3Dfx and Voodoo3
Yesterday, Wednesday, January 27, 1999, 3Dfx held a press conference on top of the Zugspitze, the highest Mountain in Germany. The conference was held by Greg Ballard, CEO of 3Dfx, Brian Bruning and Nick Pandher. We heard the latest news about Voodoo3 and what will happen in the 3D-card business now that 3Dfx acquired STB.
Voodoo3
- Voodoo3-3000 silicon in final and running fine at the promised 183/183 MHz core/mem-clock. Supposedly 3Dfx managed to clock it even higher already. Voodoo3-3000 will be the high-end Voodoo3, and 3Dfx is saying that it will easily be the fastest 3D-chip available.
- Voodoo3-2000 is also on track. This version will be the same as Voodoo3-3000, but only running at 125/125 MHz core/mem-clock. The 2000-version is targeted to the lower-cost OEM-market and also supposed to go directly on motherboards. All features will be identical to the 3000-version, only the clock is less. This includes a 128-bit memory interface.
- As you already know, the Voodoo3 will not be able to do 32-bit rendering. 3Dfx thinks that this feature is impacting performance too much, so that they didn’t even bother including it into the silicon. Many other journalists agree with me when I say that it would have been nice to leave the decision to the consumer. 32-bit rendering could have been an option, which we could have turned off by ourselves.
- Voodoo3 will be able to offer digital output for flat panels. It will need a special 3Dfx-chip for that, called ‘LCDfx” (I may be wrong with that name, sorry). Normal Voodoo3-cards will not include this chip, so you will require special boards that offer digital out.
- Voodoo3 will support AGP, but it will not support AGP-texturing. This is an important thing to consider! AGP-support as we know it from ATI, Matrox, NVIDIA or Intel does always support AGP-texturing as well. For many of us, AGP-texturing is what we all used to talk about when talking about AGP in general. Voodoo3 will still not be able to texture directly from system memory, thus large textures will only be displayable as long as they fit into the 16 MB onboard memory of the Voodoo3. However, don’t forget that Voodoo3 will be using 3Dfx’s own texture compression, which should make sure that pretty large textures will be supported by this chip.
- At the end of June there will also be a Voodoo3-4000. This version of the Voodoo3 was expressively ‘de-hyped’ by Greg Ballard yesterday. Unlike what some publication say about the ‘4000’, it will be nothing but a Voodoo3 that will include support for 4x AGP. Intel’s 440JX (‘Camino’) chipset will launch by the end of June and 3Dfx wants to have a 3D-chip ready that can support this new feature. Voodoo3-4000 will NOT REQUIRE 4 xAGP, as Greg Ballard pointed out yesterday.
- 3Dfx targets the release of Voodoo3 into April 1999, but things seem to be running very well, so that Greg Ballard mentioned the possibility of a release date in March already.
STB and the Graphics Card Market
- STB will get a name change, so that graphic cards produced by STB will be called ‘3Dfx-cards’ in the future.
- STB will exclusively produce Voodoo3-cards in the US. Neither Creative nor Diamond will receive any Voodoo3-chips. Greg Ballard called the former relationship with those two “former lovers who broke up”. He particularly pointed out that 3Dfx does not feel safe doing business with Diamond anymore, since Diamond lost $50,000,000 in the last quarter. Greg Ballard expects that Diamond could possibly pull out of the graphics market completely very soon.
- The STB product names like e.g. ‘Velocity’ and the STB brand name itself will possibly still be used for the OEM-business.
- In Europe 3Dfx is currently negotiating with Elsa and Guillemot. STB did not use to have a strong name in Europe, so that 3Dfx needs other strong OEMs in this area of the world. The future plan of 3Dfx however is to build up a strong brand name recognition for ‘3Dfx graphics cards’ in Europe soon too, which could be a problem negotiating with the two above mentioned OEMs.
- The motherboard makers that will use Voodoo3-2000 onboard will mainly be Taiwanese motherboard makers who 3Dfx has a good relationship to.
