by Andy Yaschenko
03/15/2004 | 02:11 AM
With a slight delay Mercury Research summed up the results of the Q4 of the last year: Intel’s 82.8% (+0.2) against AMD’s 15.5% (-0.2). This is a very negligible change, in spite of the beginning of Athlon 64 shipments. Well, this was only the beginning, and the new processor will have much more weight in the current quarter results. The whole picture for the entire past year also remained the same, although with some advantage to AMD: Intel’s 82.9% (-0.7%) against AMD’s 15.4% (+0.5%). This is obviously due to AMD’s dumping at the beginning of 2003 when the Athlon XP reached its evolution peak, while Intel didn’t have the Prescott as a worthy response.
<%BANNER[article]%>These are deeds long past, although necessary to emphasize. Now we are more interested in the last month, rather than last year. The notorious 64-bit processor architecture was the hot topic of February. To be more precise, there was a sudden twist in the development of the plot: 64-bit addressing had been considered as absolutely impossible in Intel processors, but this attitude changed with the time. You have surely heard persistent rumors about the Yamhill project. The crowd of disbelievers got smaller too as the very idea of desktop 64-bit computing started making some sense. The Athlon 64 received a warm welcome from the market and Microsoft announced its starting mass testing of Windows XP 64-Bit Edition (of course, solely for the Athlon 64). With such things going on in the background, Intel had to listen to the mass market.
In February, they made another step in this direction having announced Intel Clackamas technology at Intel Developer Forum (IDF). The technology is not an exact copy of the AMD x84-64, but close enough to run the same Windows XP 64-Bit Edition with different dynamic-link libraries. At least, AMD and Intel have different names for the new registers.
Anyway, I hadn’t been expecting things to turn this most favorable way. That has now become a kind of tradition: Intel utterly opposes every demand of the market, but then gives up and does everything right. That’s what we saw with Rambus when Intel was grating the industry nerves for a while, but then launched the i815E and even kind of headed the standardization process for PC133 SDRAM.
The company is wasting no time now. The first quarter of 2004 is passing by, and the second is going to bring us a Xeon with the 64 bits. 64-bit chips on the Prescott core are expected to arrive by the end of this year. This is a final confirmation to the earlier rumors about the Prescott core having the CT technology originally. Once again, this is Intel’s way of doing things. By the way, now that the technology has been officially announced, we should use its official name – IA32E where “E” stands for “Extended”, of course. Frankly speaking, “x86-64” seems more winning from the marketing point of view.
As for the launch timetable of IA32E processors from Intel, we should regard such chips as the Athlon 64. Until Microsoft has finished its Longhorn (that is, for the next couple of years), no processor will really need 64-bit memory addressing. At least this will be true for 99% of PC users who are not engaged into scientific calculations. In theory, the 64 bits should provide some benefits in multimedia applications, but that’s not too critical. Intel’s introduction of IA32E into the Prescott has been provoked by purely marketing considerations: they must not let AMD sleep well.
Server Xeons on the Nocona core are quite another matter. Sophisticated server applications which move around colossal data amounts do need 64-bit memory addressing – this is the field where Intel has started losing to the successful Opteron as this chip from AMD has met no worthy rival. Now that Intel has introduced an alternative, the company’s image should provide some advantage over AMD in the server market. We will see if it’s not too late for Intel to try to regain what has been lost.
This news was really hot, but we should still wait for its consequences to spring up. There’s another piece of hot news, and about a real product! The Prescott, the first 90nm x86 chip for desktop computers, brings about 13 new instructions called SSE3 for processing multimedia data.
Prescott features a double (1MB) L2 cache and 2.5 times more transistors (125 million against 55 million). I wonder what unknown-yet technologies this surplus of transistors conceals! The chip also has a very, very long pipeline, which affected the performance negatively if we take the frequency as the criterion. As a result, Prescott appears losing to the same-frequency Northwood across a number of tasks.
Intel rolled out five models in a row, with the frequency ranging from 2.8 to 3.4GHz, although the junior, 2.8GHz, model appeared in stores only in the middle of the past month, while the 3.4GHz version should have gone public this month only. By all evidence, Intel hasn’t yet solved its problems with the 90nm manufacturing technology. On the other hand, AMD doesn’t have such a process at all, while Intel is going to deal with those problems immediately. End of February, chips with the new core stepping (D0) should have started shipping. They are more economical and ready to come to the market in early May. Curiously, Intel is only speaking about new versions of 2.8-3.2GHz chips, so it seems like there is no such thing as a 3.4GHz Prescott with C0 core stepping at all, notwithstanding the announcement.
That’s the order of appearance of the new processors in the retail market: the 2.8GHz Prescott was soon followed by the 3.0GHz model. By the end of February we suddenly saw the unannounced 2.4GHz chip. Such chips seem to be defective Prescotts, which are unstable at their planned frequency. Well, Intel always has an escape way in every situation: they squeeze the Northwood core to the last drop releasing two new chips on the old core along with the announcement of the Prescott: Pentium 4 3.4GHz and Pentium 4 Extreme Edition.
