August 29th 2000

A Megahertz Too Far.

   The News just gets worse for Intel almost every day.  After a whole bunch of Intel Pentium III 1.13GHz processor samples sent out to web sites failed, Intel has recalled the CPU.  In the chip game, it looks like Intel went a MHz too far.  This new Intel debacle calls into question the reports earlier this week that the Pentium-4 would scale to 5 or 6 GHz.  If the PIII can't get above 1 GHz reliably, what makes people think the P-4 will go from 1.4 GHz to 6 GHz in it's lifetime???  It won't.

  But let's stay in the here and now.  This latest recall by Intel is more than an embarrassment.  It points to Intel's questionable marketing practices of announcing non-existent products, and pushing products out the door before they are ready.  It's time for Intel to clean up it's operation, and hold products in house until they are 100%.  It's also time for them to quit with the 'paper releases' of products that are not ready.  Intel is loosing consumer confidence in droves, and only a solid, reliable roll-out of upcoming product will reverse this trend.  If they pull another Caminogate with the P-4 debut this Fall, they will be in serious trouble.

                               Dr. John

NVidia Sues 3dfx.

   In a strange move for the current winner in the 3D graphics card wars, NVidia is suing 3dfx for alleged violations of hardware patents.  NVidia claims that technology in Voodoo3, Voodoo4 and Voodoo5 cards violates 5 different patents that NVidia owns.  You can check out what patents are in question over at The Register.

  It seems at least a little odd that NVidia is claiming now that 3dfx cards back to the Voodoo3 violate these patents.  Exactly why did they let this matter slip for almost 2 years before filling the law suit?  Hmmm.

  The patents seem quite basic to modern AGP bus systems, which makes me wonder why NVidia has singled out 3dfx for this suit.  In one particular patent granted in 1995, "Register array for utilizing burst mode transfer on local bus", it's clear that a very fundamental aspect of sending non-sequential addresses to a single destination utilizing "burst mode" is in question.  This kind of patent case seems Rambus-esque to me.  It's as though NVidia managed to patent technology that was at the heart of the bus interface for all modern video card's.  It will be interesting to see if this goes to court, or is settled beforehand.

                               Dr. John

Micron Sues Rambus!

   Finally the tides may have turned on Rambus, the intellectual property king of the memory business. Rambus has been bringing law suits against memory makers one at a time, trying to force all DRAM makers to pay Rambus royalties for memory they neither designed or produced.  This is because Rambus was able to obtain patents on memory technology they did not design, and they did so by questionable means. 

  Now, for the first time, an American memory company has stepped forward and filed a law suit against Rambus' patent claims.  Micron filed the suit yesterday in the US District Court in Delaware, which is where Rambus incorporated their business for tax purposes.  Now the fun starts.  We will probably not hear the details of the law suit because the case is pending.  But the basic thrust will be to make unenforceable the Rambus patents on SDRAM, DDR DRAM, and memory controllers, based upon the questionable methods used by Rambus to obtain the patents while participating in standards committee meetings.

I loved this quote from Van Smith's report about Rambus on Tom's Hardware last week:

  Recently a respected industry insider stated to us poignantly that RDRAM is "a memory solution with very limited application because it solves the wrong DRAM performance problem using fragile, expensive, and unwieldy technology that exacerbates the real problem. Thankfully its joint effort with Intel to drive the third generation of its niche technology into the PC space using slick and misleading marketing and the prestige of Intel has been exposed and is faltering under the harsh light of day. Its subsequent attempt to financially attack open industry standards using patents obtained under questionable circumstances seems to underline their failure to establish DRDRAM [aka RDRAM] as a standard and that the company seems to have chosen a kind of apocalyptic showdown with the DRAM industry as its end game strategy."

  Right on Van!

                               Dr. John

August 26th 2000

NVidia Detonator-3 Problems.

   The latest drivers for NVidia-based graphics cards (Detonator-3 ver. 6.18) were released earlier this month, and many people have since reported having trouble with them.  An article over at Tom's Hardware discusses the issue.  Apparently, TNT-1 and TNT-2 owners have experienced a performance hit, while some GeForce-2 owners experienced stability problems. So if you haven't installed them yet, it may be better to wait for the next release.  We are still having generally good luck with the 5.30 and 5.32 driver sets (Detonator-2).

