by Andy Yaschenko
12/19/2003 | 05:36 PM
In spite of the problems with the 90nm manufacturing technology, Intel met the Christmas sales season with waving banners, having announced the Pentium 4 Extreme Edition. The timing was perfectly chosen. First processors of the new series had reached the shelves by the end of November to be widely available in December. I can only find one fault about the new processor: it is not very new from the technological point of view being just a slightly cut down Xeon on the Gallatin core.
<%BANNER[article]%>Well, I just feel like fault-finding, but the result is all-important: the new 3.2GHz CPU with 2MB of L3 cache (added above the Pentium 4’s standard 512KB of L2-cache) is the most powerful Intel’s creation in the processors-for-mere-mortals class so far. It is arguable whether the 2MB of SRAM cost a few hundred bucks (the volume price is $925), but it is a fact anyway that some major PC manufacturers have already reacted to the new processor with announcements of game stations (I can’t call it a computer!) based around it. Just in time for Santa to have made his shopping before the holidays.
For those who already have a computer, the new Intel product should appear advantageous over the last AMD’s creation: you are not supposed to change your i865/875-based mainboard, memory and so on. The lifecycle of the new processor is expected to be long. It will probably be at the same performance level with first Prescott models, thus remaining a top model for at least half a year. Not topmost, as the 3.4GHz version of the Pentium 4 Extreme Edition (on the same core) is likely to come out somewhere in early 2004.
If the Pentium 4 XE 3.4GHz does come out instead of the Prescott 3.4GHz, this will imply severe problems with the Prescott project at large. The nature of problems remains secret: leakage currents, heat dissipation or, according to some fresh guess-work, incorrect operation with the 800MHz FSB of the i865/875 chipsets. By the way, if the latter supposition is true, we are very likely to see the Prescott in the Socket T variant only, while the Socket 478 niche will be left for the P4 XE to till up.
The Celeron project seems to be free from any problems, so Prescott-core Celeron 2.8 and 3.06GHz CPUs (a curious coincidence – they work with the 533MHz FSB) will surely show up in the second quarter of the next year. So far, the topmost representative of the family is the Celeron 2.8GHz (Northwood core, FSB400) that appeared in November. The volume price is quite traditional for a value product – $117.
The mobile Celeron series also rejuvenated in November with the arrival of the Mobile Celeron 2.5GHz with a maximum power consumption of about 35W. It’d be interesting to compare this parameter with those of the upcoming desktop Celeron on the 90-nm Prescott core. On the other hand, the 2.5GHz model is likely to be the last from its generation, which is to be substituted by a new family of mobile Celerons on the Banias core with frequencies starting at 1.2GHz.
The Celeron’s elder brother, Pentium M, is transitioning to the 90-nm Dothan core in the February of 2004. First samples work at 1.6-1.8GHz. The first 90-nm Ultra Low Voltage mobile processor, the Pentium M 1GHz, will appear around this time, too. It means that notebooks on the Centrino platform will work longer from the batteries, while Transmeta is up to hard times.
Other “big M” processors from Intel are looming ahead. The Madison-core Itanium 2 with a doubled L3 cache (9MB) is scheduled for 2004. Well, this is just a trifle compared to the dual-core 90nm Montecito with 24MB of the level 3 cache onboard. Interestingly, the Montecito was originally planned for 2007.
Well, it’s all natural: in spite of the periodically emerging problems, technologies are progressing. The above-mentioned progress is mostly due to the strained silicon technology employed in the 90-nm tech process. In 2007, Intel is intending to use polysilicon instead of metal as the material for the gate electrode, while the 1.2nm insulating layer of silicon dioxide will be changed for a 3nm layer of dielectric with much higher insulating properties. According to Intel, this will help to reduce dramatically the leakage current that has now become quite a problem in the industry.
That said, the company somehow managed to overcome the following problem (thanks to the metal gate?): when using high-k dielectrics today, the leakage current is reduced, but the operational current of the transistor is reduced, too, since the efficiency of the channel charging goes down. So, Intel promises us two-in-one: low leakage current and high operational current.
Taking the future at large, Intel seems to have chosen itself one: in the world of communications. November, Intel’s Craig Barrett said at his press conference that most of the company’s total R&D spending would go into this market sector. So the Centrino is just an initial milestone down a long road.
AMD doesn’t find anything useful about the Centrino’s wireless capabilities so far. Well, it’s easier to say Intel is fooling the market, when you can’t offer anything like that “foolishness” yourself. People who buy notebooks don’t think to agree with AMD. Added the ultimate Intel’s goal – control over the communications market – AMD may find itself in a desperate corner one of these years.
And I can’t even say AMD is feeling well today. The anticipated launch of 90nm processors never happened this year. Moreover, according to the latest news, it is not going to happen in the first half of the next year. Only the optimized 130nm version of Athlon 64 (Newcastle core) is awaited by then. That’s all. AMD confirmed this officially by telling in November that it’d be ready to manufacture 90nm processor in a couple of quarters. At its Dresden-located Fab30, of course.
November, the company decided at last where to build their new fab. There were two alternatives: Dresden’s outskirts where the Fab30 is located, and East Fishkill, New York, where an IBM fab is located. AMD has established close relationship with IBM, but they still preferred a time-tested variant – the Fab36 is to be erected in Germany in a couple of years. The Germans who have offered AMD credits for $1 billion must have managed to offer better terms and conditions than the Americans.