These are going to take the blow until April-May. Moreover, unlike its Prescott counterpart, Pentium 4 3.4GHz on Northwood core appeared in shops in the beginning of February and for the same money: $417. Intel is playing unfair: the cache is twice as small, why is the price so high then? Well, they just set the same prices for all Prescott and Northwood chips of the same frequencies and even had to drop the prices of 2.8-3.2GHz Northwood processors: Prescott production costs very little. For example, the official price for the 3.2GHz Northwood was $417 at the time of the announcement, but then plummeted to $278. I hope that while you can’t buy a 3.2GHz Pentium on the Prescott core yet, the lower price of the 3.2GHz Northwood should say its word.
The server market lives on without any turmoil (again, this calm provokes some suspicions about the 90nm production technology) with a bright future ahead. By the way, “bright” should be taken literally as Intel announced the first optical chip where data flows between transistors along miniature optical fiber instead of copper wiring.
This bright future is still in the distance, though. As far as short-term plans are concerned, they talked about reasonable and predictable things at the last IDF: the new Xeon MP processor comes on the Gallatin core and the Xeon DP on the Nocona (the economical version of the chip will also be based on this core). These chips will be followed by the next generation at the end of this year (on Potomac and Jayhawk cores, respectively, with frequencies around 4GHz). This will surely come true, if Intel polishes off its 90nm process and the Tejas core. I repeat it all once again only because I would like to stress the serious nature of the topic.
Itanium, with all the fuss about IA32E, is not yet abandoned altogether, but rather otherwise: we are waiting for the Madison with 9MB of the L3 cache this year as well as for the economical Fanwood. In the next year, for the first time in Intel’s history, the processor will propagate through fission to give birth to the dual-core Itanium on the Montecito core with an up to 24MB L3 cache and its “light” version, the Millington. In other words, the latest events in the IA32 field haven’t had any dramatic impact on the company’s plans for IA64.
AMD’s preparing its response: implementation of DDR2 support into the Athlon 64/Opteron by the end of the year in 90nm San Diego, Winchester and Toledo cores (again, those high hopes about the 90nm process!). As for the current day, and the rival for Intel’s Prescott, the company’s updating the stepping, just like Intel. Whereas Intel updates it in new chips, AMD goes the more traditional way.
In February, the company announced the CG stepping of the Athlon 64 processor in all its variations. The stepping, just like Intel’s, is intended to reduce the processor power consumption. Well, I won’t call the current Athlon 64 a power-hungry chip, actually. Some matter for an argument may arise if the latest rumors about the Newcastle core come true: AMD is rumored to reduce the L2 cache size from 512 to 256KB. That’s a very questionable solution, but maybe the speed of the integrated memory controller is enough to make the influence of the cache memory on the processor performance nearly invisible.
They may make up for that by reducing prices at the same rate further. Well, computer manufacturers have already got a strong liking towards Athlon 64. In February, FIC introduced one model like that, only Dell keeps loyal to Intel. As announced, Dell is not satisfied with the current generation of AMD processors.
It seems like this situation won’t change with the transition to the Socket 939. This socket is rumored to be somewhat delayed, with some “help” from Prescott and Prescott-supporting chipsets. Thanks to Intel’s problems, AMD has a small timeout and can relax for a couple of months, polishing off its new form-factor. First Socket 939 processors are moved from March to the end of April in the schedule. This may be just a coincidence, though, as AMD may have its own problems: even after the announcement, in the entire Q2, the company will produce only about 10 thousand chips. Anyway, that’s sad, as Socket 939 is likely to reduce the average cost of powerful Athlon 64-based systems.
So far, Athlon 64 FX had to be used in Socket 940: by the end of March, the new processor of this family, the 2.4GHz Athlon 64 FX-53, will appear and probably it will be the last representative of the Socket 940 form-factor. Again, AMD is rumored to cut FX-51 shipments severely in the second half of February to prepare for the launch of the FX-53.
The Opteron is getting ready for the hot summer, reducing its power consumption and heat dissipation: judging from the latest moves of both AMD and Intel, this is the most urgent topic in the server market. In February, AMD announced six new Opteron CPUs of this class, with HE (Highly Efficient) and EE (Energy Efficient) suffixes. The first three have a power consumption of 55W, while the second three – of only 30W. They are designed for the standard form-factor (Socket 940) and at standard frequencies (2GHz for the first group and 1.4GHz for the second one). The only non-standard parameter is the price: for example, the Opteron 840HE costs as much as $1514, while its ordinary 840 mate costs only $698.
