Processors

Even without Microprocessor Forum this month would have been overloaded with events, and with it - even more. But first let's look at things closer to real world.

Celeron 766 MHz that appeared on Japan markets in the beginning of September became a month's mystery. And in addition Celeron 700 MHz still is the most speedy of the line. Looks like "time gap" that Japan got into continue to widen more and more.

But Celeron is the last in Intel's plans. Pentium 4 is their main. Concerning the fact that 1.13 GHz P-III release was moved to the second quarter of '2001, P-4 is the only competitor of Thunderbird and Mustang, but it's still unreleased and its destiny is unknown. And though Intel solved release problems - a correct version of ICH2 is on markets already, perspectives remain foggy.

Preliminary engineer model tests showed very bad results: 1.2 GHz P-4 is slower than 1.1 GHz P-III and Athlon. It's clear that Intel must rapidly increase clock rates and it would possibly work in the beginning: extrapolating the results we can say that 1.4-1.5 GHz P-4 would be like 1.2-1.3 GHz Athlon, but the second would be cheaper because P-4 will need at least two RIMM memory modules and it's $400-500 for 64 MBytes, etc.

But another situation would be in the second half of '2001 - here Intel will surely lead with its 2 GHz 0.13 micron Pentium-4 and AMD doesn't have anything to beat it, because K7 would be "depleted" by this time. And then Intel, having put out its P-4 and assured that P-III architecture can't beat it, will release 1.26 GHz Tualatin as effective way to put AMD away from Mid-end markets.

We wonder what price would be declared for this processor, because by the spring 1.4 GHz P-4 will cost about $510 what is $90 more than price of 1 GHz P-III. And it outruns P-III in a good pace, though the price of other components bites a bit. However we should say that P-4 would be a processor for the masses next year and P-III won't be an obstacle.

We can't leave aside a fact that difference between Celeron and P-III begins to disappear. By the end of January with 800 MHz Celeron release and lowering prices for P-III we get such a situation: 800 MHz Celeron costs $170, Pentium III 800 MHz - $183. As a result we get a nice logical chain in which 800 MHz FSB100 Celeron becomes a link between FSB66 Celeron 600-766 MHz and FSB100/133 Pentium III 850 MHz-1 GHz and all this line turns out to be a number of Low-end processors that we know as Celerons, and P-4 takes niche of P-III and everything is ok.

And speaking about mobile processors, we should note that Intel works seriously in this part. Mobile 1 GHz P-III is to be released in the first half of '2001 and Intel would have been the first to produce 1 GHz mobile if there wouldn't have been AMD with its mobile K7s. And they show good speed as well. But Intel is one step ahead: there's a new mobile core being developed from the blank. Speaking theoretically, if it'd show the same potential as P-4 core then Intel would win this round as well.

And AMD is happy in the present. Having triumphed in the last quarter, having declared remarkable profits, this company is going to step this way in the future as in quatities (28 million processors this year vs. 18 million previous year), so in technologies (Duron will reach 1 GHz and Athlon - 1.5 GHz in the first quarter of '2001, Palomino and Morgan are declared - Athlon and Duron with the Mustang core). And now there are 1.2 GHz Athlon and 800 MHz Duron that cost less though working with the same clock rates.

And in the very end of the month AMD released analogs of FSB 133 MHz P-III - Athlons using 266 MHz system bus clock rate, i.e for working with KT133A, AMD-760 and other new chipsets.

And now let's talk about the main event of this month - Microprocessor Forum'2000. We should say it was rather dull this year due to vast amount of network procesors that are "in fashion" today.

Here you should pay your attention to SiByte's products - new Mercurian processors created with guidance of Jim Keller, creator of Alpha and Athlon. Even the first model looks very impressive: 2 parallel 1 GHz cores in one chip with one common L2 512 KBytes cache that has memory with 50 MBytes/sec capacity. The productivity is 4400 Dhrystone MIPS and this is good for a 2.5V chips. It also includes all System-on-a-chip features: PCI interfaces, Ethernet, etc., and LDT system bus in addition. This model would be released in the middle of '2001.

By the way, something about System-on-a-chip. STM made a good step and released three products at once: STPC Atlas, STPC Consumer-II, and STPC Elite. All of them have ISA/PCI controllers onboard, SDRAM interface, IDE port. Two elder models (Atlas and Consumer-II) have a graphic cre in addition, and Elite has 16-bit common I/O block. Altas also has Super I/O module, mouse&keyboard controllers, 2 UARTs, USB hub and a PCMCIA controller. Consumer II and Elite models would be available in October, and Altas - a couple of months later.

Speaking about classics we attract your attention to IBM that presented two server processors at once: Power4 and Freeway. First of these was previously described - two 64-bit processor cores on one chip, each with it's own L1 cache (32+64 KBytes), common L2 cache - three modules with 512 KBytes and it's own controller each. As a result L1<->L2 capacity is 100 GBytes/sec. Theare are also 32 MBytes L3 cache controllers. This product is to be released in the second half of '2001.

The second combines maximum reliability with high productivity: doubled instruction/execution modules, cache parity control, L2 cache ECC, data compression module included, quadriple pre-execution buffer and a lot more for making this 64-bit product "run" faster. It would work with 770 MHz clock rate, but will reach 1 GHz next year like Power4.

Motorola at last presented it's G4+, Apollo-G4 with extended conveyor, made using SOI technology. As a result base clock rate should increase to 700 MHz (that's 1.5X more than today's) with a perspective of reaching 1 GHz. Apple haven't said much about this product, but Jobs is Jobs and the way of Apple is much unlike all others.

VIA acted well, having presented several generations of x86 processors at once. 0.18 micron C5A - 500-667 MHz that is produced now; recently announced 0.15 micron C5B - 600-733 MHz that is mostly mobile; 0.13 micron C5C - 733-867 MHz with common 192 KBytes cache that would be produced in the spring; and 0.13 micron C5X with renewed architecture, extended conveyor and SSE support, clock rates would begin with 1.2 GHz. So VIA quickly understood that GHz are in big demand now.

Motheboards

It's hard to name "an event of the month" in this part of our digest, but still we'd say that the most interesting information is about Standard Microsystems Corp. (SMC) - a new PC chipset manufacturer that has a license for Intel's 440BX that would be produced as Victory66. 66 means ATA/66 support in South bridge, but the question about AGP 4x support remains unanswered. But we think that these products would be in good demand,because 440BX is in ideal state now and chipset manufacturers will continue to produce it for a long time. But here appears another competitor - Apollo Pro133A and KT133 that were produced with old 686A South bridge replaced with 686B with ATA/100 support. So motheboards like Chaintech's CT-7AJA/100 appeared - the same CT-7AJA but with ATA/100 support. Though classics is not in fashion because now is the time of their integrated "brothers". SiS630E, PL133, PM133 based motherboards continued to appear, though i815E remained favourite. Manufacturers like to invent something with this chipset, add RAID support (Abit's SA6R) or even dual processor mode support (Epox EP-D3SA), and in addition: Intel included SMP support in its chipset, impossible and undocumented but true. Intel also annuonced two other versions of this chipset: i815EM and i815EP. First is mobile to replace 440ZX and 440BX and to make notebooks real competitors to usual PCs. The second continues 440BX's traditions - graphic core was excluded. Intel decided to leave cheap i810E2 in the integrated chipset sector of the market and to make i815EP an alternative to DDR SDRAM VIA and ALi chipsets. Company wants to push this product quickly to the markets and concerning that manufacturers won't have to change i815 design, concerning parameters like AGP 4X, ATA/100 and PC133 SDRAM support, and price of less than $30 - we'd be presented a hit of spring season.

Though there's an opposite opinion, because this winter/spring season will have many products released that pretend to gain much attention. We speak about nVidia and ATI's integrated chipsets. Both companies have come a long way to this part of market (it took a year a half already), but they walked in like real triumphants.

Just look at nVidia's plans about Crush-MCP1 line base on NV11 core - in February and MCP2/MCP3 based on NV17 - in the end of spring/beginning of summer. We should await DDR SDRAM support, Pentium III/Celeron and Athlon/Duron support as well. ATI presents its ArtX2 in February, we hope it'd be interesting to look at. And motherboards manufacturers speak with enthusiasm about these two "newcomer" companies.

Micron is another new company on the chipset market and it seems to be ready to enter it. On the last MPF it traditionally presented newly deveoped products - an 8 MBytes L3 North bridge cache Athlon chipset Samurai that is to replace Mamba. And Micron seems to get father than just demonstrations: another 3 chipsets are being prepared for release in '2001 - Scimitar, Mamba and Shogun. But these are not integrated solutions, maybe opposite - just remember Samurai and Mamba specifications and you'll see that Micron likes server shipsets more. In addition there's no analog of Serverworks for Athlon, so Micron has chances to make one.

And AMD itself doesn't seem to go father than basic set. On MPF it presented a new dual processor chipset based on Athlon and AMD760MP that allows Athlon to reach server platforms. It's basic because dual-processor chipset isn't good enough for powerful servers, but in the same time there's a possibility of uniting several North bridges on one board each of which would support 2 processors (as Intel does). And, besides, Intel's chipsets aren't much better in this situation.

Intel will ahve only few of such chipsets next year - i870 and Plumas, everything else is taken by ServerWorks and VIA's PX266. And i870, made for using with McKinley and supporting Foster, won't work with this last in multiprocessor systems. So the only good things in Intel's plans are Almador and Brookdale - chipsets with DDR SDRAM support (it's potential for Almador and Tualatin will have it anyway - Apollo Pro 266T) for Pentium III and Pentium 4. So Intel begins to restore it's position in the field of classic chipsets though there's still a lot to do.

Because first - there's an announced Apollo Pro266, manufacturers already begin to release boards based on it: Epox, Transcend, LuckyStar, MSI, Chaintech. And second - all manufacturers awaited AMD-760 for the whole October and when it was released at last they presented several boards at once: Asus A7M266, Gigabyte GA-7DX and FIC AD11. All these models are based on combined AMD-761 + VIA 686B. It's much alike the situation with first generation Athlon motherboards when there was a AMD-751+686A combination.

As chipset was released in the end of the month, you should expect motherboards based on it only next month, as well as KT266 boards. And if you note i850 release in the end of October you'll see that autumn will be very interesting. Besides Pentuim 4 motherboards are presented even now: MSI MS-6339, AOpen AX4T.

Memory

Everything is silent here. Memory prices are still crawling down all the month. By the middle of it, 64 Mbit chip price was less then $6 and then - even $5 per chip. There are many reasons to that: general state of the market with companies continiuosly declaring negative profits, not so big demand that remained still since the beginning of autumn, cancelling of antidamping quotes for korean DRAM production in USA, and change to 128 Mbit as well.

The situation resembles that of last year - supply surplus, price crash making production unprofitable for some manufacturers, their leaving market, and last autumn crisis. But it won't be the same again: 0.18 micron technology greatly helped DRAM manufacturers that remained. And also X-mas is at hand.

Rambus continued its strange activity annoying all industry. There were some unusual juridical moves when company asked FTC to cancel case of Hyundai and then took it from one federal court to another.

But it's time to build own defences. Because one of Rambus's "enemy" - Infineon began an attack, having declared that Rambus violated it's own patents. The second case belongs to Hyundai that accuses Rambus in violating antimonopoly laws. We wonder if Rambus that is a rather small company will have enough money to pay for all these court things. Company's profit for the last quarter was 14.5 million dollars - first payings for SDRAM from Hitachi, Toshiba, etc.

And Rambus shouldn't rely on Intel because it quickly cuts all threads that were connecting them. There were some serious declarations about using DDR SDRAM, Barret's phrase about RDRAM's losing its positions, etc. So Rambus went against everybody and haven't gained any profit from it, and Intel haven't get much from RDRAM as well.

And DDR SDRAM continues to strenghten its positions. Transcend, another memory brand, declared the beginning of 128/256 Mbit DIMM DDR SDRAM modules production - PC2100 and PC2600. There shouldn't be any problems with it in winter when first motherboards with DDR SDRAM support will appear.

Video

It became a tradition already that nVidia makes all news here. Or those that produce cards based on it's chips. So a half of October news would be about this. Asus tried hard to release a whole bunch of products: V7700 Pro and V7700 Ultra two High-End lines that are much similar to each other, with slight differences in core and memory clock rates. And a new hit that will outrun Matrox Marvel and ATI All-in-Wonder - V7100 All-In-Deluxe: GeForce2 MX based card that that provides all features of those two: videocapture, digital VCR mode, etc. This card's parameters are closer to Marvel 450 due to TwinView (DualHead analog).