It’s all right, but the low cost of the fab seems dubious (a figure of $1.3 billion was voiced). It’s too cheap considering we deal with 65nm manufacture, not even 90nm. A bit later they estimated the necessary investment to $2.81 billion, and this seems more true to life. On the other hand, it was originally announced that AMD would build the fab in partnership with some unnamed company (IBM, Infineon?). As they already agreed on Dresden, they can register the Fab36 not as a fully autonomous manufacture, but as an expansion of the Fab30. Something like Intel’s Fab11X in New Mexico.
It’s also interesting that AMD is so very close with IBM in collaboration on the building of this fab that the tech processes on the Fab36 and IBM’s East Fishkill fab should be practically unified. This means that AMD will be theoretically able to lure customers away from the same IBM. I can well imagine a situation when AMD is manufacturing chips for NVIDIA, for example. The fab is already under construction.
These are perspectives, but what does the company have now? First of all, AMD lost 0.8% of its share in the x86 processor market in the last quarter; it is now 15.8%. It’s not all that dramatic in a longer term: AMD had only 12.4% a year ago.
Well, both Athlon 64 and Opteron were warmly greeted by the leading manufacturers. IBM started using the Opteron in its blade servers; November, Sun officially announced its intention to follow IBM and also port Solaris to the x86-64 platform. Hewlett-Packard also supported the Athlon 64, announcing November a series of Compaq X gaming systems on this processor.
Of course, this benevolent attitude should be maintained; that’s what AMD was busy with in November. At least, with respect to Sun and IBM, as AMD released the Opteron 148, 248 and 848 2.2GHz, priced from $733 to $3199. Well now, AMD can quote beautiful prices, too! There’s only one question – for how long? The 130nm Opteron has nearly exhausted its frequency reserve (just 0.2-0.4GHz left), while 90-nm samples are only going to appear in the second half of 2004. On the other hand, Intel is still grappling with its 90nm processors.
Talking about the new Opteron models, I can’t omit the fact that they now support registered PC3200 DDR SDRAM – previously, AMD said the Opteron couldn’t principally support anything faster than PC2700. The next step in this direction is going to be made early next year in the way of the economical version of the Opteron with a power consumption of 33W! Compare this to the above-mentioned Mobile Celeron 2.5GHz having a peak power consumption of 35W.
As for the Athlon 64, I can say close to nothing about it as it simply was not in November. There’s only some preliminary info about the Socket 940 Athlon 64 FX 2.4GHz and its Socket 754 analog, Athlon 64 3400+, coming in December. It must become a good repartee to Intel’s Pentium 4 XE 3.2GHz in a fight for the Christmas shopping. AMD’s processors should be one or two hundred dollars cheaper, to the bargain.
People at ASUS Computer should be proud of the last year’s results: 19.6 million mainboards in the three quarters only. Considering the fact that the company has reached an output of 3-3.8 million mainboards monthly, ASUS is likely to end this year with a record of 30 million shipped mainboards. Excellent management did the trick – just a year ago ASUS was far behind Elitegroup in numbers.
ECS, on the contrary, was quite confusing in its marketing decisions, with the ensuing results: 1.71 million mainboards sold in October against ASUS’ 3.78 million. And ASUS doesn’t intend to slow down a jot, as it is going to boost its mainboard shipments by 20% more. This means 35 million products per year, and ASUS’ one third of the world’s market. If the competitors don’t take measures, that is.
Elitegroup may try. Although forecast to have poorer showings in November than in October, the company claims to maintain and even push up the October bar as it has taken part of ASRock’s market share.
Characteristically, ECS also promises a growth of 20% for the next year, that is, about 20 million mainboards annually. Well, I have no great doubts about ASUS, but Elitegroup’s promises are another matter. At least, this year the company suffered a profit reduction of 76.4% in the first nine months compared to the same stretch of 2002!
Now, chipsets. VIA Technologies may strike a Taiwan-wide celebration. Having made it up with Intel, the company reported a profit for this quarter, however small, for the first time in months. It also boosted its sales volumes by 35.3% over the previous quarter. This can be viewed from the opposite point: it’s indicative of the company’s failure in the past rather than a success in the present. Other companies have increased their sales, too. For example, SiS had a 9% sales growth compared to the previous quarter.
There’re a few interesting things about the whole situation. VIA Technologies has become a sales leader, although questionable, while SiS outdid it in pure profit, in double. First thing to come into my head is that VIA is desperately dumping fighting for the volumes and the market share. There’s one more fact to confirm the point: Chinese manufacturers set FOB prices for PT800-based mainboards (it’s VIA’s analog of Intel’s 848P, as you know) at about $40-45. That’s the mainboard’s price, not the chipset’s one, of course. So the chipset should cost them about $15-17.
As for Intel and its chipset prices, the company is going to drop them again in December, so that prices for all i865 family chipsets and for the 845GE went down a couple of dollars. The relatively fresh i848P is to be selling for a dollar less, for $28. It seems like the planned preparations for the Grantsdale series chipsets (to replace the i865 family) are combined with a routine opposition to the pressure from SiS and VIA.
The i875P keeps on at its $50 price, as it is demanded for this money (in its niche). November, DFI reported a new version of its i875P-based mainboard, the LANPARTY PRO875B. The “B” signifies the availability of an onboard Gigabit Ethernet controller (in addition to the standard 10/100 one), a heat-spreader on the South Bridge, free space around the AGP slot for better cooling and around the processor socket for mounting a water-cooling solution, and a 4-channel SATA RAID controller. As usual, it’s all for enthusiasts.