                               Dr. John

August 25th 2000

How Fast, Pentium-4?

   Rumors are emerging from the Intel developers forum that the Pentium-4 will be substantially faster than a comparable PIII.  But when pressed for numbers, it was revealed that a 1.4GHz P-4 was 1.5 times faster than a 1 GHz PIII.  That suggests a 7% performance boost over the older chip.  Longer pipelines mean more latency, even with Intel's new "Netburst Technology". Oh my.

   The P4 is also supposed to have oodles of 'headroom'.  This means that it can be scaled to very high MHz levels. I've heard talk about the P4 scaling to 5 or 6GHz over 18 months.   But in my humble opinion, it will be a cold day in Hell before the P4 makes it to 5GHz.

   My guess is, (based on past performance improvements in previous chips in the line), that the original P4 die will max out in the 2GHz to 2.5GHz range.  A die shrink could then boost that limit to maybe 3GHz (but I doubt it).  Remember, the Pentium started at 90MHz (with the 60MHz bus) and went to 233MHz, the PII started at 233 and went to 400, the Katmai PIII started at 450 and went to 600, and the Coppermine PIII started at 500 and went to 1.1GHz (sort of).  So Intel's chips at best only double in MHz ratings before they fade away.  Anyway, aren't we going to be talking about the next chip in 18 months?

                               Dr. John

August 21st 2000

Did Memory Makers Forget to Patent SDRAM?

   For those of you following the Rambus wars, the latest news installment is intriguing. An article by Jack Robertson over at EBN discusses the legal fight between Rambus and the memory maker Infineon.

   Now remember, the scenario started with all the memory makers, including Rambus, holding hands and singing "We are the World" at JEDEC standards meetings throughout the early 1990's. The love fest started to break down when other JEDEC standards committee members began to suspect that the Rambus delegates were acting as corporate moles, deep within the standards body.  Eventually, Rambus quit the JEDEC committee, stating that JEDEC's role in the memory industry was not consistent with Rambus' corporate goals.  

   Now we jump ahead to the year 1999, when Rambus filed amended patent claims, not on Rambus memory, but on SDRAM and DDR DRAM devices and their controllers.  The rumor in the industry then was that Rambus was filing patents on technology that was already considered "prior art" in legal jargon.  In fact, even at this relatively early stage, memory makers were pointing fingers at Rambus and suggesting that they had used their position in JEDEC inappropriately.  It has since been suggested that Rambus used JEDEC simply for gathering data on their upcoming suite of patent claims.

  Now we get to  late 1999, and the year 2000, when Rambus is granted patents on SDRAM, DDR DRAM and logic controllers. They then decide the time is ripe to cash in on their newly awarded patents. They start filing legal actions against several smaller memory makers. At first, memory makers like Hitachi decided that they would fight, and they prepared legal documents that outlined their legal position.  But for undisclosed reasons, all memory makers that were initially sued by Rambus decided to quickly give in, and began paying Rambus royalties for products that Rambus neither designed, nor manufactured.

  By mid-2000, even Intel, who is Rambus' biggest partner in crime, decided that this was going too far.  They started working with the remaining memory makers that had not licensed their old technology through Rambus, to try to figure out a way to relieve Rambus of their ill-gotten Patent gains.  As soon as that story broke (at EBN), Intel got in such hot water with Rambus, that they had to quit the talks.

  But now, other memory makers are apparently joining Infineon in the fight against Rambus' patent claims.  Even without Intel, the remaining memory makers form a formidable group.  Their lawyers are currently pouring over the legal documents and patent claims to see what approach might be best to tackle the problem.  But the fact that the previous companies gave in so quickly, and the fact that the remaining companies seem uncertain as how to proceed, begs a very simple question.  Did the memory makers forget to patent today's standard memory architecture?  If they did, assuming that JEDEC was making such patents un-needed, and unenforceable, then maybe Rambus' lawyers saw the glaring omission, and decided to go for the jugular.

