by Andy Yaschenko
01/08/2004 | 11:51 PM
This time it was Intel who won the annual race. It owes this victory mostly to the newcomer – the Centrino platform. The Pentium M was the only processor made from scratch in the last year and it proved excellent both in performance and power consumption. The combination of the two is absolutely winning.
<%BANNER[article]%>But let’s be unbiased, though. Intel had enough of mobile processor series in 2003 (Pentium 4-M, Mobile Pentium 4, Mobile Celeron), among which the Pentium-M could easily and completely get lost. So the Centrino platform owes its glory particularly to the Wi-Fi unit and those $300 million spent on promotion of the “notebook with an integrated wireless network” concept.
A couple thousand hot-spots around the globe by the end of the year is a drop in the ocean, but the concept appealed to the customer (with a little help from with the massive advertising campaign), and the rate at which those hot-spots are popping up everywhere is increasing. Don’t also forget the fact that the well-designed processor architecture allowed to reduce dramatically the power consumption/heat dissipation parameters, which immediately resulted in the smaller dimensions and longer battery life of notebooks. So we’ve got a unanimous customers’ award, considering the sales volumes. Meanwhile, AMD has nothing to pit against the Centrino right now, and is unlikely to have anything in near future.
As soon as Pentium-M moves to 90nm production technology, its parameters will get even better, although this very 90nm technology was the main failure of the last year for both Intel and AMD. Intel is closer to solving this secret: the Dothan (90nm Pentium-M) will surely come out in the first quarter (although had been expected to arrive in the fourth quarter of 2003!), but the Prescott may be delayed until the second quarter as it is rumored to have problems with both the technology process and the 800MHz QPB when working with i865/875 chipsets.
AMD is down and out as to far as the manufacturing technology is concerned. Originally the company scheduled 90nm processors for the fourth quarter of 2003, but now they are not promising good news until the middle of 2004. I think Intel will surely have launched its Prescott by that time. Until then, AMD can only rely on the superiority of the Athlon 64 architecture over the Pentium 4 in their 130nm and 90nm implementations, respectively.
Well, there is nothing wrong about the Athlon 64 itself – it is a cute processor, although the half-a-year delay (capping the previous procrastinations) couldn’t have earned it a good repute. Such a delay sometimes proves fatal (as is often the case in the graphics processors world). Anyway, the bunch of technologies sealed into the Athlon 64 turned to be impressive enough for Intel to consider it the most feared competitor to the yet-unreleased Prescott rather than to currently available Pentium 4 models on the Northwood core.
The advantages of the architecture are numerous: integrated memory controller, 1MB L2 cache, improved integer and floating-point units, SSE2 support. I don’t mention such a defining property as 64-bit memory addressing, since this feature is still unavailable for an ordinary user (I don’t count Linux users in). It is a very powerful 32-bit processor until Microsoft wakes up with its 64-bit version of Windows.
AMD’s processor was really much faster across a number of applications than the current Pentium 4, so Intel’s reaction was simply required. Intel’s move was quite a surprise: they released a Pentium 4 Extreme Edition on the Gallatin core (the core the Xeon MP is based on). Roughly, it is the same Northwood with minor changes, but equipped with 2MB on-die L3 cache working at the full core frequency. Well, it was a kind of compensation for the high memory operations speed the latest AMD product boasts.
Of course, there would be no need for the Pentium 4 XE, if everything were as it was supposed to be, with the Prescott materializing in Q3 2003. Intel had to act extempore, using the best x86 130nm core they had at their disposal. The future of the Pentium 4 XE is vague enough, but I doubt this impromptu is going to enjoy a good long life. At best, if the problems with the Prescott persist, we will see a higher-clocked Pentium 4 XE (like 3.4GHz instead of 3.2GHz), before the Grantsdale chipset family and the Socket-T Pentium Something (I guess it is still going to be “4”) come out in spring.
On the other hand, if Intel likes the sales volumes of the Pentium 4 XE, we may see this series continued as a Prescott with an enlarged cache.
If the Pentium 4 XE was the surprise of the year in the product field, the main surprise in technologies was Intel’s announced intention to jump from 193nm lithography to EUV, skipping over 157nm equipment and production processes. This news was a shock for the equipment makers as they had already invested heavily into their 157nm products. The whole industry was in an uproar as Intel has enough weight in the technological field.
Intel’s position is logical, although not quite obvious. EUV development always encounters various difficulties, so it is not at all certain that everything will work like a well-oiled engine when the technology is really called for. And they won’t have 157nm technological process as an intermediate variant. Still, I hope all is going to be well: Intel will solve all problems or agree to 157nm technology without much noise, just in case.
While it is more or less clear with Intel, the situation with AMD looks more like a mess. The last year was quite characteristic in this respect: the company got divorced from UMC, which had been chosen as a partner to develop the 65nm production process and to build a joint fab in Singapore. AMD took the problem apart and dealt with the splinters independently. In the technology field, the company doesn’t rely on itself anymore, having licensed 130nm and 90nm technologies from Motorola, and stopping by IBM for the 65nm one. That’s a nice choice, and the company is solid and respected, although NVIDIA and Xilinix have been complaining about IBM’s 90nm production lately. That’s a disturbing signal, on the other hand.
As for the fab, AMD had to look for one by itself and proposed Fab36 (rather than Fab35, as it has been announced before). As you know, the number stands for the year since the foundation of the company, so we’ve got a year-long delay from the original plan. Adding 36 to the year AMD was founded in (1969), we get 2005. That’s the year when the new fabricating facility should be put into operation near Dresden, in the German land of Saxony. It is far from East Fishkill, New York, although that variant was also seriously considered. Money won again: the German offer was more enticing. I really wonder why Infineon wants to change its headquarters?