Intel has prepared something to replace the i875P, too. The Canterwood (the codename for the i875) should be ousted out by the Alderwood in the second quarter of 2004. We’ll have the 800MHz FSB and dual-channel DDR-II, and no significant differences from the Grantsdale P (well, they mention “Turbo Mode”; the name reminds me of something…). On the other hand, there are no fundamental differences between the i875P and i865PE, as we’ve shown numerous times. November, Soltek offered one more confirmation of the point by releasing their SL-865Pro-FGR. The product is intended for entry-level servers and workstations, right where the i875P is targeted.
The top-end i875P has enough of competitors, besides Intel’s own lower-end chipsets, though. The notorious SiS R659 is doing its best to become one. The chipset was again announced by SiS in October, but this announcement was different. We are now able to see this four-channel RDRAM PC1200-supporting chipset in action as ASUS released samples of its R659-based P4S13G at last. They promise a 10% performance advantage over the i875P across a number of tests.
SiS is also competing with itself, releasing another product for practically the same niche. November, there appeared first SiS655TX-based mainboards, although the chipset is not yet officially announced! Anyway, the fact of the Gigabyte GA-8S655TX is substantial and pleasing. Just to remind you, the SiS655TX differs from the announced SiS655FX in its Advanced HyperStreaming Engine (the other has just HyperStreaming, without “Advanced”). Theoretically, the newer chipset should more efficiently handle peak workloads on various data busses inside itself.
After the peace treaty with Intel, VIA Technologies just couldn’t help responding with its PT880 with the same basic characteristics like the 800MHz FSB and dual-channel DDR400 support. Regrettably, the standard DDR SDRAM is only supported, as VIA still cannot realize the QBM memory developed by Kentron. In theory, such memory would provide a double growth of the memory bus bandwidth at no cost practically. So we should wait again, until the new revision of the PT880, expected in the beginning of the next year.
ATI is just making promises, too. Among the company’s current products, only the RADEON 9100 IGP chipset enjoys some acknowledgement from the manufacturers. Biostar announced the P4ARA mainboard based on that chipset, but the mainboard is not much in itself. As for the future plans, ATI promises to give us a new chipset codenamed RS400 in the second quarter of 2004: integrated DirectX 9-compatible graphics core, dual-channel DDR and DDR-II (?) memory, PCI Express support. This chipset is going to better ATI’s position in the chipset market, which is downright poor nowadays.
ATI Technologies is also going to test its muscle at building chipsets for the Athlon 64. It’s all the same: integrated DirectX 9-compatible graphics core, dual-channel DDR-II and PCI Express. Besides that, the company follows NVIDIA’s example and announces a version of the chipset without the integrated graphics. If ATI’s dreams come true, we’ll see all of this in just half a year.
NVIDIA is not sparkling in the chipsets market lately, too, especially in the calm waters of the Athlon XP chipset niche where the long-anticipated new South Bridge, MCP-S, has not appeared yet. The “S” stands for Serial ATA support, which is missing in the current NVIDIA chipsets making the mainboard manufacturers use extra onboard controllers to implement this function. The ABIT AN7 had been promised to use this Bridge, but it came out in November with the old one, MCP-T, and an integrated Serial ATA controller from Silicon Image.
ABIT AN7
NVIDIA has nothing to be proud of as concerns chipsets for the Athlon 64, while the competitors were more successful. SiS announced its first Socket 939 chipset, that is, for Athlon FX mainboards with the dual-channel memory mode. The chipset is called SiS755FX, and first Athlon 64 FX processors for the Socket 939 are only scheduled for the next year, so that’s a kind of “paper” announcement.
Elitegroup was more practical releasing its 755-A mainboard for the Athlon 64. November, these mainboards were spotted in the not-so-cheap Tokyo retail stores for about $100. Considering the rate of cheapening of the processor and mainboards, the Athlon 64 may become a reasonable option in a little while, especially if you forget that getting a 32-bit performance comparable to the Athlon XP and Pentium 4, you pay a lot for the 64-bit capabilities that you can’t use yet.
Anyway, NVIDIA may strike back as soon as December when the new chipset series for the Athlon 64/Opteron, nForce3 250, should be announced. They have been promising it from summer, but there have been problems with the Serial ATA realization. Strangely enough, this interface has become a real nightmare for the NVIDIA engineers. Overall, the new chipset will bring nothing particularly new with itself, differing from the well-known nForce3 150 in this very Serial ATA support only.
It’s a nice tickling feeling to see your forecasts coming true. I was repeating the first half of this year that there would be none of the back-to-school season, and the second half – that the market wouldn’t bloom up before Christmas. In theory, we should have witnessed the peak of the Christmas sales, but there was none. Moreover, I again disagree with Dataquest that promises a memory deficit for 2004-2005. There’ll be none of that as there is no demand for memory because there are no offers to stimulate the demand.
Well, Intel may find something at last (they are looking for ways to boost the demand for their processors, but that’s quite close). Or Microsoft may make a miracle of releasing the Longhorn (which is too raw so far) in 2004 rather than 2006. If there’re no such things, why there should be any deficit? Added a number of 300mm fabs to start production of memory chips in 2004-2005. There’s only one variant feasible: half of the manufactures get fed of all this and switch to flash, which is becoming ever more demanded product.
Overall, there have been no glimpses of sunlight in this market for long. As usual, the price of a 256Mb DDR400 chip went down resolutely from $4.7 to $4. There’s only one piece of good news for the manufacturers: the main demand seems to be shifting to DDR400. Well, it should be as the i865 has been here for quite a while already.