  If there is any merit to this speculation, then JEDEC is at least partly at fault here.  They set up a standards committee to avoid legal disputes, and ensure a standard architecture for modern computer memory.  It was up to them, in my opinion, to make sure that anything worked out at their meetings would be considered "prior art" by the time it was implemented in the market place.  Or perhaps JEDEC should have filed the patent claims themselves, and then just given the technology away to all it's members.  But letting Rambus subvert the process, and letting the situation get so out of hand that Rambus claims it owns patents on all current forms of computer memory is inexcusable.  That's the opposite of what standards committees are constituted for. 

                              Dr. John

August 17th 2000

Allgames Network Gone.

   Allgames Network, including AGN3D, AGN hardware and AGN News have ceased operations, and the web sites are gone.  This is very depressing news.  They were good sites, started by the talented and energetic Jeremy Allford.  We will miss the sites, and Jeremy's reviews.

                              Dr. John

August 14th 2000

Rambus Sues Infineon.

   Infineon is the memory spin-off of Siemens, and is one of the larger memory manufacturers in the world.  It was reported in EBN last week that Infineon is being sued by Rambus over alleged violations of Rambus' sweeping memory patent claims.  So far, the 3 memory manufacturers that were sued by Rambus have settled out of court, by agreeing to pay Rambus royalty fees for all SDRAM, DDR DRAM and synchronous memory controllers that they manufacture.  

   The "pick 'em off one at a time" strategy has worked brilliantly for Rambus so far.  They declare that "negotiations have broken down", with a single memory maker, and then file a legal action against them.  The memory maker then gets it's hackles up and threatens counter-suit.  In all 3 cases so far, the memory maker then quickly surrendered out of court, and agreed to pay.  Infineon is a pretty big player, so the result of this Rambus suit is going to be pivotal.  If Infineon caves, Rambus may be on Easy Street.

  This is also a defining time for Rambus.  The announcement by Intel that they might pursue memory interfaces other than Rambus for the upcoming Pentium 4 processor may have put Rambus' legal juggernaut into high gear.  They don't have that guaranteed position as the only Intel memory interface for the P4 any longer.  So they need to get tough with the remaining companies that they believe are violating their alleged patents.  If they can make it stick, then they will get royalty payments for any type of memory that Intel uses with the P4.

  Maybe Rambus never expected Rambus DRAM to become the dominant memory type.  Perhaps it was their plan all along to use the JEDEC meetings to submit ad hoc patent amendments which would give them sweeping patent rights over all memory and memory controllers in current PCs.  While memory makers were fighting against the Rambus DRAM Straw Man, Rambus would be able to pick off the memory makers one at a time in court.  Then, whether Rambus DRAM becomes the memory standard or not, Rambus still gets paid for sitting on their butts.  Don't you love conspiracy theories?

                              Dr. John

Intel Price Drop Coming.

   In response to AMD's move to lower CPU prices, Intel may follow by moving their scheduled price cut forward by a couple weeks.  So far, the new prices have not shown up at distributors.  Intel is in a tricky situation.  They don't want to give up their huge profit margins, but they don't want to appear to be the "high-priced alternative" either.

NVidia: New Card, New Drivers.

   In an unbelievably short period of time after they released their GeForce-2 GTS cards, and almost concurrent with their release of the new low-cost GeForce-2MX cards, NVidia has announced the release of the GeForce-2 Ultra.  This one will give the still-missing Voodoo5 6000 a real run for it's money.  And again, NVidia looks like they will have it to market before the competition.  3dfx has some real catch-up to do.  The Voodoo5 4500 does not compare with the GeForce-2MX, the Voodoo5 5500 doesn't compare with the GeForce-2 GTS, and the V5 6000, when it debuts, may not be much faster than the GeForce-2 Ultra, despite it's substantially higher estimated price.

  To go along with their new card, NVidia has also released the new Detonator-3 drivers (version 6.18).  These add improved performance, and better full scene antialiasing (FSAA).  You can get them here.  NVidia seems poised to take over the 3D world.