Well, you have always had to pay for compactness and economy of server systems. In February, Sun Microsystems introduced its long-anticipated Sun Fire V20z, the first company’s server on the Opteron CPUs. The dual-processor system has a starting price of about $3,000 and is going to be followed by four- and eight-processor servers.
I hope they will also be so inexpensive, considering that AMD lowered the prices for all Opteron and Athlon XP processor models in February. Opteron suffered most (up to 53%), while Athlon XP 3000+ stepped down from $203 to $163. Athlon 64 remained untouched, but its pricing is modest enough as it is.
As usual, February started with the January reports. Many mainboard makers don’t have much to be proud of, as they lost in shipment quantities compared to December, although not very much, by 4% in total. Again, we should make allowances for the strong growth in the fall, so the situation may be just getting back to the norm. Short-term forecasts are contradictory. While analysts predicted a growth of 3-15% for different companies, the first quarter overall is expected to result in a slump of 10-13%. That’s a strange discrepancy, so let’s wait for the February results at least to make any guesses about the outcome of the situation.
Chipset makers haven’t had much fun lately, too. VIA Technologies reported a February sales reduction of 17.8% compared to January, in money, not quantity. This is also the indication of the too-low price they set for their chipsets. Of course, I mean chipsets for Intel processors – VIA doesn’t have to resort to dumping as far as chipsets for AMD’s CPUs go.
Mainboards for Athlon 64 are becoming cheaper, partially because of cheaper chipsets, too. For example, the not-very-inexpensive Leadtek now offers an nForce3 150-based mainboard for less than $100. The chipset itself is no low-end, as the dual-processor DK8N on the nForce 3 150 from Iwill, to be presented in the second quarter, indicates. Of course, it supports the Socket 940 Opteron.
The just-revealed nForce3 250 has already become nearly obsolete as it belongs to the passing generation of chipsets. The first representative of the new generation – the K8T890 – is already here, for example, in the A8V DX mainboard announced by ASUS: 1GHz HyperTransport, PCI Express 16x in parallel with an AGP port, 7.1 audio. Overall, it is a typical Socket 939 mainboard to support appropriate processors. By the way, the availability of AGP 8x and PCI Express 16x ports on one board creates a curious precedent as far as dual-monitor configurations are concerned, in which every monitor works with its own dedicated and fast 3D accelerator. I’m curious to see if manufacturers and software developers will really use this opportunity.
I should remind you that it’s not only VIA Technologies that produces chipsets for AMD processors. ALi officially announced its M1689 chipset that belongs to the same generation as the K8T800, but includes all functions in one chip. This fact is only of some value for the mainboard makers, rather than end-users, so ALi will have a hard time trying to promote their product in the market. Their only real opportunity is setting an alluring price.
Overall, everything was frozen in the chipset market for Athlon 64/Opteron processors, waiting for the spring. However, Pentium 4 didn’t enjoy much attention in February, either.
Of course, there are super low-cost options here, too. It was hard to imagine just a little while ago that SiS655FX chipset would enrich the category of “everyone-can-afford-it” products. Now we have ASRock P4S55FX+ mainboard: the specification indicates that it is an acceptable option for the Prescott, while the pricing policy of ASRock means you won’t have to pay more than necessary.
It’s the same with VIA, which released its PT890 and which reference boards have two graphics interfaces, although it’s still uncertain if PCI Express and AGP cards can work along together. Regrettably, PT890 doesn’t support PCI Express 1x as VIA doesn’t have the necessary South Bridge.
SiS, on the contrary, can’t stop producing them, announcing its second chipset in February, SiS965, which is a variation of the earlier SiS965L with an integrated Gigabit Ethernet controller. This is just announcement, though, and we should wait until the mass production starts.
In fact, SiS was very active in February. Besides the new South Bridge and claims about full support of the newly-born Prescott starting from SiS648FX (actually, even the 854 would bear the Prescott provided there is well-designed power circuitry), they continue with their strategic support of RDRAM in the near future. They are preparing new chipsets with support of this memory type till SiS659TX scheduled for 2005, which is going to work with PC1600 RDRAM (its 6.4GB/s bandwidth only equals 800MHz of PC6400 DDR2…), PCI Express and other advanced technologies. I’m really confused about the purpose and targeting of this product.
These chipsets will be produced for SiS by UMC, as SiS abandoned its fabrication past completely, selling SiS Microelectronics (its only 200mm fab) to UMC. That’s a positive fact for the company, as SiS had been feeling the lack of production facilities and the construction of their fab solved the problem. Now that they have normalized their relationship with UMC, there appeared the opportunity of making chips cheaper at UMC’s fabs and SiS feels no need for a fab of their own anymore.
Other companies, like the heavy-weight Intel, don’t shun placing orders with third-party makers, like TSMC. Intel also had its new chipset in February, the first from the spring line, called E7210 (Canterwood-ES). It was the first probably because it was the simplest of all. And it is the simplest because it is the well-known i875P with a new South Bridge and support of the 64-bit PCI-X bus. Manufacturers of server mainboards soon revealed their products on this chipset and I suspect they didn’t have to work hard to do that. First mainboards were already available in February.
It’s more complicated with the Grantsdale and Alderwood. Of course, they showed mainboards on these chipsets at IDF, but we are unlikely to see them in mass quantities in shops until the end of the spring, even if the chipsets are announced as promised, in the end of March. Again, announcement and production are two different matters, even for Intel. Otherwise, there were no surprises in the specs. Overall, February was the month of waiting for further events. I hope March is going to be somewhat more interesting.
Winding up the mainboards section of our overview, I’d like to recall some basic things, that is, buses. In February, there were two remarkable events. They ratified the HyperTransport 2.0 standard with a clock rate increased from 800MHz to 1-1.4GHz, and a higher bandwidth (the bus width remained the same). Among major innovations: HyperTransport now is compatible with the PCI Express. The second event was the Infiniband developer forum, which discussed the problem of increasing the bandwidth of this bus, aiming at 100GB/s.
Overall, February was an example of the slow beginning of the year with a low demand. The prices may have fallen lower, if the memory makers hadn’t switched to demand-enjoying flash and started playing with DDR2. As a result, the reduced supply compensated for the low demand and the memory prices went down (for example, from $4.75 to $4.45 for a 256Mbit DDR400 chip), but still remained above the production cost. I think a majority of makers will enjoy the results of the first quarter.
Traditionally, March is the month of growing prices, as the demand for memory from computer manufacturers increases. This March won’t be very splendid, though, according the mainboard makers’ forecasts: a growth of contract prices by a few percent and a relative stability (like plus/minus 3-5%) in the spot market.
As memory modules continue becoming cheaper, their manufacturers are enjoying themselves issuing products, sometimes quite different from the JEDEC standards. This February was less exotic and brought us standard modules, although not quite widespread.
OCZ showed up with its kits of two PC3500 registered modules for Athlon 64-FX systems and then unbuffered 1GB PC3200 DIMM modules (thus, a kit amounts to 2GB of memory) with OCZ’s proprietary system for prevention of parasite noise aka ULN2. They set the maximum timings, though, so the goal behind this technology seems rather obscure.
Kingston also revealed similar-capacity modules (1GB PC3700 and 2GB kits) with even slower timings, but higher frequency – they come as an extension of the HyperX series. There’s only one problem for the buyer of such products – to find a mass application where so much system memory would be demanded.
This niche has practically exhausted itself and the module makers are casting intent glances at DDR2. Corsair has already got the brand, XMS2, and posted diverse info on DDR2 modules on its website. The speeds are long known, though, while the capacities won’t be much of a surprise: the company is planning to show 256MB-1GB modules for personal computers and 512MB-2MB for servers. Curiously enough, such modules were already available for ordering in some places, although there are about two months left before mainboards capable of working with them appear.
Moreover, you can already go shopping for DDR2 modules in Japan, although Buffalo, their manufacturer, asks inadequately much for them - $915 for a 512MB PC2-3200. It’s even funnier as there are no mainboards for such memory and the difference in performance with PC3200 is negligible. No wonder then that serious manufacturers prefer to make a pause.
Micron goes on certifying its chips and modules with Intel, becoming the first company to certify successfully 1Gbit 400MHz DDR2 chips. It should be no surprise, though, when we recall that Intel is a major Micron stock-holder. 1Gbit chips allow making DDR2 modules with capacities up to 4GB and such modules were announced by Micron in the end of February. Intel is busy testing them now.
In February, Elpida went the opposite way, having introduced its 256Mbit DDR2-533 chips for PCs and servers and 1GB SO-DIMM modules for notebooks on 512Mbit DDR2-533 chips. Again, you won’t have a chance to use them until the end of this year when Intel releases the second generation of the Centrino platform.
Overall, as I emphasized it numerous times, there is no need to be hasty in moving to DDR2 until the arrival of 800MHz DDR2 chips and modules. I think the end of this year or the beginning of 2005 will be the proper time to start considering DDR2 for your own system. For example, AMD, which has always been careful about the optimal price/performance ratio of its solutions, is going to implement DDR2 support in its 90nm chips just about that time (if there are no technical problems).
In 2005, according to Intel’s plans voiced at IDF, we will see new modules, FB-DIMM, that would enhance DDR2 with a new pack of technologies like connection of the chips in the module directly to each other (point-to-point) rather than along a common bus like it is today. The basic advantage of the new format is the easier increase (as far as the number of pins is concerned) of the memory channels in the system and further frequency growth. Of course, Intel has this technology in mind for future processors, of 2006-2007, while the current chips will be quite OK with the current memory bandwidths.
