Introduction
The first day of MPC Forum 1997 consisted of two different seminars which you’ve got to choose from. Whilst Larry was attending Michael Slater‘s ‘Microprocessor for PCs: A Critical Look at Technologies and Future Directions‘, I went to ‘3D Graphic and Multimedia: Chips and Choices‘ by Peter N. Glaskowsky, satisfying my new strong interest into this exiting subject.
3D Graphic and Multimedia: Chips and Choices
After the introduction, which didn’t really supply us with any new insights, Peter was going into 3D processing detail. As you all know, currently the FPU of the used microprocessor is of paramount importance for the 3D performance of a system, even if 3D accelerators are used. This is shown fairly extremely if you benchmark systems with 3D Winbench 97. Now Peter’s opinion as well as mine is to move the FPU intensive geometry calculations from the main CPU to the graphic processor on the 3D accelerator. Peter says on slide 31 “Fundamentally, however, the CPU is the wrong place to do geometry processing.” Now you can imagine that this is not exactly in the interest of Intel. The strong FPU of Intel CPUs is one reason why Intel CPUs are the best ones to use for 3D applications. If the FPU is losing its importance in 3D geometry processing, Intel’s competitors would look much better in 3D applications than they do now. One of the keys to the enabling of 3D processors doing the work of geometry processing instead of the CPU and hence supplying the FPU power needed for this topic is the 3D API that’s used, since it has to allow this kind of ‘work load sharing’. The most important 3D API today is Direct3D and this will most likely stay this way. Microsoft is currently deciding if they want allow the shift of the geometry processing from the CPU to the 3D chip whilst planning their next version of DirectX, revision 6.0. Obviously the decision of Microsoft will take place tomorrow, October 14th, and so far the favors seem to lie at not supporting 3D chips processing geometry instead of the CPU. This may be based on Microsoft’s close relationship to Intel, at least it doesn’t seem too hard to guess. I guess that most of you would prefer competitors of Intel to be able performing well in 3D applications as well, particularly as it seems possible. Peter Glaskowsky’s words were “if you’ve got any influence on Microsoft’s decisions, please tell them to implement the 3D chip’s geometry processing support into Direct3D 6.0. The CPU shouldn’t have to do this calculations.” I think I’m doing what I can to influence this decision as good as I can by asking you to let Microsoft know what you think. If you want to give non-Intel CPUs, that might have a weaker FPU, a chance in the future, then let Microsoft know that you want them to implement the geometry processing support of 3D chips.
Peter also shares my concerns about the insufficiency of the AGP architecture. He demands that textures should be processes from the local graphic memory, since the bandwidth of local graphic memory is still much higher than what AGP offers. It’s actually even worse than what you could read in my ‘AGP – The Theory‘, section ‘Critical Thoughts‘. Even with x4 mode (probably not available before 1999) there’s one peculiar problem. Whilst the AGP device can read from main memory with 512 MB/s (x2 mode) or even 1 GB/s (x4 mode), the CPU can only write to the AGP device via a 66 MHz PCI like technique, offering not more than 264 MB/s bandwidth. If the x2 or x4 modes shall be used, the CPU would have to write the e.g. geometry data first to main memory, let the AGP device know that it’s there and then the AGP device has to read it from main memory via the x2 or x4 mode. That is obviously kind of crazy and jeopardizes the whole idea of the fast x2 or x4 modes.
Evaluating 3D chips, Peter Glaskowsky was covering the several products. I will go into more detail about his evaluation of the chips as soon as the Microprocessor Forum is over, but I’d like to mention a few of his basic opinions. He didn’t include Matrox into his graphic chip overview, because he also doesn’t think that Matrox has got any real 3D accelerator available right now. The add-on card for the Millennium II which will released in the future is actually using the Power VR PCX2 chip, not the 3Dfx Voodoo as some of you seem to think. This chip is manufactured by NEC, where the Matrox chips are manufactured as well and this is probably the reason why they are using it … they’ll probably get a very attractive price. So far Power VR cards (Intergraph Apocalypse 3D) haven’t been too convincing. Peter Glaskowsky is currently favoring the 3DLabs Permedia 2 cards, probably because he also thinks that the NVidia Riva 128 chip is unfortunately supporting too little video memory (only 4 MB as you know). Unfortunately neither Matrox’ upcoming ‘Twister’ nor 3Dfx’ upcoming Voodoo 2 chipset were mentioned at all.
Microsoft’s ‘Talisman’ high end 3D chip architecture model was also covered and I will supply you with more information about this interesting project after MPF as well.
The rest of the afternoon was dedicated to multimedia, particularly sound stuff. So far this is not really one of my highest interests and I hope you’ll forgive me that I won’t get into this by any big detail.
A Critical Look at Technologies and Future Directions
by Larry Barber
Today was the first day of Michael Slater’s Micro Processor Forum. Michael did his usual superb job and right to the point. A some what negative point when it came some of our industries patriarchs. Michael’s view of the Power PC is clear — May it rest in peace, right next to (Michael’s words) “the sinking ship” formally called Apple. Maybe we should request a burial at sea? On the Alpha front. Due to less than capable marketing (my thoughts) DEC’s Alpha has been relegated to a niche player although it arguable is the best performing chip out there.
Netware? Yes remember that company called Novel Michael called it “fading” he was being polite as it is more like chronically anemic.
Michael’s thoughts for those of us waiting to buy that notebook are to take heed. The performance gap between notebooks and desktops is turning into a chasm. If you are waiting for the gap to narrow don’t — it is only going to get bigger. The problem is that new CPU’s are power hogs and it is getting harder and harder to get that power down to where notebooks need it Under 8 watts. A typical Pentium II 300 is 35 to 42 watts. The primary tool designers have to get the power down is lowering the voltage, however there is less and less room to get the voltage lower. Obviously you can not go much lower than 1 volt and some systems are getting close at 1.8 volts. Heat and power are show stoppers. As frequency of the CPU escalates so does the power required and heat produced.
Remember that thing called Network computer or NC. Well, today Michael announced it’s demise and gave out get well cards for the terminally ill Network PC. Truthfully, I think it is a caterpillar that is just waiting to be reborn as a Guess what PC. What a novel idea!
A few other of Michael’s observations all of which I happen to agree with are:
1. The 30 odd graphic chip companies we have to day will be ten in three years. Intel graphic products will have a major impact in taking that number to ten. PS Dump your stock!
2. Intel prices on Pentium II are going to crash in next few weeks. Do not be surprised if we see prices that are half of today’s numbers. I can not believe that will not take a hit in the stock market to their stock price.
3. Industry wide ASPs (average sales prices) are going down.
4. The computer industry is utilizing almost 100 million CPU’s per year and Intel is shipping 90% of them. It is not going to change in the foreseeable future. I really don’t believe that anyone other than maybe Jerry Sanders of AMD is going to bet against that.
More in a couple of days, particularly Micheal’s view of the future of Socket 7 …
Please follow-up by reading Microprocessor Forum 1997 – Second Day