- You may remember that Quantum3D is something like a spin-off from 3Dfx. Ross Smith (founder of Quantum3D) used to be one of the founders of 3Dfx. Quantum3D will with 3Dfx’s help focus on the Arcade market and probably only provide some super-high-end 3Dfx based cards for the normal PC. This way Quantum3D does not have to target the main stream 3D-card market anymore and can fully concentrate onto the Arcade market. 3Dfx doesn’t have to see Quantum3D as a competitor of the likes of Diamond or Creative anymore, so that they can work a lot more closely with Quantum3D as a provider of super-high-end hardware with 3Dfx chips.
You can see that Voodoo3 is pretty close. I hope to be able providing you with some test results of Voodoo3 as early as possible. The STB-deal of 3Dfx will change the graphics market considerably. Diamond is suffering badly and Creative is certainly hurting too. We will see what the next steps of NVIDIA will be, now since they had a successful IPO.
Intel’s New CPUs for Mobile PCs
On Tuesday, January 26, 1999, Intel held a large press conference in Munich, introducing their new mobile CPUs. It is interesting to see that in the mobile area, the days of the Pentium II processor with external cache are already over now. Here the facts in short:
- Intel introduced the new mobile Pentium II 266PE, 300PE, 333 and 366. Those new mobile CPUs will come with 256kB on-die L2-cache running at core clock, as opposed to the former mobile Pentium II 233, 266 and 300 with 512 kB external L2-cache running at half the core clock. Intel claims that the new mobile processors will be up to 10% faster than the old mobile processors at the same clock speed.
- At the same time there will be mobile Celeron 266 and 300 processors. The Celeron will be almost exactly the same as the new mobile Pentium II processors, the only difference is the smaller on-die L2-cahce of 128 kB.
- Those new mobile Celeron and Pentium II processors will require less power than its predecessors with external L2-cache.
- The processors will come as BGA version for cost-effective notebooks in which the processor will be soldered onto the board and thus not exchangeable, as 280-pin Mobile Module (MMC1) as already known from the former mobile Pentium II processors, as 400-pin Mobile Module (MMC2) and as 240-pin Mini-Cartridge. This means that people with PII-notebooks can possibly exchange their CPUs with the new mobile Pentium II in case the notebook is using MMC1.
I only want to say a few words to this. Firstly I was very surprised about Intel’s claim that the new mobile PII would be significantly faster than the old one. So far we haven’t seen that much of a difference between L2-cache at half the core-speed vs. L2-cache at full core-speed (desktop PII vs. Celeron or Xeon). Intel’s claim that the Celeron will be slower than the new PII for mobile is also difficult to believe, since I doubt that the difference in L2-cache size would make that much of a difference (again desktop PII vs. Celeron vs. Xeon). I will test all this as soon as I can, to back up my doubts.
The second thing that’s close to upsetting me is the fact that Intel introduces a PII with 256 kB on-die L2-cache for mobile systems just like that. Doesn’t that clearly show that Intel could offer that for the desktop too? If Intel’s claims are true and the Pentium II with 256 kB on-die L2-cache is indeed faster than it’s older brother with 512 kB external L2-cache, wouldn’t it be nice if we could take advantage of this on our desktops too? Isn’t it obvious that those new Pentium II CPUs could just as well fit into a Socket370?
Intel keeps this new technology from is in the desktop area and we have to stick to the old and obviously obsolete big Pentium II CPUs for Slot1 with external L2-cache. The same is even valid for Pentium III, it will also come with external L2-cache.
Intel proved another time how much they like playing games with us. Intel is obviously ready to ship CPUs with 256 kB of on-die L2-cache, but strategically they only offer this to the mobile market. Celeron for mobile, which could obviously run up to 366 MHz as well, will only ship in versions up to 300 MHz, so that the high-end users won’t even get to the idea of buying a notebook based on a Celeron 366, although it will probably perform close to identical to the mobile PII 366. The desktop market will carry on with Slot1 and external L2-cache, although we just learned from Intel that internal 256 KB L2-cache would be faster. All those new CPUs would fit into Socket370, but this is another thing that Intel doesn’t want to happen.
I really wonder, how stupid does Intel think we are?