Anyway, Intel schedules the introduction of FB-DIMM modules for the middle of 2005 and, according to the company, 667-800MHz DDR2 modules will come out in this form-factor. That’s why the necessity of transitioning to DDR2 400-533 becomes less urgent.
Well, there is also another candidate for the title of the “memory of the future”. It’s XDR DRAM technology promoted by the invincible Rambus and its proponents – Elpida, Toshiba and Samsung. At IDF, Toshiba showed 512Mbit 3.2GHz chips and Samsung also showed up with similar products, which promised to start mass production in 2004-2005. So the intrigue is getting tenser, and the crucial factor is the price. XDR DRAM will hardly cost little enough for the market to fancy this technology. That’s a kind of tradition with Rambus initiatives, though.
I’ve been talking about products and technologies, while the health condition of the manufacturers is a most important factor in this market sector. So far all companies have been feeling well. According to the Q4 results of the financial year 2003, Hynix was profitable and is now quite confident about its future. Nanya is also back to profitability and promises to increase its output by 52%, moving nearly all its facilities to 0.11micron process. ProMOS starts building its second 300mm fab in March, in the continental China. Overall, there is a kind of lull at the frontlines.
I doubt the people at NVIDIA ran for champagne after summing up the results of their 2004 financial year. The sales volume dropped from $1.91 to $1.82 billion this year, and the profit from $90.8 to $74.4 billion, although the GPU market itself grew significantly throughout 2003. I guess there is no need to recall the fiasco with the NV30 and the problems with the Xbox, while ATI was making good chips in the background. This is just the fact: NVIDIA lost its leadership in the market in the past year.
Judging by all evidence, the company doesn’t have many chances to regain its top position. At least, in the beginning of this year ATI and NVIDIA are both running neck and neck. ATI showcases samples of its PCI Express solutions (with an integrated support of this bus), while NVIDIA even launched a new family of products, GeForce PCX. Well, that was purely a marketing advance as the new products are actually the old chips accompanied with an AGP-PCI Express bridge chip. The GeForce FX 440 is now called GeForce PCX 4300, the GeForce FX 5200 transformed into the PCX 5300, the FX 5700 was replaced by the PCX 5750, and the FX 5950 – by the PCX 5950.
This renaming extravaganza gives you no performance gains as the bus bandwidth remains the same, while the translation can only be negative for the performance. On the other hand, the new indexes of the products indicate their novelty and, seemingly, higher performance. I really wonder if the price remains the same. Anyway, such graphics processors will only come to our computer systems in the middle of this year when there appear systems to install these graphics cards into. So we deal with a “paper” announcement – NVIDIA never resorted to this sort of thing before.
Another February announcement from NVIDIA was more honest as it dealt with new, although very slightly new, chips. GeForce FX 5500 and 5700LE chips seem to be the results of FX 5200 moderate overclocking and FX 5700 frequency screening, respectively. Thus, the former chip is a slightly speedier FX 5200, while the latter is a much slower FX 5700. Overall, there’s nothing new from the technological point of view, but a good repartee to ATI’s considerable price cuts for the RADEON 9600 SE from this very niche.
The faster RADEON 9600 XT receives more attention, though, and not from ATI, but rather from card manufacturers like SUMA, PowerColor and so on. Mainstream solutions have always been the most optimal choice according to the price/performance criterion, and they receive most attention from chip and card makers in the intermediary seasons.
Yes, there’s still some time until the arrival of the NV40 and the R420. Although the R420 has already been showcased to the public, information about this chip is still half-official. As far as we know, we are up to monster chips consisting of 160-210 million transistors each (much more transistors than in current Pentium 4 and Athlon 64 CPUs!). It’s somewhat off the topic, but anyway: graphics chips still cost you much less than central processors.
According to the preliminary info, all this transistor power will be spent to increase the number of pixel pipelines/texturing units, and the NV is rumored to have 16/1 (16 pipelines with 1 TMU per each) bringing another ideology into the GPU design, or 8/2. R420 should be somewhat slimmer than its rival, but more efficient with the 8/2 formula. Well, comparing the current top-end chips like RADEON 9800 XT and GeForce FX 5950 Ultra, I can believe that ATI has a more efficient product, but the current advantage won’t count with the next generation of chips.
I think the most interesting detail is not the performance data, but the fact that the NV40 is likely to have the AGP 8x interface. Its PCI Express variant will still be implemented through an external bridge. Of course, that’s not for the lack of trying – NVIDIA seems to be having problems with the PCI Express implementation. On the other hand, this is not a crucial drawback as the next generation of chips will do well enough without the PCI Express. It is not yet certain with ATI Technologies – whether they will launch an AGP 8x R420 and a PCI Express R423, or only the AGP 8x R420 plus an AGP-PCI Express bridge for it, or include both controllers into the R420.