                              Dr. John

August 11th 2000

AMD Price Drop Coming.

   AMD will lower prices on it's Athlon and Duron processors on Monday August 14th.  This price drop looks to be fairly large, and may bring the 1GHz Athlons down to near $500!!! If AMD really cuts the prices that far, Intel may be in for a real rough ride! 

Pentium 4 May Be Delayed.

   We knew this was coming.  Intel has not announced when the Pentium 4 would be released, but industry watchers had been led to believe it would be September.  Now the rumors have started to float that there will be as much as a 2 month delay (November?).  If true, Intel may not have a very good holiday sales season.  If the Pentium 4 does not debut until November, there will not be enough of them in the supply chain for the holiday season.  They can fall back on the newer 0.13 micron Pentium IIIs, but with the AMD Thunderbirds up to a GHz dropping so much in price, it may be a hard sell.  If the 0.13 micron Pentium IIIs overclock even better than the 0.18 micron ones, Intel may still be in the game, but only time will tell.

  If you are interested in a hypothetical Pentium 4 roadmap, check this article at The Register.

                              Dr. John

August 8th 2000

VIA's HDIT Chipset. 

   VIA announced yesterday that it was working on a new DDR chipset that could be tailored for either Intel's or AMD's line of processors in the year 2001.  The so-called "High-Bandwidth Differential Interconnect" (HDIT) architecture will support DDR memory up to 266MHz. Changes include improvements to the Northbridge and Southbridge chips, and the addition of PCI-X companion chips, which will boost PCI speeds up from 33MHz to 66MHz.  The design also includes memory buffers, dual ATA/100 controllers, and an improved audio-communications riser connection.  VIA expects to have new P-4 and Athlon chipsets based on the HDIT architecture ready early next year.

                              Dr. John

Voodoo5 5500 Love-in. 

   NVidia fans, avert your eyes.  Another pro-V5 5500 review is out, and it couldn't be more glowing.  I tend to agree with the points about image quality. However, the antialiased image with the V5 is not that much better than the competition's, and I don't believe that most people would be able to see a difference while playing a fast paced game.  Nonetheless, if you are interested, check out the review.  Just for the record, my Gladiac review is almost done, and it's getting a 5 out of 5 smiley faces.  The V5 5500 got a 4.4 out of 5 smiley faces in my review of it last month.

                              Dr. John

August 5th 2000

S3... Graphics R.I.P. 

   Don't you just love it when big companies like S3 buy smaller companies like Diamond Multimedia, and then grind them into the ground before shutting them down?  3Com did it to US Robotics, and there are numerous other examples in the computer industry over the last few years.  Well, S3 has just announced that they will shut down the graphics card division of Diamond Multimedia.

  I'm not sure what these folks are talking about behind closed doors when they do this kind of stuff, but the disturbing trend seems to be "Destroy them... or buy them, and then destroy them."  Competition is the key to keeping prices down, so every time a graphics card maker or modem maker goes out of business, it means one less competitor in that market.

  S3 was never a very good graphics chip company, but they sure knew how to leverage a market.  Diamond Multimedia did make good video cards, until S3 bought them.  Well, now S3 and Diamond are completely out of the graphics chip/card business, and will focus on MP3 and networking stuff.  The end result is less competition for the few remaining graphics chip makers.  

   The whole thing may not bode well for 3dfx, who made a similar move last year.  In both of these cases, a graphics chip maker bought a graphics card maker, in order to become the sole supplier of their brand of graphics card.  These decisions were made before this year's spate of component shortages, which have left companies from Intel to 3dfx without sufficient parts to produce their products.

  But companies who just produce graphics chips, like NVidia, do not have to worry about parts shortages.  That's a problem for the companies that buy NVidia chips, like Creative and Asus.  However, since there are many companies that use NVidia chips for their graphics cards, that assures that some of them will be able to get the necessary parts and produce the boards.  So by getting a better chip to market faster than 3dfx, and by sticking to the "old business model' for the graphics card industry, NVidia has positioned itself to take over the graphics card market.