But back to processors, to our ordinary, desktop processors. Speaking about the Athlon XP and the Pentium 4, though, we’ve got no sensational news. The system bus frequency grew up a little bit: to 400MHz for the Barton based Athlon XP and to 800MHz for the Northwood based Pentium 4. The processor clock-speeds went higher, too. The topmost Athlon XP is now 3200+ rather than 3000+. The Pentium 4 is 3.2GHz instead of 3.0GHz.
That’s quite an unprecedented thing, although quite understandable. Both Athlon XP and Pentium 4 processors on their 130nm cores have practically exhausted their frequency reserve, while the 90nm technology process never succumbed to either of the companies in 2003. I don’t think we should be sad about the slow progress; it’s rather good since both companies had to win their customer with the price factor. The all-out price wars of 2003 brought us price cuts by 30-40% in spring and fall. So, although the frequency cap has gone just a little up, we’ve got much more performance for the same money by the end of the year.
Interestingly, it was the Celeron that added in frequency more than anybody else, from 2.4GHz to 2.8GHz. This is also quite reasonable as it has to fight for its place under the sun with low-end Athlon XP models. By December, Celerons of 2.4-2.6GHz frequency cost roughly the same money as an Athlon XP of the same performance rating, but of course their performance is simply incomparable. By the price-to-performance ratio, mainstream processors from AMD looked perfect in the end of 2003, while the Pentium and the Celeron don’t even stand close to them.
For those who fancy “value” products, AMD presented a surprising, but anyway very pleasing gift in 2003: the new generation of Durons that were in fact Athlon XP processors with a smaller cache-buffer (64KB). Yes, they have the appropriate performance, which is quite enough for many chores. But they have an appropriate price, too. About $35-40 for 1.4-1.6GHz models!
The Athlon 64 had settled in the price range of the Pentium 4 by the end of the year, but its availability is still low: half a million items shipped during the fourth quarter. Intel found itself in a quite unfamiliar situation when its Pentium 4 3.2GHz costs about 10% cheaper than AMD’s Athlon 64 with the 3200+ PR.
There are some unpleasant moments about the new AMD processors, though. First of all, there is a confusion with the sockets (the Socket754 for the Athlon XP and the Socket940 for the Athlon 64-FX), and the Athlon 64-FX is in fact the Opteron 1xx with another marking. This kind of a disorder has never been considered an advantage. Even if we take the “pure” Athlon 64, there are questions concerning the future development of the series. In fact it had been planned keeping in mind the upcoming release of the 90nm production technology in the end of 2003. As the technology hasn’t yet been finalized, the processor was rated as 3200+ rather than 3400+, and there are grave doubts about the remaining frequency reserve of the 130nm core and further core frequency growth.
In 2003 AMD looked more advantageous in the server processor field (that was downright strange and weird). Still, this is an undeniable fact: Opteron practically met its schedule and the manufacturers welcomed it warmly. Among them are IBM, which is now AMD’s partner, and Tyan, which is not a minor player in the server and workstations market. This market is a good playground for the Opteron with its 64-bit architecture. With the added support of eight-processor systems, AMD processors look very attractive from all points of view compared to Intel Xeon.
So, Sun has already announced its intention to use Opteron in its x86 servers and will port Solaris for the x86-64 platform, while cluster manufacturers are complementing the list of the world’s highest-performing machines with systems based on Opteron. The unanimous acceptance brings money in: having launched Opteron x48 in the end of the year, AMD priced the series from $733 for the 148 model to $3199 for Opteron 848.
Meanwhile, Intel has no ready ersponse. The Xeon is basically not 64-bit. And the traditional solution of price reductions (by one third in July) pushed AMD to do the same with its Opteron. Intel also boosts the size of the L3 cache; it is now 3MB in the two-way Xeon DP. The Itanium is a specific and narrow market due to its unique architecture. This explains why the Opteron sales soon volumes after its release grew much higher than those of Itanium.
At the same time, it is the Itanium 2 family that’s going to resist the onslaught of the Opteron (that is, if Intel doesn’t come up with their own 64-bit x86 processors, as rumors suggested). The family was developing quite nicely during the last year, with the release of 130nm processors on the Madison core (a 0.13-micron version of the McKinley) and the Deerfield (a “light” version of the Madison). The latter is really very light: 1GHz Itanium 2 LV with 1.5MB L3 cache requires only 62W of power and comes at an official price of $744. Who said the Itanium means bulky and pricey?
It is the opposite end where Intel meets no competition (at least, from AMD) – miniature processors for smart-phones and PDAs. The XScale has grown mature and in the beginning of 2003 Intel announced its abandoning the oldie StrongARM SA-1110 it had received from DEC. This processor seemed quite a piece of antiquity against the Manitoba – XScale with its integrated flash and DSP, raised to conquer the smart-phone processor market. The Manitoba is expected to arrive in the first quarter of 2004, so we have to wait just a little. Still, it is good that there is something to wait for. For example, there is no hope left for AMD’s Au1100, since there appeared not a single PDA model on this processor.
Now that we have discussed the positions of the two companies in the market and sorted out their product lines, it’s time we summed up the information in numbers.
Intel
The fall of 2003 is most remarkable as Intel’s processor, chipset and mainboard departments reported record-breaking results. The company said that this was mostly due to the growing markets in Russia, India and China.
AMD
In the end of 2002 AMD underwent serious restructuring (its cost accounts for the high loss), but its effects told quickly on the company’s profits. The launch of the Athlon 64 in the fall, and the Opteron’s success helped AMD to see some quarter profit.
I guess there is no need to draw graphs as AMD’s share of the x86 processor market increased by 2% only in 2003 (but it increased anyway!), but I will present you the rankings as of the fall 2003:
In the end of 2001, however, AMD had 18% of the x86 processor market. I think Athlon 64 may help the company to regain their position.
Those remaining 1.6% belong to the “kids”, Transmeta and VIA. Yes, there are still two of them, although Transmeta lost about $90 million in the last four quarters with sales volumes amounting to $20 million (per year!). Well, what do they have for sale?
TM5600 is not interesting anymore now that we have Centrino. Transmeta’s response to Intel – TM8000 called Efficeon this year – didn’t reach the stores in 2003. Transmeta once again promised to make the processor available in the third quarter of 2003 and did ship… sample chips in September. They are 130nm of course, but Transmeta made arrangements with Fujitsu to produce 90nm Efficeons starting from the second quarter of 2004. Samples of 90nm chips are expected to appear in January.
VIA Technologies is doing somewhat better. Early in 2003, the company released the long-anticipated Nehemiah core as a 1GHz Socket370 C3 processor with SSE support. In spring they presented the Mark project, which is a single chip including the CLE266 North Bridge and the C3 core. A new mainboard form-factor was announced to support this chip – Nano ITX – with dimensions of 12x12cm. They also sampled the mobile C3 (Antaur). What’s most important, all of this was launched into mass production throughout the year. VIA also fosters high hopes for the 90nm production technology.
Once the chipset maker Number 1, VIA Technologies at last pocketed its pride and got reconciled with Intel. The quarrel started a few years ago, but is now perfectly over: all past sins are forgiven plus they agreed on cross-licensing for the future. VIA took a long time to count up its losses and understand that it’s much cheaper to pay Intel for licensing rather than suffer drooping sales volumes.
VIA managed to enroll into the next round of technological wars Intel had started with its i865/i875 in spring. I can’t deny that after Intel wound up its affair with Rambus, it again bakes up excellent chipsets. Both series were embraced by the manufacturers and ended 2003 with a status of the market leaders. Talking about the mass market, it is the i865PE, although the other models were quite confident in their respective market niches, too.
For example, the i865G with the integrated graphics core (and average performance) became the OEMs’ favorite, while the i875 was quite successful where it had been supposed to: in the workstation market and in systems for enthusiastic users. Suddenly, the i875P faced severe competition from i865PE! Mainboard manufacturers found a way to enable Performance Acceleration Technology (PAT) in i865PE although it had been advertised as the i875P’s unique and distinctive feature. The sales of the expensive i875P were under a serious threat.
Intel was furious at the sign of the riot so the manufacturers had to come up with their own names. As for the winner of this naming contest, I guess MSI should be awarded for its “Memory Acceleration Technology – MAT”. Intel ultimately gave up and banged the door saying they didn’t recommend buying such mainboards from its own partners. PAT was rooted out of the i865PE completely.
However, this is not the only trimming this chipset model underwent in its history. Notwithstanding all its advantages, it was quite expensive at first, before traditional price cuts, and didn’t meet Intel’s expectations concerning the sales volumes. The junior member of the 865 family, the i865P chipset, proved to be of no use for the market. So a new version of the i865PE was presented in the fall with a disabled second memory channel. The new i848P, an ideal mainstream offer, resembled the old i845PE, and differed from it mainly by the support of 800MHz system bus used in the new Pentium 4 processor models.
Intel took its time with licensing the new bus, so there appeared rumors about Intel having no plans to license it to any of the Taiwanese manufacturers. In April, however, the pause ended abruptly as Intel settled its argument with VIA Technologies. The cross-licensing they agreed upon, gave VIA the access to the 800MHz QPB.
Immediately VIA pulled up prices for all of its now-legitimate chipsets so that the price reflected the increased self-cost (because of the licensing fees), and also renamed them (the company played with chipset names throughout the entire year). Anyway, those chipsets belonged to days long gone, and some interest to VIA’s products arose only with the announcement of the PT800 (that was a good old P4X400 with the 800MHz QPB support). As a result, big sharks like MSI or Soltek issued their mainboards on the VIA chipset for the first time in years.
The PT800 is an alternative (quite interesting due to its much lower price) to the i848P, but for VIA to compete with the i865PE, they needed a dual-channel chipset with the 800MHz QPB. This they planned to present in May-June as samples of the PT880. In fact, the chipset only showed up in September, but without much sensation. Probably, VIA had problems implementing the SerialATA interface in the VT8237 South Bridge – there was something wrong there. As a result, first mainboards on the PT880 somehow made it into stores by the end of the last year only.
Now that the legal problems are over, there is hope that the next generation of VIA chipsets for the Pentium 4 will arrive according to the schedule. As for VIA’s processors, there are no problems at all and they are quite active in producing mainboards for them as well. The EPIA platform, which is now the focus of VIA’s VPSD division, has been developing quite rapidly with an interesting result: the Nano ITX form-factor (12x12cm). I am eager to see how it will develop in 2004 as I am sure this form-factor is destined to grow popular.
Curiously, the 55xx platform family from SiS never made a foothold in the market as the EPIA did. The basic shortcoming of the platform is the weak x86 processor core borrowed from Rise. As for ordinary chipsets for ordinary PCs, SiS is quite good at building them. On the other hand, SiS got a license for chipset technologies for the Pentium M CPU and already showed two models: SiSM661MX and SiS648MX. MicroATX mainboards with the Mobile Pentium 4 appeared last year, so we may even see desktop systems on the Pentium-M this year, who knows.