It was also as usual with SDRAM: the price was growing up, but slowed down to a halt in the end of the month. That’s all right. This is a normal behavior of a demanded memory type before a Christmas Eve. All involved made their shopping in November to be able to offer finished products before Christmas. This is mostly consumer electronics, which, unlike DDR SDRAM, enjoys a stable demand.
On the second thought, I may be a bit wrong about Dataquest. There is one and only variant (if we exclude wars, floods, earthquakes in Taiwan and other acts of God) when computer memory may become a deficit in 2004-2005. This may happen if the packaging and testing firms don’t cope with the amounts the manufacturers and the customers would like to deal with. The transition to the 0.15-micron tech process and 300mm wafers has outlined the problem. What will there be after they have all of their 300mm fabs working at full capacity and build new ones? The next year the manufacturers will shift to DDR-II and this process requires the testers and packagers of the new memory to reequip their facilities. This may become a real problem.
Under such conditions, those manufacturers who have better negotiating skills will win, not those with the biggest production facilities. So the market shares may change somewhat. So far, according to the third quarter results, summarized by Dataquest, Samsung’s inevitably declining market share shrunk from 35.6% to 27% in the first three quarters. That’s quite indecent for so short a term. And it’s not quite clear why. Well, it may be that the company that always knows which way the wind blows has transferred some of its lines to flash. The other plausible version is that the competitors’ activities made the company’s share in the South Pacific market decline, which resulted in the shrinkage of the total share.
Anyway, Samsung remains the leader with its 27%, followed by Micron that maintained its share throughout the year – from 18.3 to 18.4%. Of course, Micron cannot go much higher, but Samsung may meet it continuing its downward movement. Micron is doing its utmost to earn money (but not to expand the chip production). Particularly, they have started building a new plant for producing and testing memory modules, including DDR-II.
If there’s a company to aspire for the world’s first place, it is Infineon that has managed to grow from 9.2 to 16.6% in three quarters. This is partly due to the company’s advancement in China where it is at equals with Samsung already. Characteristically, the last quarter was the first profitable for the German company in two years – €49 million with a sales volume of €1.75 billion. Computer memory accounts for €765 million of the total, that’s 75% more than in the nine months of the last year and 34% above the showings of the second quarter. That’s what I call progress.
“Hynix” and “progress” are two words that rarely meet in one sentence nowadays, but even this company managed to keep its market share and even enlarge it from 11.4% to 14.9%. It’s very close to the third place in the world rank list. November Hynix was reported to consider licensing its 0.11-micron technology of DRAM production to the China-based GSMC under the standard formula “chips for technologies”. (It was also rumored in November that they intended to build a fab in China, but that’s not going to happen). It means there will be a second sturdy alliance a-la Infineon + SMIC (and Infineon is now doing well in China, as I’ve said above). Hynix will get one more advantage from the proposed collaboration in the way of additional elbowroom in its anti-antidumping measures. The measures also include a law suit filed by Hynix against the European Union to eliminate the protective duties of 34.8%, as Hynix claims the money loans it received from the Korean government were absolutely legal. Hynix has something to offer in the technology department, too. They have just started mass production of 256Mbit 0.11-micron DRAM chips in spite of all Micron’s insinuations about Hynix’ being practically unable to master the 0.11-micron process. Well, Micron has reasons to dislike the Koreans.
There’s also another sweet couple, Elpida and Powerchip, who are most ambitious and aspire for an entry in the “top five” list. They are far from it, but do what they can. November Elpida founded a new division to produce memory chips for consumer electronics (a tasty morsel nowadays) and called it “Digital Consumer Division”. And, as I’ve said earlier, the company announced cheap and economical SDRAM chips.
That’s all about the high life chronicles, and I am now supposed to cover some finished products. Well, I really doubt the purport of doing this. It’s all the same month after month: a couple of overclocker modules under the pseudo-standards like PC4000-PC4200 (this time they came from OCZ and the new one in the filed, SimpleTech, who unveiled its Nitro series) and a pair of SO-DIMM modules for notebooks and 1U servers (both from Kingmax, both are standards-compliant, PC2700 and PC3200).
512MB PC3200 SO-DIMM from Kingmax
One more company has started to sample new DDR-II-533 modules. This time it was Elpida with its 2GB models (it already has 512MB and 1GB variants, just like the rest of the companies). They say they can start mass shipments right now, but such a rush is unnecessary. So we wait until the Grantsdale comes, closer to the next spring. Anyway, this is a good opportunity of reporting success to one of Elpida’s stockholders. The one who made this very Grantsdale thing.
DDR-II 533 memory from Elpida
So it’s all calm and quiet in the memory market. No prices jumping, or attempts at merging or bankrupting anyone, and no legal quarrels. This is all gone and now we can only dig in the new (or better, newest) technologies area for anything interesting. As for new technologies, Samsung released a second generation (288Mb) of Network-DRAM chips (a.k.a. FCRAM). These are actually DDR SDRAM chips with reduced latencies for working in network equipment. Samsung also produced another multi-chip package (MCP) product in November. It is 6-layer rather than 4-layer. The miniature casing now conceals 256Mb of NAND flash, 128Mb of NOR flash, 64Mb of UtRAM, two units of 128Mb DRAM and 16Mb SRAM. That’s a perfect solution for a smartphone or PDA.
In the newest technologies category, we have HP Labs offering one-write permanent memory. That’s a second, after the molecular memory, offer from them lately. They suggest using a polymer film as the medium, and the main disadvantage, like with similar proposals of other firms, is the access time. They work upon the problem. Interestingly, Intel has become active in this area of research, too.
