :)