iTotals - September `02 (IT Monthly Roundup)

Processors

as per the 26th August presentation, prices for Pentium 4 2.6 and 2.8 GHz dropped by 22%, while by 55% for Pentium 4 2.4;

• presentation of the Xeon 2.6 and 2.8 GHz processor having 512 K L2 cache;
• launch of the first low power consuming Xeon processor: 1.6 GHz at 30 W;
• release of the Pentium 4-M 2.2 GHz and 10 more various mobile processors;
• start of sales with a further official advertisement of the Celeron 2.0 GHz;
• announcement of the 933 MHz Intel LV Pentium III whose dissipated power is as low as 12.2 W;

Such a formidable list would have been more than enough, wouldn't it? What is more, the IDF was held, a really remarkable event with numerous demonstrations of Banias-based notebooks as well as a Pentium 4 @ 4.7 GHz on the base of the new 0.09 mk Prescott core. The former proved more attractive: its 77 mln transistors which for the very first time aim at the mobile use have every reason to become a true brilliant solution for the sphere, and first Banias-based notebook prototypes are proving that. Intel may even have to use their processor rating to make the buyers aware that a 1.3 GHz Banias is by no means worse than, say, a Pentium4-M @ 2.0 GHz, although Intel's office in Russia strongly deny the fact.

At the IDF, the server front was not passed over, although in the form of outlined prospects: already in this quarter the release of Xeon with the FSB533 is planned, and early next quarter the speed of these processors will reach 3 GHz - as it should be, some time later after a Pentium 4 with the same clock speed has been released. In the fourth quarter this year a 0.13 mk version of the Xeon MP will come out - the long-awaited Gallatin core in which the L3 cache will reach the customary 2 MB. The following forecast was also made public: during the upcoming five years Intel will bring out a 64-bit processor with two processor cores on the same chip. This is quite logical a step which is in line with the most recent trends in the industry of super-powerful processors.

And finally, there's one more sphere where Intel also has left traces - it's new technologies. At the International Solid-State Devices and Materials Conference the company proclaimed their new proposal in the part of creating new chips which solves one of the basic problems still on the agenda of the semiconductor industry: the leakage current. A three-dimensional triple-gate transistor allows to reduce it by 100-200 times relative to the current level.

By the way, a bit earlier AMD who created the world's smallest transistor having two gates and based on the Fin Field Effect Transistor (FinFET) technology manifested their contribution in the progress development. Potentially, this technology allows to create elements with the gate as tiny as 0.01 mk in size. Let's wait for the coming January though when the details of the development will be made public.

But on the real-life front the company's affairs are not that bright, and it absolutely does not make sense arguing that AMD is at least in the tactical crisis. With the recently released Pentium 4 2.8 GHz the consumers have spent a whole month in vain waiting for the release of Athlon XP with FSB333 - this processor has not been produced so far. And in the middle of the month it was announced that the Clawhammer release has been deferred from the fourth quarter this year to the first quarter of the next.

At the same time the Barton has also been moved to the first quarter of 2003 because of the delayed release of Athlon XP FSB333: it is too much producing the former and Barton at the same time. What is more, it turned out that SOI will not be applied in Barton because the technology is still poorly debugged. This reveals something about problems with Hammer. The overall picture appears to be rather sad: with all the good will, it is very very hard to see bright prospects for AMD in the competition '0.13 mk Barton and Hammer vs. 0.09 mk'.

Among the real advancements AMD can boast are merely a couple of new mobile Athlon XP processors which are in fact nothing out of the ordinary. Their XP 1900+ and XP 2000+ offer nothing more than support of PowerNow!. At the same time, prices for the outdated predecessors in this line have gone down.

By the way, as far as the mobile market is concerned, Transmeta has woken up. The company first announced the HP plans to represent their Crusoe-based portable PC on 7 October, secondly, they boasted having a 0.13 mk 1 MHz Crusoe TM5800 sample, and thirdly they promised to produce a 0.13 mk 1.4 GHz 256-bit Crusoe TM8000 next year.

And the last novelty of September was a new 64-bit Sun - UltraSPARC III Cu 1200 processor. As usual, it is a Texas Instruments produce, but what is unusual is this - the 0.13 mk copper process technology was used for the very first time, which gives the following result: compared to the predecessor, 1.05 GHz UltraSPARC III, as the frequency rises by 14 per cent, the power consumption drops considerably, from 75 to 53 W. In fact, this is no more promising than the same results of transition demonstrated by other manufacturers.