                              Dr. John

August 4th 2000

Intel to Increase Bus Speeds? 

   A rather cryptic article by Mike Magee over at The Register talks about Intel's plans to move the Celeron to 100MHz, and the PIII first completely to 133MHz, and then to 200MHz.  The goal... crush AMD.  This "information" from some deep mole in the far East, (obtained by a good rubber hosing and gallons of beer), bears some scrutiny.  

  We, and many others, have speculated for over a year that the Celeron would move to 100MHz, and that the PIII would move to 133MHz.  Well, as you know all too well, the "Coppermine-core" Celeron-2 is still at 66MHz, and it's performance suffers greatly from the sad fact.  And while some of the PIII line has moved to a 133MHz front side bus, all this has done is limit it's overclockability.  The 100MHz models are less expensive per MHz, and overclock much better.  So the plain fact is, Intel is still moving at a snail's pace when compared with AMD, and much of this is just talk.  Intel also seems to still be stuck in Rambus-waiting-mode, hoping the price of RIMMS will drop to within 5% of SDRAM (holding your breath anyone?).

  Now let's move to the next tidbit of info.  Intel will be moving the PIII to 200MHz on the bus frequency, to keep pace with the AMD EV6 bus?  Well, unless something has happened in hardware land that I have not heard of, only double data rate DRAM is capable of going to 200MHz, or beyond.  If we were talking SDRAM, the top speed would be 166MHz.  So if this rumor is true, it's just re-stating the fact that Intel is looking into a DDR memory interface for the Pentium 4.  They are way behind VIA and AMD on integrating a DDR Northbridge into any chipset design.  So my take on all of this is... we've heard it before, now show us.  By the time Intel is testing it's first 200MHz DDR chipset, AMD will be rolling out a 266MHz DDR chipset.  Intel's got lots of catch up to do, and talking about it won't cut the mustard.

                              Dr. John

UUNet Outage Hits Again. 

   Our web site was down for much of Tuesday and Wednesday due to another UUNet service interruption.  So far, I can't say their service has been reliable.  If this happens again, it's probably time to look for another connection provider.

August 1st 2000

Rambus Spinmeisters... Ho!. 

   As Rambus' stock price slips further downward, the chief corporate spinmeister, Avo Kanadjian,  has come out swinging.  An SBN article details the recent Rambus announcement that they have perfected 1.066MHz (1GHz) Direct Rambus DRAM.  According to the article, Kanadjian says that new testing equipment will allow 1GHz parts to be selected from current runs of Rambus chips. 

  Well, it's obvious that hanging with Intel has really rubbed off on Rambus.  This tactic is straight out of Intel's "How to Fend off AMD" manual.  When you start to loose the engineering war, send in the 'ol spinmeisters. Remember Intel's announcement of 1GHz processors?  Well, those parts are still about as plentiful as flying elephants, months after their "release".  But Rambus liked this idea so much, that they decided to try it out themselves.  As people who have followed Rambus' roller coaster ride know well, Rambus is having trouble getting decent yields of 800MHz.  So exactly how many of these 1GHz parts will they be able to make with the same fabrication process?

  But it gets funnier.  If you recall, many memory makers got mad at Rambus early on, because retooling their fab plants for Rambus production was a major, expensive undertaking.  All new equipment, including new testing equipment, was needed.  Now that they have done that, Rambus is telling them that they will need to go out and get even newer, faster (533MHz) testing equipment, which you can read about here.  So the price of production goes up again!  That's going to make reducing the cost of Rambus harder still.

  But there's more!  It turns out that to get this faster Rambus to work, you need to solder the chips directly onto the motherboard!!  Oh, I get it, with 1GHz Rambus DRAM, I pay much more money for a completely non-upgradeable, high-end computer system.  Can't wait.

  Well, Kanadjian did his job.  He announced a non-existent product in an attempt to slow the decline of the stock price.  You'll notice that no additional engineering was required for this "new" product.  That's because Rambus no longer appears to have any engineers in their employment, only a bunch of lawyers and spinmeisters.

                              Dr. John

Copyright 2000, KickAss Gear