ATI is said to have a chip revision that can be launched into production even today. NVIDIA will have it no sooner than in April. In April NVIDIA is also going to start production of the NV45, which is supposed to be the NV40 plus an integrated PCI Express controller (with some variations in frequencies, probably).
As for manufacture, NVIDIA, or rather its partners, is beset with troubles. As soon as NVIDIA made friends with IBM after some technological problems with TSMC then the Big Blue encountered them, too. The corporation is very slowly introducing its 0.11micron technology, while TSMC, on the contrary, has it right and the order for producing 0.11micron chips has been given back to Taiwan.
These are the global things that happened or are likely to happen in the nearest future. As for particulars, I should say it with regret that Hercules left the graphics card market. They did a good job making high-quality and non-standard products, beyond reference designs – maybe that was the cause of their downfall. The second and last piece of corporative news is about the ambiguous ABIT. On the one hand, this company has become a key partner of ATI, offering graphics cards on RADEON chips, but on the other hand, ABIT is still friendly to NVIDIA, going to unveil an NV40-based card.
This time we will return to the past, from 3D monitors to CRT dinosaurs. This approach is no worse than any other and Hitachi announced its Transpost technology in the end of February that allows producing true three-dimensional images viewed from any point without any glasses or other gadgets.
This technology, however, suits mostly for static images as it requires 24 snapshots of the object, which are then projected onto a cylindrical system of rotating mirrors. In fact, we deal with optical mechanics here with a small pinch of modern technologies.
Intel’s offer, Cayley, is closer to reality. Of course, the semiconductor giant is not interested in display devices, but chips for such equipment are another matter. Cayley is Intel’s implementation of LCOS technology (Liquid Crystal on Silicon), which starts to be employed in production of HDTV projection sets with diagonals over 40 inches. The screen brightness should be higher than with the competitor offers, and at a lower cost. Of course, Intel seems not in its place in this market, but they caught at the opportunity of using the side results of their research concerning communication chips working with optical fiber.
Such things don’t concern computer monitors much, as classical LCD panels rule this market. They are not rock-solid in their design, though. February, Mitsubishi introduced an interesting variant adding another backlight unit to the classical combination of the matrix and the backlight, but on the other side of the matrix, and making both backlight units transparent. As a result, we have an LCD monitor, perfectly transparent from both sides.
This would be an ideal solution for cell phones with two screens as well as for PDAs and notebooks. For example, it would be excellent for a notebook to be able to display images on the closed lid, giving the tablet PC concept another push. It’s also pleasant that none of this should tell negatively on the price, as there are no extreme innovations to be implemented, just Mitsubishi’s resolution in pushing the technology into the market.
Toshiba Matsushita came up with their own vision of a notebook screen: a 14” matrix in versions with XGA resolution (1024x768) and SXGA+ (1400x1050). Their appeal is in the power consumption: record-breaking 2.5W and 2.7W, respectively, while current SXGA+ screens consume about 3.5-4.3W. A good accompaniment for the Centrino.
Nothing very interesting happened with respect to ordinary LCD monitors for desktop computers. 17 inches have become the minimum standard from which manufacturers start up. In February, Sharp unveiled two such models: LL-T17D4 (analog and digital inputs) and LL-171A (only an analog output). They refreshed the design, improved the contrast ratio of the 1280x1024 image to 430:1, and have 170 horizontal and 150 vertical viewing angles with a response time of 20msec. Overall, there is nothing extraordinary, so the prices of $500 and $480 seem to be somewhat too high.
The 17” LCD monitor from NEC, the F17W81 model, belongs to a different category, as the integrated TV tuner and the TV-set-like design indicate. Its functionality embraces integrated 5W speakers, digital, S-Video and composite inputs, an USB 2.0 hub, and so on. The price is impressive, too, being about $915.
Higher up, we see Sharp again with its LL-191A and LL-T2015 models with 19” and 20” diagonals, respectively. Their only distinguishing feature is the 16msec response time, and the prices are appropriate: $700 and $1300, respectively.
LCD monitors with integrated TV-tuners, also known as LCD TV-sets, occupy the next step of the stairs. Once again we see NEC with its F23W11, resembling its junior brother in its parameters and costing about $2300. The FORIS.TV SC23XA1 from EIZO Nanao costs $2350, but it has an integrated DVD-player and two 10cm speakers – just like a regular TV-set.
Guillemot/Hercules will offer us nothing new anymore as the company left the LCD monitor market just like it abandoned its graphics card manufacture. Their reasoning is the same in both cases: low profitability. That’s true as this market is filled with players no worse than the DRAM market. In January, NEC sold its plasma panel manufacture to Pioneer and in February they gave up producing OLED screens, selling to Samsung its share in the joint venture called Samsung NEC Mobile Display. The joint venture for producing LCD monitors between NEC and Mitsubishi seems to be working on, luckily.