SiS651 remained OEMs’ choice and SiS replaced it in 2003 with the SiS661 supporting 800MHz FSB (and with the SiS661FX including an integrated DirectX 9 graphics core of their own design). Yes, SiS was long negotiating with Intel on the bus licensing terms, but quickly settled the matter after the news that VIA had already got it. The company managed to be the first to produce a Pentium 4 chipset supporting 800MHz FSB. It was SiS648FX, a modified version of SiS648. That’s quite a well-done chipset, accepted by many mainboard makers.
Other products from SiS were quite successful, too. They are two improved versions of the dual-channel SiS655: the FSB800-supporting SiS655FX and the SiS655TX that appeared in the end of 2003 and supported accelerated memory access mode (something like PAT from Intel), which makes it a competitor to the i875P. The above-mentioned chipsets enjoyed a steady demand, too.
SiS undertook a few other attempts to get accepted in the high-end sector. SiSR658 was mostly an image-making product, and SiSR659 announced in 2003 is the more so. They announced it as many as four times! By the end of the year, this setup with the RDRAM-supporting chipset looked like a farce rather than anything serious. By the way, Intel stopped taking orders for the i850E and i860 – the Rambus affair is all over now.
RADEON 9100 IGP was much more real than SiSR659. ATI’s second attempt to design a chipset appeared at least partially successful. Although the chipset was released without the South Bridge, it looked good enough. Some middle-range mainboard makers, and ASUS and Gigabyte tried to use it in their products (as an addition to ATI’s GPUs?).
Their third attempt may be even better, although it’s going to meet a tougher competition. There are three basic topics: DDR II, PCI Express (South Bridges), and transition to the Socket T. These three features are expected in a bunch in the spring of 2004 when the following chipsets should come out: Grantsdale, SiS656, PT890, ATI RS400. They all support DDR II and DDR as well as the PCI Express. The Grantsdale adds Socket T support, others are uncertain on that matter, but anyway they could implement it in next chipset revisions.
I doubt that we will get a serious performance gain from first DDR II modules and PCI Express expansion cards compared to current solutions. Faster solutions will appear later. We will also see the next generation of integrated graphics in North Bridges. They are all going to be DirectX 9-compatible with a performance of current MX440-RADEON 9200.
The last year brought considerable progress in South Bridge building, although the potential of the new solutions was not always used to the full. ICH5/ICH5R, VT8237, SiS964 – they all offer SerialATA, RAID 0+1 and 5.1 audio. Well, the latter thing is rather about audio codecs, but integrated sound traditionally refers to South Bridges.
Unfortunately, all these new capabilities have had different fates. While many manufacturers are already implementing SerialATA ports as there are SerialATA drives in the market, RAID is less popular (maybe it deserves it). As for 5.1 sound, mainstream mainboards often come without it or without a bracket with the S/PDIF output.
In the new year, and in the new South Bridge generation, we are likely to see the integrated Gigabit Ethernet becoming a standard. That’s most welcome, as higher bandwidth may reveal new uses for local networks. Next, I sincerely hope for the new audio codec from Intel – the Azalia. I also hope that the manufacturers will slow down on 7.1 audio and wait until demand catches up with the supply, i.e. when 7.1 audio systems and 7.1 sound in applications (especially in games) get widespread enough.
Talking about South Bridges, NVIDIA was most unlucky with them in 2003. They were promising us a South Bridge for the nForce2 with SerialATA support, but never released one, and they also did the same with the nForce3 Pro 250 chipset that was to differ from the nForce3 150 in this exactly SerialATA support. Well, by an odd coincidence, VIA Technologies also spent much time implementing SerialATA in their VT8237 South Bridge.
By the way, these two companies, VIA and NVIDIA, were grappling in 2003 for the shrinking market of Athlon XP chipsets. Of course, AMD tried to help them the best it could offering the Athlon XP with a 400MHz system bus in spring, which required new products from chipset makers, just like with the 800MHz QPB for the Pentium 4. NVIDIA got the upper hand, showing its nForce2 400 Ultra, quite a cute chipset (well, they actually had to add FSB400 support into the official specs, as this support had already been unofficially implemented in the nForce2).
VIA was quite confusing about the generations of its chipsets, announcing at once the KT400A (the KT400 plus PC3200 support), and the KT600 (the same plus FSB400). Of course, the manufacturers who announced mainboards on the first chipset later waited until the arrival of the second one. However, KT600 never enjoyed the popularity of KT400. So, here NVIDIA wins the round, no doubt.
NVIDIA also feels quite confident in the chipset market for the Athlon 64/Opteron where the nForce3 Pro 150 seems more prominent than the VIA K8T800, although it is hard to say that any of them is a definite leader. It seems more like equilibrium with a certain advantage on NVIDIA’s side. The competition should get fiercer in 2004 with another round of chipsets: the nForce3 Pro 250 (nForce3 Pro 150 + SATA) against the K8T800Pro (HyperTransport sped up to 1GHz plus Socket939 Athlon 64 support). This time NVIDIA will be somewhat late as it is going to roll out a Socket939 chipset later.
SiS may jump into the fight, although it looked less brilliant in 2003 in the market for both: Athlon XP and Athlon 64. SiS741 (a thorough revision of the good old SiS740) passed by without anyone noticing it, while SiS755 for Athlon 64 was delayed and unassuming. The first mainboard to use it appeared in October, and you could count up all mainboards on this chipset with the fingers of one hand. On the other hand, SiS announced SiS755FX (with Socket939 support), and scheduled SiS756 for the coming spring (PCI Express support). I think the new year will bring some improvement in SiS’ relationship with AMD.
Well, I suppose some facts discussed above can be better illustrated with a diagram:
The graph emphasizes such events as VIA’s shutting up its production of its own mainboards for the Pentium 4 in May when the SiS’ sales exceeded VIA’s. There is a noticeable improvement in the financial showings of VIA after it began selling “legitimate” products. The disturbing slump by the end of the year is caused by delays with PT880 and the active competition from NVIDIA in the Athlon 64 and XP chipset market.