But others were active as well, makinhg accents on High-End products. There were enough cards, based on GF2 Pro and GF2 Ultra, announced in October: Hercules 3D Prophet II GTS Pro, AOpen PA256 Pro II, ELSA Gladiac Ultra, Canopus Spectra 8800, VisionTek GeForce2 Ultra, Gainward CARDEXpert GeForce2 Pro/400TM.

Creative showed everybody that 32 MBytes cards based on GF2 Ultra are meaningless, and announced another AGP/PCI 32 MBytes line - based on GF2 MX. The declared price was $110-120 and it is good enough. But why take one of these if the same company will release another 32 MBytes card by X-mas, but it'll use DDR memory for a price of $129. And having in mind that these cards will be mostly used not on 1 GHz machines, we can say that 64-bit memory won't be the main restricting factor and there would be a balance between CPU power and graphic subsystem productivity, providing nice price/quality rate.

So nVidia has enough business for manufacturers and they are not eager to work with new solutions during following months. And that is the reason due to which nVidia doesn't hurry to release NV20. Though technical side of the question is important anyway, because first test model working on 250 MHz instead of 300 had been gotten only in the middle of September.

But they don't have to hurry as well, because noone would be able to compete with GF2 Ultra, not speaking about NV20. So the company can take some rest this time if there's any with all these monthly upgrades. Now turn to competitors. 3dfx is trying to restore its financial position and it doesn't have time for some new products. Though CFO changing isn't so helpful in this case, and release of Rampage or GP-3 would help more.

But of course VoodooTV line looks very nice. These cheap TV-tuners have their own niche and combined with any videocards present better solutions than All-in-Wonder or Marvel. So these tuner line can give 3dfx a chance to withstand hard life until Rampage release.

After nVidia only ATI looks better than everybody else, because many products are to be released in the first half of '2001: RV100 (Radeon Value - GF2 MX competitor), R200 (second generation Radeon), mobile Radeon, R300 that is planned for the second half of '2001 and that promises to NV25's strong competitor. By the way, more about it: the company announced it's "predesessor" - first mobile chip with AGP 4X support - Mobility M4. ATI did it before nVidia that is going to release mobile GF2 MX.

Who's left? Matrox? It earns money selling G450 and winning OEM competitions over manufacturers like Dell. G600 is missed for sure. And G800 is becoming a problem even now. It won't beat productivity records so company is preparing a dual-chip solution - Matrox Fusion, that would be bad variant, remembering Rage MAXX and Voodoo5 5500, because cost would be too high.

And STM with it's Kyro are losers again. This card is better than previous but still bad. Raw drivers, bad productivity, price higher than that of GF2 MX, though it is faster. So manufacturers are not interested. Try again?

Storage devices

A very eventful month in all branches and fields. October is surely "the month of storage".

Not speaking about improving old models (12/10/32 drives from AOpen, Iomega, Imation or Plextor, or even "budget" Ricoh 8/8/32X MP7083A for $179), we should note that industry began to change for new 16/10/40X controllers. Yamaha was the first to announce LightSpeed CRW2100 line. There were two models released in October - internal ATAPI and Ultra SCSI, and 16X CD-R discs were announced as well.

Looks like HDD manufacturers want to beat all records. Maxtor with its Maxtor External Storage was the first in the month. This drive is a usual 80 GBytes DiamondMax with a FireWire interface. Then goes Fujitsu: MPG3xxxAT and MPG3xxxAH-E lines that use 20 GBytes discs with spindle rotation speed of 5,400 and 7,200 rpm. There are some other differences like 512 KBytes and 2 MBytes cache, standard versions and quiet with liquid bearings, etc.

The further the better. One after one manufacturers began to announce 73.4 GBytes models. First was Quantum with its Atlas 10K III, shippings will begin in the 1st quarter of '2001. This model: 10,000rpm, Ultra160/Ultra320 SCSI interface. Then IBM announced Ultrastar 73LZX. Same as the above plus additional 1-2 GBytesit/sec FibreChannel interface. A couple of days later Hitachi announced DK32DJ-72: 10,000rpm 73 GBytes and a half, first quarter of '2001.

Only Seagate didn't announced such a drive. It just presented Ultra320 SCSI HDD. A real one, with full support of Ultra160 SCSI, including cables, terminators, etc. So accepting Ultra320 wouldn't be so hard.

By the way Seagate announced new Barracuda 36ES, 18/36 GBytes Ultra160 SCSI HDD that is a Barracuda ATA modification with SCSI interface and nice price. And 73 GBytes model - Cheetah 73LP would be released in the same 1st quarter of '2001. But look what's ahead. Real monsters! 140 GBytes Barracuda, 15,000rpm Cheetah X15 36/18 GBytes with Ultra320 SCSI interface.

140 GBytes impress but don't surprise: in October Samsung and Read-Rite announced a new record - 60 GBit/sq.inch or 80 GBytes per 3.5'' disc or 0.5 TBytes with 6 discs. And there are technologies that allow to write 300 GBytes per disc or 18Tb HDDs with 6 discs.

But back to Seagate. Except HDDs it announced some exotics: shippings of Viper 200 streamers and Ultrium 1 cartridges for them - 200 GBytes per cartridge using autoloader provides archiving of up to 2Tb of data at speed of 115.2 GBytes/hour. This is more than enough for today.

There's some information about flash cards. Viking announced sales of its extremely "storageable" models of CompactFlash cards - 192 and 256 MBytes. In the end of the month Simple Technology again announced the most "storageabe" CompactFlash card - 512 MBytes. Doubling space for one month is not just good - it's outstanding. The card is based on 512 MBit flash chips that are already being used by manufacturers. In October Hitachi presented 128 MBytes MultiMediaCard based on this type of flash chips - 2 chip s along with 32-bit RISC SuperH processor, relatively cheap and simple.

Now let's observe new technologies as there were plenty of them. All concerning optical discs. CD-RW drive compatibility was in demand, and the most cheap variant is MultiLevel Recording technology from Calimetrics, that is supported by TDK since October, 2 GBytes per disc, 36X write speed. All this after adding a small microchip to today's CD-RW drives.

What about DVD? On CEATEC Sony presented a prototype that uses GaN laser whose wave length is enough to create DVD discs of 22.5 GBytes per side. And this is 2 hours of HDTV video or 8 hours of usual. But this is still a prototype though it wotks ok. But the question is at hand already. So on the closest conference of DVD technical coordination workgroup forum would be a question about DVD specifications.

If there are no problems with compatibility then take a look at C3D that slowly but assuredly prepares everything for Fluorescent Multilayer Disc. Even on the Comdex there was first drive for such discs, though official release of the product using cheap red laser and compatible with CD/DVD is set only at the next year. There's already a partner for disc manufacturing - Zeon Chemicals.

By the way, FMD idea continues to expand. A new technology of this class was developed in Romania, its creators promise 10Tb discs already in this year. And the period of safe data keeping is not less than 5000 years.

Now some high life chronicles. The main event is merging of Maxtor and Quantum. In about a year Quantum trademark will cease to exist and we'll face an industry giant producing 50 million HDDs a year. We can expect that Maxtor models would be used as base for ATA lines and Quantum models would be High-End because Maxtor just doesn't have any Atlas level solutions.

And we should note another "asian" merging: LG and Hitachi decided to unite their optical drive industries. Hitachi was always good with DVDs and LG is one of CD leaders. As a result we'll get a giantic combination on optical drives market - Hitachi-LG Data Storage, with most of quotes held by Hitachi. LG computer industries slowly dissolve into various countries LCD screens are produced with Philips, memory - with Hyundai, and now optical drives.

And the last. DataPlay, creator of 500 MBytes optical minidiscs found another powerful partner - Toshiba that will set industry for making and selling these discs all over Asia, excluding Korea. And Toshiba is right: DataPlay discs is a very interesting technology with a potential of filling many market niches, i.e as a data storage for next generation digital players. If it won't be stopped by 5 GBytes Fluorescent Multilayer Card from C3D. But both technologies still have to show themselves.

Write a comment below. No registration needed!