As for the longest-term perspective, the scientists from Dresden reported their dealings with nano-tubes of a 10nm diameter that are filled with various magnetic materials like iron, cobalt and nickel. Of course, it’s all very raw so far, but the potential growth of the memory density in hundreds times above the today’s densities cannot be underestimated.
I’d like to start this section out by making clear the current standings of the market players, the current products to follow. The standings have ultimately reached equilibrium. NVIDIA’s ended its third quarter with a sales volume of $486 and profit of $6.4 million, while ATI is going to end its first 2004 quarter with sales volumes of $440-470 million. I estimate the quarter profit to be around $30-40 million, but let’s wait for the official report.
Anyway, it is clear that ATI Technologies enjoys the results of the last year, while NVIDIA should feel the opposite. I won’t quote the financial results of the two companies in the same period of 2002, as they agree with the above statement. This only applies to the first nine months of this and previous years, though. This year NVIDIA earned less than then. Talking about the last quarter only, the current situation is much better: a year ago they had a catastrophic situation with the (non)arriving NV30, so it just cannot be worse than then.
Overall, ATI should overtake NVIDIA in sales volumes by the end of this year. Note also that NVIDIA profited much on its nForce chipsets, while ATI is only developing this direction, lagging for about a year. As a result, according to Mercury Research’s data for the third (calendar) quarter, NVIDIA has 25% of the computer graphics market, ATI – 22% and Intel – 35%. So after the fourth quarter is over, the second and the third companies may well change their places.
As for the chips for graphics cards, NVIDIA looks well enough with 62% against ATI’s 32%. On the other hand, the former lost 2% in a quarter, while the latter gained 4%. If the rate remains the same, it may take just half a year for ATI Technologies to close the gap. The mobile graphics market is unlikely to save NVIDIA if that comes true as it has 12% (+1) there against ATI’s 56% (-3).
There’s an interesting nuance, though. The current leadership of NVIDIA in the desktop GPU market is due to its supremacy in the DirectX 9-compatibles sector (72% against ATI’s 27%), while it’s all quite otherwise with DirectX 8 GPUs (ATI has 80% there). This situation seems to evolve from the inertia of the market as many customers don’t view ATI Technologies as a supplier of high-end products. If the Canadians keep up their tempo, though, the negative aura about the company’s image will get thinner with every passing quarter.
ATI is also steadily enlarging its partnership net. November, the Hong Kong-based Xpertvision, Diamond Multimedia (a division of Best Data that bought the trademark) and Exair Computer got caught in it. Besides that, ATI got stronger in South Korea where big sharks like LG and Trigem chose RADEON-based graphics cards (RADEON 9200 and 9800) for their systems. Samsung also signed an agreement with ATI to use ATI XILLEON chips in its digital TV-sets.
Meanwhile, NVIDIA is tearing itself apart between two suppliers, IBM and TSMC. The GeForce FX 5700 Ultra is manufactured at the IBM plant, while the GeForce FX 5950 Ultra – at TSMC facilities, both by the 0.13-micron tech process. It is rumored, though, that notwithstanding the positive words expressed by NVIDIA about its partnership with IBM, they have to transfer their orders to TSMC and UMC as IBM has problems with the chip yield with 0.13 and 0.09-micron tech processes using the low-k SiLK dielectric. The Taiwanese, according to the same rumors, provide a 20% higher chip yield with their 200mm wafers than IBM. On the other hand, IBM has a shorter production cycle – 65 days against 80 days with TSMC. So it is not quite clear who’s going to produce the NV40 and NV44 scheduled for the spring. So far, TSMC seems to have got the order, but things may change.
So these two companies, ATI and NVIDIA, have got hold of the whole desktop graphics market. S3 is only planning to launch its DeltaChrome one of these days, and XGI should come up with its Volari. The latter release looks more promising and there appeared previews of the Volari Duo V8 Ultra. That’s the topmost solution from XGI and preliminary tests put its performance somewhere in between the 5700 Ultra and 5950 Ultra (or between the RADEON 9600 XT and RADEON 9800 XT, if you like). Graphics card manufacturers (like Club 3D) will ask about $400 for this. It’s a bit tough, considering the inevitably raw drivers and the unknown brand.
This should add some competition into the market, though, and that’s the good about the whole affair. I hope XGI will do according to their schedule, although the company will find itself in dire straits: it’s going to get right into the middle of the next round between NVIDIA and ATI.
The current product lines from the two giants have the monstrous GeForce FX 5950 Ultra and RADEON 9800 XT hovering above everything else. They’ve been for quite a while in the market, and managed to get slightly cheaper, even. And the quantity of such cards from various manufacturers is on the rise. In November, ASUS, Gigabyte, Sapphire, SUMA, PowerColor, FIC and Chaintech announced such products. There are also some intriguing options like the Innovision GeForce FX 5900XT, which is an ordinary 5900, but with reduced frequencies (390/700MHz) and (supposedly) costing only about $200.
So we have the upper echelon merging with the renewed middle one. November, first cards on the GeForce FX 5700 Ultra made it into the market, priced the same $200 as the cut-down GeForce FX 5900. The choice of the RADEON 9600 XT from the same niche is enlarging, too. Curiously enough, the mainstream sector is mostly occupied by other manufacturers. Of course, we have the same ASUS and Gigabyte, but there are overall fewer well-known companies here. You can only spot Gainward, FIC and Chaintech. That’s quite logical since the price factor is of more importance here, and the set of brands should change accordingly.