I’d like to end up this section of the article nominating the Technology of the Year Award. The choice is limited and thus is simple: Performance Acceleration Technology from Intel as it provoked much sensation (and nothing more). The technology of the year, which is to “fire” in the next one, is the new case form-factor, BTX. There have been too many changes in the computer world for old ATX to adapt itself to them.
The year 2003 was absolutely black for memory makers. The SARS epidemics and the Iraq war were accompanied with a catastrophically low demand. They hoped for the dual-channel chipsets, for the end of SARS, for the back-to-school season, but they never saw any melioration in the market, only a total and hopeless downfall.
Some glimpse of hope shone through the clouds in June, when the SARS epidemic was over and the new mainboards hit the stores, with the sweet September looming ahead. It transpired, however, that the demand never reacted to all of this, and the prices went down again. The price of a 256Mbit DDR SDRAM chip fell from $6 to $4 during 2003. And that’s even not as dramatic as the things we used to see in the past.
The prices for different specifications of DDR SDRAM memory were quite curious to watch: PC3200 didn’t enjoy any significant demand in 2003, unlike PC2100 and PC2700. As a result, its price stopped at $4-5 for a 256Mbit chip and remained there thereafter. Junior specifications were more demanded, so their prices grew up until they hit the ceiling of PC3200 and stopped at that: of course junior memory types couldn’t cost more than PC3200! So we watched the three lines converged into one oscillating languidly throughout the year.
Yet another small sensation: the price of SDRAM was increasing in 2003 so that PC133 SDRAM is now more expensive than PC3200 DDR SDRAM. This absurd situation is well grounded: SDRAM is widely used in consumer electronics, which is selling quite well, unlike computers.
It was all more interesting with the modules. Of course, memory modules price dropped along with the memory chips price. Quite expectedly, the module manufacturers didn’t like the situation and started a show we saw coming back in 2002. It is called “overclocker modules”. All started out quiet with something like “PC3500” or PC3200 with improved timings. At first, only specialized firms like Mushkin, but then Kingston and Samsung, got involved into this production.
Then the technologies continued improving, and we saw modules like “PC3700”, “PC4000”, “PC4200” and “PC4400” appearing. “Older” modules like “PC3700” also improved their characteristics: extremely high timings became normal. Plus, such modules often come in pairs: for dual-channel memory configurations. Moreover, they have cool looks: gold or platinum coating on aluminum or copper heat-spreaders. Summing up the results of the year, I would name Corsair the winner in this race. By the end of the year, there opened another niche – registered modules for the Athlon 64 FX. This is a specific product for the last year, as we hadn’t seen registered modules in the consumer market before.
Meanwhile, the DDR2 standard was steadily developing without any extravaganza or sensations. In the process, it transpired that DDR2 looks much similar to DDR, so the transition to DDR2 is going to be less advantageous than the SDRAM to DDR SDRAM shift. Anyway, all leading manufacturers are ready to start mass production of DDR2 chips with capacities ranging from 256Mbit to 2Gbit and frequencies of 400-533MHz (PC3200-PC4300). These have all already been tested by Intel for compatibility with the company’s upcoming chipsets.
It’s not quite clear, though, what will become of 667MHz PC5400 as Intel never produced a Pentium processor for 667MHz FSB (and that was the reason for this memory to appear). In all probability, this memory will never show up, and the manufacturers will skip it and pass to 800MHz PC6400. The only uneasy fact with regard to PC3200-PC4300 is how well the packaging and testing infrastructure is ready for DDR2. It would be quite a stupid situation if the manufacturers do ship chips in necessary quantities, but the testing and packaging firms cannot cope with the amounts as they haven’t invested into necessary equipment. This way we could get a grave shortage out of nothing!
As for the manufacturers, I have no doubts about them: they have long and easily moved to 0.10-0.11micron production process necessary for DDR2 chips manufacturing. Even Hynix raised up enough money for the shift (Hynix was feeling surprisingly well in 2003, by the way). But as for particular names, let me first present the following table to you, Semico Research’s version of the annual results:
According to the table, both Samsung and Micron lost something of their market shares in 2003, although Samsung shouldn’t be dismissed altogether – the company just switched to flash. Micron, received a nice gift from Intel as the processor giant bought Micron stock for $450 million. Micron just has to invest the money in a smart way…Infineon skyrocketed up but is unlikely to repeat the trick in 2004. Hynix lost its seat in the first trio, but has a chance of returning: the company has been showing very nice dynamics lately in spite of all the protective duties in the European Union and the USA. Nanya won’t hold to its fifth place – they were hasty in breaking apart with Kingston, which was a billion-dollar client. ProMOS, Elpida (who enrolled Intel and Kingston as its stockholders) and Powerchip are likely to go up in the rankings: their ambition is expanding along with their production facilities.
As for our High Life Chronicles, the quarrel about ProMOS was something unheard of before. The year 2003 started with Mosel Vitelic’s attempt to grasp control over ProMOS in the board of directors. Infineon got it back in court, but lost its patience on the way. So, the Germans started selling their ProMOS stock and withdrew the license for their production technologies. After that, ProMOS said they wanted to be friends with Infineon in the future and had no claims at all and were ready to call back their suit, if Infineon started buying its own chips from ProMOS. Quite cynical, as our modern age goes.
Infineon of course put a full stop at that, and ProMOS had to dispose those chips they used to sell to the Germans. In summer, the Taiwanese negotiated with Elpida the technological collaboration and selling up to 10 thousand wafers to Elpida monthly. It seemed quite a logical variant, but in October it was revealed that ProMOS was building its new 300mm fab without any help from the Japanese company. This was recognized as a rift in the relationship.