Motherboards

Basically, there has been more than enough said about these chipsets to bring up the topic here as well - we can only establish that judging by the first tests the products appeared to be in a very Intel-specific style, in the good sense of the word: no extravagance or frills, but what is there is implemented at full 100%. Of the features to get excited about - not the 845PE itself but the motherboards based on that - it's worth mentioning that the manufacturers have already started even using Gigabit Ethernet controllers in them. (By the way, manufactured by Broadcom. And ServerWorks, their daughter enterprise, is beginning to integrate Gigabit Ethernet into their new southbridge chips).

It's future versions of the chips that arouse growing interest because not so much time is left until they appear live on the market. Springdale, a successor of the above mentioned trinity, became available already in September as a pre-production model for motherboard manufacturers, but we still have to be patient for at least nine months until boards based on this dual-channel DDR chipset have been released. Augmented with support for Serial ATA and AGP 8X, it is going to be one more 100% hit.

We don't have to wait until the Springdale has been released though to see the new chipset live: Granite Bay, the first Intel's dual-channel DDR chipset is to be announced already in October although it is aimed at other niche - solutions for workstations etc., i.e. it is a direct competitor to i850E. AGP 8X is already present in it, but still there are problems with Serial ATA because today's southbridge chip ICH4 is used, so of the benefits offered we've got to restrict ourselves to USB 2.0. The chipset is coming up very soon, so more than enough motherboards based on that were demonstrated at the last IDF. Placer will appear a bit later, and its level is much higher: the dual-channel DDR is augmented with support for dual-processor configurations based on FSB533 Xeon, PCI-X and again AGP 8X.

VIA is continuously losing its positions, which is seen from the fact that nothing worth of attention was presented in September. But what is the criterion for 'attention'? Cessation of KT333 supplies? Or release some new KT400-based motherboards? Maybe declared intentions to produce KT400A early in 2003 with a new revision of the unsuccessful P4X400? In any case, VIA is following its former route.

By the way, this has been proven by the August results - owing to KT400 AMD increased sales by 6.4% compared to July but anyway failed to achieve the formerly planned figures, while SiS succeeded in increasing sales by 17.6%! All that became possible due to providing solutions for Pentium 4. So we have every reason to expect the same trends onwards because recently new SiS 648-based motherboards started being produced and proved quite successful; in October, a mass production of SiS746FX chipset for FSB333 Athlon XP supporting AGP 8X is to be started, and in November the first ever dual-channel DDR chipset made by SiS will come up. As you can easily guess, it will be supplied for Pentium 4, with features better than those offered by Granite Bay. For example, it is support for PC2700 DDR and a more flexible operation with various DIMM modules configurations.

There is one more player on the chipset market, almost forgotten completely, but who's been very active all this time: as it has turned out, early in October mass supplies of new Socket-A chipsets made by NVIDIA, nForce2, will begin. This time NVIDIA seems to have done a much better job at the "price-performance" ratio for their new product (by the way, also a dual-channel DDR), so the chip is having quite good chances for success. What is also eroding VIA's share of the Socket-A market is their fading profits.

Memory

No doubt DDR is de-facto the most popular type of memory modules (some largest DRAM manufacturers, with Samsung among them, say the DDR share in the total production volumes already exceeds 50%, and by the beginning of this autumn the overall market average DDR share has slightly outnumbered the SDRAM share - 45.5% vs. 45.1%, respectively). While PC2100 is still the king, in the nearest future as motherboards based on new Intel chipsets get more common the demand will be increasingly shifting towards the PC2700. With PC3200 the situation is far more tense, although in general it's already clear as noonday: Intel are strongly against the standardization of this specifications, and the arguments being advanced are cogent enough: the performance gain is in fact negligible. In this regard it is significant that SiS and VIA who initially tried to proclaim support for PC3200 in their chipsets had to give up their statements very quickly.

Therefore, we've got to wait for DDR-II, while some of us already does not have to wait for that a long time: ATI and NVIDIA intend to present the prototypes of graphic cards which use these chips by the end of this year, although the vast majority of DRAM manufacturers will be able to present pre-production models of their DDR-II memory chips by no earlier than the beginning of 2003. JEDEC also intends to adopt the graphic version of DDR-II as the reference standard, and the work at it is already underway.

So in one way or another, the DDR will survive, and however strange that may sound, this seems to hold true for the RDRAM either. See for yourselves: at IDF, Intel officially announces support for the PC1066 RDRAM through i850E, and Europeans acknowledge lawfulness of one of the patents to do with SDRAM/DDR, and memory manufacturers start producing 32-bit RIMM modules. Following suit, motherboard manufacturers announce their RIMM-enabled models! So everything is not going on the way it seemed quite recently. Moreover, Intel's mood has been changed again - the company is still hesitant while choosing the best options while making up new strategies of market entry. You never know - technologies are improving and RIMM prices will go down anyway, but even RDRAM rivals can't deny that in terms of ideas behind it this memory type is rather promising.

This month is also remarkable in terms of the "high life chronicle". Finally acquired by the board of creditors, Hynix is again trying to buddy up with Micron on the same subject. Micron seems to get bored to death with that, so further talks on Hynix acquisition are very unlikely to bring any fruits. For the time being, Hynix, a Korean company, is still an independent entity, due to which South Korea is an undisputable leader at DRAM production - in August, Samsung and Hynix released 49% of all the overall world's memory chips produced. By the end of the year Samsung alone will grab over 30% of the market.

Other players are consolidating: rumours have it that Mitsubishi are selling their DRAM business to their nationals, Elpida, while Intel and Powerchip intend to cooperate somehow with Elpida. Since Mitsubishi owns a formidable share holding at Powerchip, the information seems quite as logical as the list of companies not discovered in whatever alliances of any kind. Although there is in fact one company, NEC, one of the founders of Elpida, who intended to develop magnetic RAM jointly with Toshiba that combines the benefits of both DRAM, SRAM and flash-memory which seems like a "sacred Grail" for memory manufacturers today.

Despite being quite ambitious, this development can't catch up with the advancement achieved early in September by HP - creation of a nanoelectronic RAM chip. Interlacing of 8 vertical and 8 horizontal threads produced 64 node molecules whose condition can be altered and read electronically. Totally, this gives 64 bit per 1 mk. What is more, there is a technology ready for the industrial application.

Video

By the way, a familiar name - Gainward - was added to the list of ATI admirers in September. This is one more confirmation that ATI's business is on the up. There are inverse symptoms though - the volume of sales for the past quarter as it turned out has reached merely $240 mln, with zero profits gained. All this is far much lower than planned and 30% lower than a year ago. In fact, it's ATI to be blamed for that, since they underestimated demand for the Radeon 9000 product line.

Not all is going smoothly with Radeon 9700 Pro though, because it turned out that the 128Mb versions do not always run correctly with the AGP 8X motherboards which sometimes even fail to start up. That could be a splendid marketing trump for NVIDIA which has always been distinguished for the aggressive marketing towards video cards manufacturers. In this regard, I wonder what the 256 Mb Radeon 9700 Pro video cards already announced by some manufacturers will be compatible to?

What is left is to hope there won't be similar problems with Radeon 9500 whose specifications are already appearing in the Internet. AGP 8X and DirectX 9.0 will stay there, with the number of graphics pipelines being "merely" four, and all the standard technologies supported as well; the speed of 128-bit (not 256-bit) DDR SDRAM will slightly drop to 550 MHz. At $180 (Radeon 9500) or $220 (Radeon 9500 Pro) - isn't it the best choice!

In the meantime, two more new products have been presented: at the very end of the month a new generation of AIW video cards - ALL-IN-WONDER 9700 PRO was brought out. This is all clear without any comments. A week before that a new graphics chip for workstations, MOBILITY FIRE GL 9000, was announced. Also goes without saying for MOBILITY RADEON 9000 whose drivers are optimized for CAD/CAM and DCC applications.

For NVIDIA, September was not so fruitful a month since they had nothing much to boast with - the company announced its AGP 8X chips on as late as 25 September. On the other hand, the first respective cards were announced by some manufacturers already in the middle of the month although they are expected to hit the shelves no earlier than in October. Basically, there is nothing of interest here - the speeds and functionality are the same, only support for AGP 8X has been added which in fact gives merely 4-5% performance boost.

Generally, softly speaking, a bit low on this. It's worth mentioning the court pleadings with Microsoft who are demanding deliveries of chips for XBox at maximum reduced prices, though Microsoft is still unsuccessful at it. It's understandable that this by no means can have a positive effect on NVIDIA suffering currently. All the hopes are pinned on the release of NV30 which is coming up no that soon - no sooner than Comdex Fall takes place in November than official advertisements with demonstrations of pre-production models will be made public.

A few words about AGP 8X and AGP in general. IDF left its traces in here as well - the final AGP 3.0 version was proclaimed which defines the AGP 8X mode. So, on formal grounds Radeon 9700 Pro is a bit of an illegitimate child in contrast to the NVIDIA produce. This immediately reminds of the above mentioned Radeon 9700 Pro incompatibility to some motherboards running in the AGP 8X mode:

Data storage

In fact, there is not much to speak about the CD-RW: all the same as before. A few models of 48X CD-RW drives have been announced, and the prices for low-end models have gone down: in Taiwan the retail prices for 16X models have gone down to $40. So have the prices for CD-R drives, and the trend will stay: in September, three largest disk drive manufacturers received a permission from the government to build three new factories in China at 30 mln disks/month of total production output. Also keep in mind that Sanyo, HD-BURN have presented a second technology for doubling the capacity of standard disks (up to 1.4 GB). Reading such disks will be effected by any of the currently existing disk drives - it suffices to update the software only. All in all, the CD branch is evolving at its very natural pace, and it's still a too long way to go until it dies away.

While the DVD is still at the beginning of the road, it is evolving at the same pace. Sony has produced their first comprehensive drives which are able to read (and write!) all the DVD formats at the same time, although the practical sense of that is not so evident - there are hardly enough users willing to write DVD-RW disks on even days and DVD+RW on odd days. In any case, the user would rather prefer a certain format, and it's up to the manufacturers of DVD-ROM drives and other reading devices to decide whether to support one format or the other suggested by the competitor. This should be better addressed to Panasonic who developed a chipset supporting the DVD-Multi, i.e. providing right what we mentioned above: operation with DVD-RAM, DVD-R and DVD-RW.

There is one more benefit in writing devices - new DVD-RW drives have appeared recently on the market which provide the 4X recording speed and 2X in re-recording, although the respective specifications were established no earlier than in August. What is more, Mitsubishi has already released a laser providing the 8X recording speed, which will allow to write a standard 4.7 GB disk in as fast as 7 minutes.

As to the hard disk drives, the past month proved extremely attractive, with Maxtor being the undisputable leader who early this month presented their new 40 GB models Fireball 3 and DiamondMax Plus 8 as well as professional DiamondMax Plus 9, DiamondMax 16 models having up to 160 GB in capacity, and then a brand-new line, MaXLine, aimed at operation on a 24/7 basis. This is still UltraATA/133 but the capacity increased to 320 GB and the mean-time-between-failures being over 1 mln hours. And finally, in about the middle of the month new models of fast hard disk drives, Maxtor Atlas 15K (15,000 rpm Ultra320 SCSI), were announced having the capacity up to 73 GB.

At the same time, the guarantee issued by Maxtor for their HDDs has been reduced from three to one year. On the other hand, almost the same was announced by Western Digital and Seagate, so this is at least not a unique occurrence but more likely to be a trend. Not that the disks turned less reliable - on the contrary. With the competition and price wars getting fierce, the manufacturers simply found a new more niche for savings. Let the reliability be high at short guarantee term rather than the contrary. A reminder - Fujitsu had to withdraw up to 300 thousand of their HDDs produced during the last year.

By the way, Maxtor was far not the only manufacturer of hard disks who showed activity in September: Western Digital informed they launched their next model of the Caviar SE family (with a 8 Mb buffer) having already a 200 GB capacity and 7200 rpm speed. A real monster among the ATA monsters. WD although have something to offer in the sphere of USB 2.0, but the maximum capacity of the new line is reduced to 120 GB. Anyway, this capacity will be more than enough for 99.9% of tasks.

With all this going on, Seagate has had to demonstrate only their Serial ATA product. Yes, this time it was Serial ATA II! The prototype of such a HDD was presented by the company at the recent IDF, and we are looking forward to seeing the first generation of Serial ATA Seagate HDDs already in this quarter. But SerialATA II products will appear on the market in no earlier than two years - in the second half of 2004.