GeForce FX 5700 Ultra from Gainward
Take a step down and we see the RADEON 9600 SE and GeForce FX 5600 XT. The different suffixes actually mean the same thing: reduced core and memory frequencies. Anyway, these solutions should be demanded, considering their relatively high performance, DirectX 9 compatibility, and pricing below $130 for a 128MB version. Add the lack of noise – the heat dissipation is too low at such speeds and a passive heat-spreader can handle the GPU. November, Creative Labs pitted its Radeon-based solution against the GeForces from Leadtek and SUMA in this field.
Creative Labs Blaster 5 RX9600 SE
The RADEON 9200SE and its competitor, the veteran MX440, stand on the lowest step. The latter is doing nicely, although obviously needs some face-lifting. NVIDIA seems to be willing to update the oldie and name the result as “MX4000”.
Among the innovations, we’ll probably see support of 256Mb memory chips, although across the same 128-bit memory bus, plus the integration of TDMS and S-Video controllers into the chip. The core frequency is rumored to equal that of the MX440-8x, while the memory – that of the MX440. So we have 275/400MHz, capped with AGP 8x support. In other words, NVIDIA slightly adjusts the chip for the current market situation. The cost of the end solution may be lower due to the optimal memory use. Let’s wait to see whether the rumors are correct.
The graphics card has no intrinsic value apart from the monitor, so this section of the review should have appeared. Well, it might never be here if the market was not boiling in the last month to give some matter for discussion. And the question is not about the choice of CRT or LCD models anymore. We should be forgetting about CRT displays already. I’m going to talk about LCD monitors and the next generation of technologies (plasma, OLED and the like).
The most important factor for evaluating a monitor remains the same – the diagonal length. And 15” LCDs are rarely spoken about, too. I could only remember 15” models from BenQ, FP557s v2 and FP567s v2, showcased at Comdex with their characteristic suffix and improved response time (to the today’s required 16ms), and DoubleSight DS-1500 from the namesake company, which consists of two 15” panels in one case separated by a thin strip of plastic. The “miracle” costs $800 and makes the option of two 15” LCD panels quite reasonable.
In the end of this year, and, accordingly, in the next year, 17-19” models of LCD monitors were most optimal for the manufactures and were promoted accordingly. This is related to the wafer size the production lines of the 4-5th generation operate with, the number of panels of different diagonals that can be sliced up from one such wafer (that is, the self cost of the panel) and their market price.
So we saw several new 17” models in November from respected manufacturers. BenQ was among them showcasing its FP757 v 2 and FP767 v2 with 16ms response time at Comdex.
Looking at the specs of the FP767 v2, you notice an average brightness of 260nit, but a quite normal contrast ratio (500:1). The viewing angles are 140 degrees, only an analog input is available. This is not excellent by today’s standards. The moderate technical characteristics are somehow balanced by a cute two-color case with a thin bezel. This is obviously a model for the low-end sector.
EIZO Nanao targets the other end of the price range rolling out its FlexScan L550-R. The case design is two-color, too. Note also the brightness auto-adjustment depending on the lighting in your room and better technical parameters (300nit, 450:1, 16ms, 160 degrees viewing angles).
Samsung’s assortment covers the entire range, but the latest two models, SyncMaster Magic CX710P and SyncMaster 177X, are unconditionally aiming the very top. The first device allows adjusting all of its settings directly from the computer, so it lacks any buttons on the case (well, we’ve waited long for this, since the advent of the DVI interface). The parameters are impressive for a 17” monitor (700:1 contrast ratio and 180 degrees viewing angles). The other model is a humbler one, priced at $550 rather than $670, but draws the spotlight due to its record-breaking specified response time, 12ms! Who’s going to tell now that LCD monitors are not for watching movies or playing dynamic computer games?
SyncMaster Magic CX710P
19” LCD models are no less popular, at least among the manufacturers. They take more pains developing the design of the product here, as the diagonal alone serves mostly image purposes and a nice exterior of the device is most welcome. The technical parameters should match the design, though. For example, BenQ’s FP951 has a contrast ratio of 700:1 and this is quite ordinary for 19” monitors, although close to the limit. This is a top-end model, while the IBM Think Vision L190p and Hercules ProphetView II 191 don’t offer anything transcendental in technology, but have their own peculiarities: Hercules’ monitor features a cute design, while IBM’s one allows rotating the screen by 135 degrees around the horizontal axis.
We take a step further to find that 20” LCD monitors are less demanded by the users and less frequently renewed by the manufacturers. Times are changing, though, and Apple released another iMac version with a 20” LCD screen and a resolution of 1680x1050 (the iMac is equipped with a GeForce FX 5200 that doesn’t quite match the screen).
Iiyama presented two new models in this sector, ProLite H510 and H540S, which are close to each other in their main parameters. Both have the same resolution – 1600x1200. They will compete with the Belinea 102015, which is the biggest monitor from the company having a diagonal of 20.1”. Its design is very conservative, although the basic characteristics are normal, matching those of the H540S.
Let’s take another step up the stairs. We have the record-breaking 57” LCD panel from Samsung here shown in November. It comes from the sixth-gen production line, something like a Rolls-Royce in the car market. With its 600nit brightness, 1000:1 contrast ratio,170 degrees viewing angles, and 8ms response time, the monster feeds on 350W of power only. Well, that’s rather about HDTV-sets rather than computer monitors.