However, the information dated December about ProMOS finding another partner, Hynix, was much of a surprise. Simultaneously, ProMOS announced its purchase of 50% of Mosel Vitelic and of the patents and technologies related to memory and flash. Overall, this is an active preparation to the company’s great goal scheduled for the second quarter of 2004: selling 100% of produce under their own brand. If the plans come true, the ProMOS sales volumes will grow in 2004 no less than in 2003.
The last sensation of 2004 has to do with Rambus. For the first half of the year, the Japanese fans of the company (I really wonder why the Japanese like Rambus so much – maybe some personal relationships?) licensed the well-designed high-speed serial data-transfer interface aka Yellowstone. They were Sony, Toshiba, Elpida…It means we may one day see this interface implemented in the Cell processor and, accordingly, in the PlayStation 3. And Rambus will keep on raking the licensing fees in.
In the second half of the year, Elpida and Toshiba announced a new memory architecture called XDR DRAM, which is of course Yellowstone-based. They talk about fantastic frequencies (3.2-6.4GHz) and bandwidth (up to 100GB/s). This is not pure sci-fi: in December, Toshiba did show a 512Mbit 3.2GHz XDR DRAM chip. So, this memory seems to be technologically plausible, just like RDRAM was a few years ago. They just need to convince Intel of the necessity of making a second try.
ATI undoubtedly won the last year. Just look at the following graph that shows the company’s stock price throughout 2003:
A few comments seem necessary, though. It seems like NVIDIA inherited the karma of 3dfx, having consumed it in the end of 2000. It took just a year for the company to lose its first place in the computer graphics market to Intel and allow ATI Technologies to get very close (NVIDIA’s 25% against ATI’s 22%. Compare to 31% and 19%, respectively, in the beginning of the year). It would have seemed purely fantastic just a couple of years ago – ATI reached NVIDIA in sales volumes and outpaced it in profits!
NVIDIA ended the last quarter with a sales volume of $486 million and a profit of $6.4 million, while ATI with $469.7 million sales and $47.4 million profit. As for the big picture, IC Insights predicts NVIDIA to end the year with $1.835 billion sales (-4% compared to 2002), and ATI with $1.135 billion (+76%). To make the story complete, ATI only has to surpass NVIDIA in annual sales for 2004. Considering the dynamics of the last year, the Canadians can do it.
ATI has learned the secrets of making good graphics chips, of writing normal drivers and of doing business together with partners. For example, such giants as ASUS and Elitegroup joined ATI’s camp in 2003! The name of ASUS alone in the list of companies manufacturing graphics cards on ATI’s VPUs can bring success to ATI in 2004. It’s a meaning fact that ASUS used to produce cards on NVIDIA’s GPUs and couldn’t afford the luxury of getting into an affair with ATI. When there is no clear leader in the field, NVIDIA just cannot command its partners, so we are waiting for MSI to roll out RADEON-based cards under its trademark, too.
There is one more proof – the Gigabyte story. When the company started making graphics cards on ATI processors, NVIDIA refused to ship GeForces to them, but later changed the attitude and Gigabyte returned to making GeForce-based cards once again, without breaking apart with ATI, of course. Again, this would have been quite impossible, if the market situation hadn’t changed so dramatically.
Ah, yes! I can’t also pass by the fact that Microsoft chose ATI to supply graphics chips for the XBox 2, saying that they “selected ATI after reviewing the top graphics technologies in development and determining that ATI’s technical vision fits perfectly with the future direction of Xbox”. They only don’t say if it is it financial or technological direction? Anyway, a close relationship with the developer of DirectX never did any harm to a graphics company. As for the rumors that NVIDIA would develop and ship the graphics core for the PlayStation 3, they were soon discarded by people at Sony saying they could do it themselves all right.
NVIDIA saw the trouble coming in early in 2003 when the long-delayed NV30 (GeForce FX 5800) GPU family came at last out. Those chips were overall slower and more expensive than RADEON 9700 that had been around for quite a while already. It was all so bad that NVIDIA had to release hastily a new NV35 chip (GeForce FX 5900) that replaced NV30 completely. As for GeForce FX 5800, the company tried to muffle the matter, removed technical briefs from their website and so on. Of course, this schedule didn’t allow inventing anything radically new, so NV35 was in fact a bit more improved version of the NV30: it had 256-bit memory controller and a number of new technologies to allow faster Pixel Shader calculation and variety of additional speed improvements.
ATI, however, had a long-prepared answer – R350 core (RADEON 9800/9800 PRO). Both high-end VPUs from ATI and NVIDIA provided the same performance for about the same money. This parity was only reaffirmed in the fall as the slightly overclocked versions of the same graphics processors, namely RADEON 9800 XT and GeForce FX 5950 Ultra, don’t differ in performance and price. On the other hand, they didn’t go far in performance from RADEON 9800 PRO / GeForce FX 5900 Ultra…
RADEON 9800 PRO
GeForce FX 5950 Ultra
It was rumored last year that ATI and NVIDIA agreed on slowing down the technology race. Well, some slowdown of the progress is evident. For example, the new chips from both companies (R420 and NV40) were scheduled for the fourth quarter of 2003, but were then synchronously postponed for the first quarter of 2004. Anyway, the graphics hardware has left the software developers so far behind, that it’s quite possible to skip one or two cycles without anybody noticing it.
The war in the upper echelon was nothing compared to the mess in the mainstream sector. Two new DirectX 9-compatible generations hit the market simultaneously: RADEON 9600 and GeForce FX 5600. Then they started their “do-as-I-do” game: ATI uncovers RADEON 9600 PRO benchmark results and NVIDIA raises the frequencies in its FX 5600 Ultra.