Well, if monitors become cheaper, this venture can cease to exist, too. According to Display Research, LCD monitors will lose $3-5 monthly in average, although this price reduction will only become tangible in the summer. The company forecasts LCD monitors to conquer about 70% of the computer monitor market in 2007.
Anyway, the good old CRT will have its 30% even then thanks to such admirers as IBM with its 22” ThinkVision C220p. Yes, this is a real monster, but where could you find an LCD monitor with a 20” diagonal (equivalent to CRT’s 22”), with a resolution of 2048x1536, excellent color reproduction and costing just $589?
Another milestone in the hard disk drive storage capacity was set by Hitachi: a 300GB model, the Ultrastar 10K300, attaches to servers across Ultra320 SCSI or FCAL interfaces intended for pumping data streams that require such storage capacities. Well, this model has only been announced; it will actually start selling in Q2.
Hitachi is also active in the opposite sector: the company is preparing a new integrated (without CompactFlash support) version of the 1” Microdrive and revealed the 1.8” Travelstar C4K40, that hit the market a year after its announcement, but anyway to the point: 20GB in the most miniature form-factor of today (there are devices smaller than 1”, but they can’t dream of such capacities yet). Considering that flash hasn’t been willing to getting cheaper lately, manufacturers of various portable device pay more attention to such products.
1U servers also feel the need for small form-factor hard disk drives, although not yet ready to get down to 1.8”, only 2.5”. In February, Seagate officially announced its Savvio family, a year after it first voiced its intention to make server HDDs of this form-factor. Now the company is ready to launch three series with different interfaces: 2GB Fiber Channel, Ultra320 SCSI and Serial Attached SCSI (SAS), ranging in capacity from 36.7 to 73.4GB. In summer, the two first interfaces will become available, and SAS drives are expected to come in the fall.
We shouldn’t confuse server 2.5” HDDs with their notebook counterparts, though. The characteristics are incomparable: 10,000rpm spindle rotation speed, 1.4 million hours mean time between failures, 4.1msec access time, and the above-described interfaces. However, Seagate is not a monopolist, as Fujitsu is also willing to be in this niche, announcing 2.5” server HDDs with a capacity of up to 73.5GB and the same choice of interfaces. They don’t voice the timetable, though.
In the middle of February, 2.5” SpinPoint M40 drives from Samsung appeared in shops. These 40GB models are exactly for notebooks, with 5400rpm speed and the UltraATA/100 interface. The Koreans are going to introduce 30, 60 and 80GB representatives of the family soon.
As for the good old 3.5” hard disk drive for the PC, the market is nearly exhausted. At least, there is no sense in growing towards higher capacities.
There was some fuss in the market of CD-RW drives. The speeds remained the same – they can’t grow any higher, too, but some new models appeared: an external PlexWriter 52/24/52U for $92 and an internal DVD-CD/RW combo (52/32/52/16) from Memorex for $80. So we have the old thing, but for less money.
DVD combos are still the developing sector where there is some room towards the ceiling, although these devices have slowed down, too. Teac DV-W58D and ASUS DRW-0804P are an example as they burn DVD-R and DVD+R discs at the same 8x speed and rewrite their RW versions at the 4x speed.
So the most interesting product of the month is Sony DRU-700A, specifically the information about its imminent arrival into the market. Anyway, this is a drive with parameters similar to those of the Teac and ASUS, but working with dual-layer DVD+R 9 discs with a capacity of up to 8.5GB. Regrettably, the burn speed for such media is lowered to 2.4x, although it remains 8x for ordinary one-layer discs.
By the way, in February, Philips released the second version of its specification for dual-layer DVD-R discs – version 1.9. It is once again preliminary, as the previous version (0.9) never made it to the official 1.0 yet. Anyway, we are sure to see numerous drives capable of burning two-layer DVD+R blanks at CeBIT. Philips is going to show its DVDRW 885K with a proposed burning speed of 4x. The second quarter will bring us the first version of the DVD+RW 9 specification. So we’d better wait for drives that support this specification.
Well, you could wait eternally long here. For example, you can wait for Blu-Ray or HD-DVD drives. Sony is producing the former drives and is set to introduce two models of external Blu-Ray devices with Ultra160 SCSI and USB 2.0 interfaces, handling 23.3GB discs. You like it? Prepare to pay $3980 and $4160, respectively. I hope HD-DVD drives will be cheaper, but let’s first wait for DVD Forum to ratify their specification first.
Surprisingly, IDF produced most news for this field, which is a very rare occasion. Although there is not much fuss about processor/chipset-related matters, but there is an abundance of mobile-concerned announcements. First of all, Intel is developing further its digital home concept with the notebook playing a major part. To provide full mobility in a house, a really broadband connection is necessary to be established in this domicile.