Fujitsu’s new technology for producing plasma panels with 1-meter long pipes (that is, the screen’s width or height) is also about television. The technology may make into reality such wonders as 200” plasma panels. The company is now busy developing one with 2x3m dimensions and 1000nit brightness and a few hundred watt power consumption. By their estimates, the development may take a couple of years.
Prototype
We have only trends in this area. Due to increased data densities, hard disk drives are shrinking in size (at least, there are ever more small-size HDDs available). The platter capacity is growing accordingly. The rotational speed remaining the same, the data access time is growing, too. The read/write head passes one degree of the circle in the same time, but has to deal with more data.
So it’s no wonder November events were mostly concerned with 1.8” HDDs (not even the 2.5” ones!). Well, it is of course a sort of visual artifact or coincidence – nobody else offered anything else. Anyway, this is a trend as not only technology is moving this way, but also the demand: notebooks, barebone systems, and consumer electronics like digital VCRs.
As for the products proper, Toshiba showed up with its 1.8” drives of “grown-up” capacities (20 and 40GB). The rest of characteristics are quite normal, too. Consider: 4200rpm spindle rotation speed, 15ms seek time, ATA-5 interface, 2MB buffer. Add a power consumption of 1.4W.
Moreover, Toshiba is planning to increase the production volume of its 1.8” series in the next year from 300 to 600 products monthly, by making arrangements with other manufacturers. The company estimates the market of 1.8” HDDs at 25 million items in 2006. Interestingly, the China-located GS Magicdrive has scheduled about the same production volume of 1.8” drives for 2006. In other words, the Chinese are going to have close to 100% of the market (owned by Toshiba so far). They are unlikely to do it, but may make up a good competitor anyway.
There’s one more promising technical innovation. Agere Systems, a major provider of HDD logics, cooperated with Maxtor on developing a chip to combine all the logic functions necessary for a regular hard disk drive. So the controller circuit board may now carry only this single chip and the chips of the cache memory, thus reducing the energy requirements, cost and dimensions of the product. Of course, Maxtor’s HDDs will be the first to feature the innovation, and this company is not yet targeting the small-size drives market.
That’s all about the dimensions, let’s now mention the other important characteristic – speed. There’s also a new chip, from Adaptec. November, the company tested its Serial Attached SCSI controller to see that the speed reached far beyond the specified 3GB/s notching 5GB/s. There’s only one question: whether this interface is destined to live, considering the inroads made by the Serial ATA interface.
Optical drives are still coming with either ATA/133 (internal) or USB 2.0 (external) interfaces. The last month was rich for optical drive news. First of all, a new generation of DVD-R drives has arrived to burn both DVD-R and DVD+R disks at 8x speed, twice faster than the mainstream products of today. I won’t even cite the names of the companies to offer the innovation as it is long enough to include even Cyberdrive and MSI that usually don’t go for innovations among the first.
Pioneer DVD-A07
A number of manufacturers are also improving their 4x model series, complementing them with new multi-format devices. The numerous announcements were mostly concerned with external optical drives with the USB 2.0 interface. It seems like the normal high-speed bus to connect external peripherals has catalyzed this market segment.
As for the prices ruling here, the pressure is high due to rapid changes in the model lines and tough competition. So we can see an 8x drive costing below $300. That’s the money you paid for a 4x DVD-R model just a little ago and 2x model half a year ago.
There’s a characteristic example from Taiwan where contract prices for one-format 4x drives (i.e. DVD-RW or DVD+RW only) dropped from $220 to $88. Multi-format 4x models have now stabilized at $120-130, while the contract price for 8x drives came down from $160-180 to $140-150 in a single month. So we are waiting for a corresponding price adjustment for multi-format 4x re(writable) DVD drives.
As we’ve landed on Taiwan, let’s notice the local manufacturers being busy with thin DVD+RW drives. That’s another proof to the statement that the market of notebooks and barebone systems is getting more prominent in the whole computer market. The list of the new pretenders for a slice of this pie includes such names as BenQ, AOpen and Lite-On. The latter, for example, should reach the level of 200 thousand items monthly in thin drives (all kinds of them), which is twice more than in the first half of the year.
Getting back to the full-size devices, I guess that 8x DVD drives are up to their own price reductions as 12x models are looming in the horizon. And the latter half of 2004 is going to give us 16x multi-format rewritable DVD drives! At least, Philips has a working prototype already. That said, the engineers suspect the 16x speed to be very close to the limit of the write speed for DVD disks. Anyway, it’s not only Philips that pushes the progress on, as Fuji Photo Film reported November that they created a material for optical disks to make this 16x write speed a reality.
In parallel, another technology will be trying to get accepted by the market. It is concerned with increasing the capacity of a DVD+R disk by making two data layers per side instead of one. If this comes true, the capacity of a standard DVD+R disk will nearly double from today’s 4.7 to 8.5GB. There’s only one uncertainty – whether the current drives have powerful enough lasers to be able to handle those disks after a firmware renewal.
Well, 8.5GB is nothing compared to the single-side dual-layer blue-ray disks from Matsushita with a capacity of 50GB and 2x write speed (this “2x” means 9MB/s). On the other hand, this is merely a maneuver war, preparations before the final engagement, which is to take place in 2004-5 between the two competing standards of optical disks of the next generation: blue-ray and advanced optical disk system.
We can also live to see one more standard, HD-DVD (from “High Definition”), which implies DVD disks with 15GB per layer. High Definition means one such disk may store one high definition movie. The proponents of the blue ray technology opposed the new standard that might become a danger to blue ray. Notwithstanding the opposition, DVD Forum accepted the specification, 8 votes against 6. However, the formal acknowledgement is only the beginning, and the HD-DVD format should prove its usefulness yet.