In the fall, the overclocked RADEON 9600 XT meets the overclocked GeForce FX 5700 Ultra. However, they are both confronted with the RADEON 9800 PRO and the GeForce FX 5900 that have come down from the top. Overall, a hell of a mess, and I hope the new year will bring us fewer, but better chips.
RADEON 9600 PRO
I also hope for a clearly-defined product policy as we witnessed in the good old times when the top-end graphics cards featured the latest generation of chips, the mainstream cards – the previous generation, and the low-end products had ancient chips in them. New products appeared less frequently, but were more balanced and comprehensible for both: users and developers.
And there was no need in those brave times to cripple new products into “cut-down” versions in such haste that it turns out no difficult task to bring them back to the original state. In 2003, ATI was spotted in two cases of the kind when users found a way to “upgrade” the RADEON 9500 into a 9700 or bring RADEON 9600 SE into its pristine and faster self.
As for the low-end solutions, we had some calmness there, as they cannot be made into anything slower. MX440 and RADEON 7500 were complemented with RADEON 9200 and GeForce FX 5200. These were faster, although MX440 co-existed with them in 2003. RADEON 7500 was evidently an obsolete chip, so ATI had to release RADEON 9200 SE, which exactly fitted into the niche of the MX440.
To end the last year, NVIDIA acted up with its MX4000, which is an enhanced version of MX440-8x with a redesigned memory controller (to support the today’s popular 16Mx16 128Mbit memory chips). Graphics cards on the new chip should cost somewhat less. It’s now hard to tell the future of the innovation. MX440 is still quite popular, and NVIDIA just has to organize the transition smoothly (in prices and shipments).
As for the future, I have already mentioned the upcoming graphics processor generations, NV40 and R420, and I do hope that we will see some new technologies there rather than the same frequency race. At the same time, it is clear that the new generation of gaming engines (Doom 3, Half-Life 2) won’t use those technologies, and all games scheduled to come out next year will be based on these engines at best. As usual, software is lagging behind hardware in the computer graphics market.
By the way, R420 and NV40 are going to become the last representatives of the AGP-compatible chips. Other solutions will use the PCI Express bus, not only the top ones like NV45 and R450, but also mainstream NV41/42 and RV370/380. In the lower level NVIDIA has some NV33 core on its roadmap, which is to become an AGP 8x solution. There is going to be another mess about AGP/PCI Express graphics cards and AGP/PCI Express mainboards, I am afraid.
Let me now say a few words about the other developers as there are other players in the graphics chip market, besides ATI and NVIDIA. Both XGI and S3 are set to enter the computer graphics market, and there is some evidence we should trust them. XGI has made more success so far, as there are already graphics cards on their chips, namely the top-end Volari Duo V8 Ultra.
Lower performance is not interesting anymore. This solution is as fast as RADEON 9600 Pro that appeared a little bit less than a year ago. Meanwhile Volari Duo V8 Ultra costs $450, like a RADEON 9800 XT. However, there is some reserve: the drivers will be improving and the prices will surely go down with the launch of new cards. On the other hand, the dual-chip solution itself seems questionable.
But at least XGI has something real to offer to the public and they have met their schedule. As for S3, it has a smaller reserve of faith. In the beginning of the year they promised DeltaChrome X9-based graphics cards of the same performance level as RADEON 9700 PRO and GeForce FX could show then and costing from $90 up. Then they promised the cards to show up in November-December, having showcased a PCI Express prototype at IDF, though.
In December, instead of the promised cards, they offered another updated roadmap, with January as the official launch date. The promised performance is now equal to that of RADEON 9800 and GeForce FX 5900, and the price has been risen up to 155 euro. Why euro? Because they chose the main manufacturer of the cards – the same Club-3D that has already started selling cards on the XGI chips. Overall, it all looks quite uncertain, but I cannot dismiss them altogether as BitBoy, since S3 showcases some samples or prototypes every now and then. I do hope the new year makes things somewhat clearer.
At a perfunctory glance, the main topic of the past year in the hard disk drive market was about the transition from Ultra ATA/133 to SerialATA. In fact, it is a very strange transition: it is still not clear if it has happened already or not. On the second thought, this may be a sign of an ideal smooth change of the standard when you cannot pin a point and say this day we all used Ultra ATA/133 and the next day we all used SerialATA. And there is one more transition ahead – to SerialATA II. And that is going to be also slow. This interface will show its strength in the current year already, while the year 2003 saw only prototypes of 3.5” and 2.5” SerialATA drives.
On a deeper level, the last year was about the increasing data density per platter and its consequences. To be precise, the data density has been increasing since 2002.
In 2003, they discovered that a combination of cobalt with platinum wafers increases the energy of magnetic anisotropy that directly affects the data density, in hundreds times. Hitachi announced a technology that’s going to boost HDD capacities by about hundred times by the year 2010. In other words, we will have consumer drives of a few terabytes capacity. By the way, Hitachi, in cooperation with Maxell, also announced a technology to create optical disks of the standard 5.25” diameter and of 1 terabyte capacity somewhere around 2007. But that’s rather not about hard disks.
As for the off-the-shelf products, the manufacturers were all busy with product series on 80GB platters. Theoretically, they could offer us 480GB HDDs in 2003 already. They didn’t, because hard disk drives, just like processors, have outpaced the current level of the software and the users’ requirements. What would a user need a 0.5TB HDD for? And there are 100GB platters approaching. The Barracuda 7200.7 200GB on such platters is already selling. Characteristically, Seagate used only two platters. The reason is simple: we don’t need more.