The presented Ultra Wideband channel with a bandwidth of 500Mbit/s is capable of becoming the carrier for USB 2.0 with its specified 480Mbit/s. Video, audio and data can be retrieved without any annoying wires. This technology clearly targets at destroying Bluetooth, which should be considered a failure because of its very sluggish invasion into the market and low bandwidth. We don’t yet know the schedule for UWB-supporting equipment (the second half of 2005 is set as a target), but the price is promised to be like that of Bluetooth.
The next step in wireless communication – within a building – is still the 54Mbit/s 802.11 standard. WiMax (802.16) can commune between buildings. The mobile Florence platform, showcased at IDF, includes only 802.11b/g and Bluetooth controllers so far. Florence is not a codename for the next Centrino generation or anything like that. It is something more wide-reaching – a concept of the notebook of the future that will come in three formats: 12” On the Go, 15” Virtual Office and 17” Mobile Entertainment PC. You can see the outlines of classification in the current notebook market already.
Besides the communication capabilities of the Florence platform, Intel emphasizes the fingerprint recognition technology, a small auxiliary LED on the lid of the two portable models, the detachable and usable as a tablet screen of the On the Go and the keyboard-integrated detachable remote control of the Mobile Entertainment PC. All this stuff looks alluring, but also reasonable considering the existing market trends.
Unfortunately, power elements as the power source are not yet written into these trends. There is too much work to be done in this field yet, although Intel introduced one concept at IDF, too. A 100ml cartridge gives a power of 12W and that’s quite enough for a PDA, but insufficient for a modern notebook. Moreover, notebooks experience power surges up to 50W. So this concept will be under consideration until 2007.
The new generation Centrino platform, called Sohoma, on Dothan core, will come out earlier, of course. If no problems arise with the 90nm process, we will see it in April-May. Good news – it’s probable that two chips, 1.7 and 1.8GHz, will come out. Until then, notebook makers have to consider Prescott: they made notebooks with the 130nm Northwood, so why can’t they do the same with a 90nm Prescott, even without waiting for the mobile version of the processor? Of course, when working with such a notebook, you shouldn’t go far from the wall outlet.
First models are scheduled for May and should be selling at a relatively low price: desktop processors are always cheaper. By the way, the very appearance of Prescott made the prices in this market go down: the official price cuts affected the Mobile Pentium 4 (on the Northwood core, of course) and quite severely: from 14% to 32%, so the 3.2GHz chip got cheaper from $433 to $294 in a jump. The Mobile Pentium 4 on the Prescott core, just like the Pentium M on the Dothan one, is likely to arrive in the second quarter. Intel didn’t stop on the Mobile Pentium, but continued with the Mobile Celeron, although the price cuts were lower here, by 10-14%. In the opposite camp, AMD was reducing the prices for the Athlon XP-M, announcing a new model on the way, the Athlon XP-M 2600+. Compare: the Athlon XP-M 2600+ for $99 against the Mobile Celeron 2.2GHz for $96.
There have been certain advances in the mobile graphics area: NVIDIA Quadro FX Go1000 is only used in the same Dell Precision M60 as its predecessor, while the ATI Mobility RADEON 9700 is a more democratic and interesting solution, although not quite low-end. In fact, this chip is not a hi-end, either, as it’s just the mobile version of the RADEON 9600 XT, but this chip is not bad for mobile computers itself. That’s most important, notebooks with this GPU are already available in shops.
Overall, the notebook market is expanding. Notwithstanding 10% poorer quarterly shipments from Taiwan (because of the disastrous January), the quarter is going to be 10% better than that of the previous year. And the manufacturers hope for a strong demand in March.
As for particular models, Centrino platform is experiencing a boom. In February, IBM offered two X’s: the ThinkPad X31 (a new version of the old model) and X40 (an extension of the X31). The result is nice: 1.3kg weight, small dimensions, up to 7.5 hours battery life, and integrated Gigabit Ethernet. This is a good business machine costing you $1499 for a configuration with a 1.3GHz Pentium M.
Toshiba Tecra M is another variation of the same theme: a business computer with a 14.1” screen and 1.4-1.7GHz Pentium M, with GeForce FX Go5200 graphics and (optionally) Gigabit Ethernet. The last feature may not be very useful, but that’s something new. Other things like 2GB PC2700 DDR and 40-80GB HDD are too trivial to be mentioned. The price is similar to that of the X40: starting from $1,500.
Some people like it heavier and cooler like the Klondike Referent 200 and Dell Inspirion XPS, although these two have different weights. The former has a heart made of the desktop Pentium 4 1.7-2.6GHz or Celeron, a 15” screen and ATI Mobility RADEON 9000. The second is a monster on the 3.4GHz Pentium 4 Extreme Edition with ATI Mobility RADEON 9700, all imaginable ports and connectors, an appropriate HDD (up to 80GB capacity, 7200rpm speed) and so on. The price is biting, though, starting from $2600, and you’d better not think about the time of autonomous work. If you need an easily transportable desktop computer, that’s the solution for you.