You may think that CD disks and CD-drives are quite obsolete nowadays. That’s not quite so. November, there were a few announcements of combo-drives (CD-RW/DVD-ROM), external with the USB 2.0 interface as well as of a couple of simple external CD-RWs. Well, it’s hard to invent anything new to the internal CD-RW, while external devices may be perfected yet with respect to the dimensions.
The event of the month in the notebook market was the jubilee of the major trendsetter. IBM Thinkpad was 20 million in November after 11 years of being in the market. That’s a solid number, and IBM has become an inventor of numerous trends in the field, including the “folded” design.
On the other hand, 20 million divided by 11 years is less than 2 million notebooks annually. That’s not at all enough to be the leader. ASUS is going to ship a total of 1.9 million notebooks this year and boost the number even further in the next year, to 2.5 items, or by 30%!
And again, as with the mainboards, Elitegroup wants to boost their notebook sales by the same percent, although it is smaller in sheer numbers – 1.1 million items. And as with the mainboards, the plans look quite dubious. Dell also voiced its plans to stimulate the demand for its products, most of which look too similar to each other. That’s the problem the company is going to solve.
Well, whatever they do, they can’t get along without the main supplier. Intel remains it and even reinforced its position with the launch of the Centrino. According to IDC, notebooks on this platform account for 42% of the total sales volume by the results of this year. Well, this is only the beginning, as the same agency forecasts the share to grow up to 95% of the market by 2006.
The main feature of the platform is its wireless capability and that’s the matter Intel should be working on hard in the near future. The number of hot-swap points is too low so far, but the WiMAX (802.16) standard should play its role in resolving the problem of the Last Mile. Wi-Fi needs further development, too, and Intel has already chosen the companies to order 802.11g modules from for the Centrino 2.
Let me remind you that this platform is going to appear in the second half of 2004 and to include the Pentium M processor on the 90-nm Dothan core (the frequency starting from 1.6-1.8GHz, and the processor itself becoming available in the beginning of the year already) and mainboards on the Alviso chipset (PCI Express, DDR-II, Serial ATA, Gigabit Ethernet, Dolby Digital 7.1) plus those 802.11g modules.
Intel has been rather poor at graphics, but, fortunately, even the current realization of the Centrino allows using external graphics chips. Two such chips showed up in November. While NVIDIA offered a solution for “simple” users – the GeForce FX Go5700, which is very close to its desktop counterpart in characteristics and consuming 8W of power, ATI Technologies targeted professionals this time, unveiling its MOBILITY FireGL T2 on the FireGL 9600 core.
Thus, the performance of the two solutions should be quite similar. As for availability, notebooks with the GeForce FX Go5700 are not supposed to arrive until the first quarter of 2004, while the MOBILITY FireGL T2 comes in the nw8000 workstation from HP. Of course, it will be based on Centrino platform and will use new ATI and IBM chips.
Well, mobile workstations are not exactly what the Centrino implies, its main advantages being a compact size and economy. A good example is the most miniature Sony VAIO PCG-X505/P with a 12” screen, based on the 1GHz Pentium M, weighing less than 1kg and having a minimum thickness of less than 1cm.
Compared to such extremities, a solid mainstream product like the Acer TravelMate 370 looks like a heavy-weight, although Acer classifies it into a category of compact and ultra-compact notebooks: 1.77kg, 2.4cm thickness, 12” panel, external drives. So the Acer loses to the Sony in all but the price. The former costs $2315, the latter – $3230.
Standing next to these two models, the Dell Inspirion 5150 (Mobile Pentium 4 3.2GHz, 15” LCD, up to 2GB of PC2700 memory, Mobility RADEON 9000 or GeForce FX Go5200, a HDD with an up to 60GB capacity and so on) looks a colossus. The whole treasure costs only about $2000, if you can carry it away (3.6kg weight, 4.65cm thickness).
Well, recently, there appeared real brontosaurs in the notebook world with configurations like Pentium 4 3.4GHz, 17” screen, i865PE, two HDDs in a RAID, Mobility RADEON 9700, integrated TV-tuner (3.9kg without the accumulator), or Pentium 4 Extreme Edition 3.2GHz, i865, up to 4GB memory, a pair of 7200rpm HDDs, RADEON 9600 PRO 128MB. Of course, such monsters cannot be used autonomously as they are the result of the evolution of the battery-less DeskNote systems that were originally intended for transportation from home to office and back again.
The evolution of tablet PCs hasn’t yet yielded anything useful. Well, they are here like the Acer TravelMate C300, but it is in fact a notebook on the Centrino with a screen rotating by 180 degrees. That is, you can fold it up the other way to have a bulky and weighty “tablet”. And pricy, too - $2995 for an ordinary configuration with a Pentium M 1.6GHz. The same holds true for the ViewSonic V1250 and Toshiba Portege M200, announced in November.
There are also “pure” tablets, like the Philips DesXapeTM 150DM. It can work as a monitor when connected to the base station, or as a tablet when detached and communicating in the “thin client” mode by means of 802.11b. This seems to be a promising direction for further development as it brings us back to the original philosophy of the tablet PC, but on a new level of technology: 15” matrix with sensor input, Intel XScale 400, 96MB integrated memory, 2.4kg weight. The price is much lower than of the would-be tablets, $1899, although it is still unreasonably high.