No wonder then, that reducing the platter diameter while keeping the same storage capacities becomes the most urgent topic. For example, 2.5” HDDs of 60-80GB capacities is an ordinary item in the product list of any notebook hard disk drives manufacturer. And parameters of such products match those of “grown-up” models: 5400-7200rpm spindle rotation speed, 2-8MB buffer and so on.
Add the Nano ITX form-factor and the RADEON 9800 with passive cooling…Yes, for the Centrino platform, the performance of the HDD should be no worse than that of an ordinary 3.5” drive. By the way, in the end of the year, Fujitsu showcased the prototype of a 2.5” HDD with the Serial ATA II interface. These are all parts of the same picture, just like Seagate’s release of its 2.5” Momentus family and Western Digital’s intention to get to production of 2.5” models. Hitachi has been manufacturing them for a long time (the Travelstar series was inherited from IBM). There is only Maxtor left.
While the manufacturers of full-size HDDs shift towards the 2.5” form-factor, the manufacturers of notebook drives are ousted to the 1.8” format. Of course, Toshiba is in the lead, proving its superiority with a 40GB 1.8” drive: 4200rpm speed and 15msec access time, but everything else is within the norm – Ultra ATA/133, 2MB buffer, and (most important!) a power consumption of only 1.4W. The company predicts the market of such drives to become worth 25 million items in 2006. Besides Toshiba, the Chinese GS Magicstore aspires for a share in it, although it has only a 4.8GB model of the 1.8” form-factor.
This capacity suits well to a 1” Microdrive: Hitachi, for instance, has a 4GB model, although mercilessly priced – just a little under $800. However, there appeared competitors in this niche that offer much more enticing prices for a megabyte. They don’t yet offer anything above 2.2-2.4GB, but the mere fact that there is competition, is nice.
Let’s also see how the 0.85” HDD form-factor proposed by Toshiba will be developing. The company targets it for mobile phones, as these devices can well use a HDD of 2-3GB capacity. And they also already use graphics chips from NVIDIA and ATI. We won’t see a ready product soon, though. Toshiba is only promising to start selling drives like that in 2005.
It is interesting that with all those records in data density and capacities, such important parameters as access time and data transfer speed are lagging behind. The only news during the past year was Western Digital’s announcement of the Raptor series – consumer HDDs with 10,000rpm spindle rotation speed. Other HDD manufacturers haven’t followed WD, though.
This is strange, since the interfaces allow higher data transfer speed. Take SerialATA II or SerialSCSI as examples. When testing their controller, Adaptec reached 5Gbit/s instead of the specified 3Gbit/s! But what can be done with such high speed? They can only continue enlarging the cache buffer more and more…
Winding up the discussion about HDDs, I would like to draw a parallel with the computer memory market. While the stagnation in the memory market resulted in a number of super-mega-extra modules, we have external drives here. There are numerous external HDDs announced last year, although the number of internal devices is smaller than usual. So, this is a way for the manufacturer to remind the public of themselves and also grope for a niche to make money in. The interfaces can suit everyone’s taste, as USB 2.0 and FireWire often come together. The trend led to NAS systems supplied as an external HDD with an Ethernet controller.
CD-RW drives have practically exhausted their growth potential, too, having stopped at 52/32/52 speeds combination. Again, we’ve got numerous external devices and CD-RW/DVD-ROM combos or CD-drives with an integrated card-reader. It is already clear that several suggestions about doubling the data density for a CD disc didn’t make it into the market: the falling prices for DVD-burners have rendered such initiatives useless.
The prices did plummet. During the fall, volume prices for one-format 4x DVD-drives fell from $220 to $88 in Taiwan. Multi-format devices got a new volume price of $160-180. 8x DVD-burners came down from the original $160-180 mark to $140-150 after a month of being around. No wonder Sony preferred to leave this market, handing over its DVD+RW drive manufacture to Taiwanese OEMs who had earlier dropped the prices on entering the market (earlier in 2003, Yamaha left the CD-RW field for similar reasons).
Let me give you more detail from now on. The last year started out with 4x DVD-R/RW and DVD+R/RW (and multi-format) drives that had just appeared. After that, there was a long silence – until the end of summer. The formats were testing each other’s strength. Summing up the results in the end of the year we can say that the “+” format won as it is now more popular among the manufacturers.
As for the speed factor, 8x models came into shops in the fall, and 12x ones are expected in the beginning of 2004. 16x models are scheduled for the end of 2004 – Philips already has a working prototype. This may be the end of the race, as 16x speed is the physical limit, according to the technicians. A minor disappointment of the year: Philips announced a DVD+RW drive with the SerialATA interface at the spring IDF, but the idea has never been put into life.
Anyway, these are no big problems, as the future is tapping on the door. There is only one question – what future should we let in? We’ve got Blue-ray and Advanced Optical Disc: the former seems more real as Sony already showcased a Blue-ray drive and a VCR, while Matsushita demoed single-layer 23GB discs and dual-layer 50GB 2x discs (9MB/s). Theoretically, this can be purchased right now, unlike AOD systems from NEC and Toshiba, that are still only prototypes.
So far the two camps used DVD Forum as the fighting ground. Some innovations like Enhanced DVD (the current DVD plus enhanced functionality like links to websites with additional information) or adding a second layer to DVD-R discs (8.5GB in total, there is a similar proposal with respect to DVD+R) passed calmly to be implemented in 2004. HD-DVD was another matter.
The Blue-ray consortium has no need in a specification that describes discs of a 15GB capacity per layer (against 33GB in Blue-ray), as it is a direct competitor to their own product. NEC and Toshiba, who proposed it, need it to get a timeout for perfecting their AOD. Taking great pains, at a third try, after promoting their own members into DVD-Forum, the couple managed to push the specification on. It’s another question whether they will make it into a de facto standard.
