by Shui-Che Lim
Letter 1: from Adrian
In everything I've read I've only heard of D3D and the new instructions. Will a K6 3D and K6+3D be able to speed up Quake2 and other games that use OpenGL instead of D3D? Adrian.
K6-3D and K6+3D will have different levels of performance on OpenGL and applications like Quake 2.
OpenGL, like D3D is a 3D graphics API. In order for anyone's implementation of OpenGL to work with the 24 new FPU SIMD instructions in AMD's new K6-3D and K6+3D, the code must be rewritten to support these new opcodes.
That being said, recent announcements by Microsoft and SGI have made it clear that D3D and OpenGL will merge and become the new 3D standard supported by both Microsoft and SGI. This will probably occur in it's first iteration in 1999. Since Microsoft has already ported support for AMD-3D in DX6, there is no reason to expect that the eventual version of D3D/OpenGL will not support AMD-3D.
K6+3D will have higher performance than K6-3D at equivalent clock speeds simply because K6+3D will have 256K of L2 cache directly on chip that is running at the core clock speed of the CPU. Therefore, given equivalent pieces of code, whether it is D3D or OpenGL, it will run as much as 15%-20% faster on K6+3D.
Shui-Che Lim
Letter 2: from Dr. Graham R. Davis
It sounds like the AMD K63d will only run faster for software specially encoded for it, including Directx6, and the performance gain will be mainly in 3D rendering, but not in other floating point operations. However, there is a growing tendancy to offload the 3D triangle setup onto the 3D graphics card, which could take away AMD's advantage. I wonder how the K6 3D would compare with a Pentium II when used in conjuction with a Voodoo 2 3D accelerator! G.R. Davis
All that you've said is perfectly correct. AMD-3D must be specifically coded into applications and into API's like D3D in order to gain any floating point performance. For applications that do avail themselves of the technology in AMD-3D the performance gain could be as much as 4x the floating point performance of Intel's FPU.
Your second assertion that future 3D hardware will have triangle setup engines incorporated into the chipset design. Voodoo2 will lead the way with this trend and I suspect that within 18-24 months, FPU performance for any CPU will become a moot point.
Keep in mind, however, that the only point that Intel currently has in it's favor is the fact that it has better floating point performance as compared to AMD or Cyrix. Voodoo2 will do more than take away AMD's advantage with AMD-3D, it will level the playing field for all CPUs.
It has been shown time and again, that except for the FPU performance of AMD vs. Intel CPUs, the K6 is really quite competent at meeting or exceeding the performance of Pentium II in other areas, all at a lower price. What does this tell you about the popularity of these chips versus Intel chips once hardware triangle setup is a defacto standard?
Also, let's not count out AMD-3D just yet. More and more, companies are supporting two 3D standards... 3Dfx Voodoo and Direct3D. For users of Voodoo or Voodoo2, AMD-3D is a moot point, but for people unfortunate enough to have to rely on D3D because native support for their chipset was not provided, the D3D can have on average 2x to 2.5x the performance on an AMD K6-3D over an Intel Pentium II.
Shui-Che Lim
Letter 3: from Gordon Berg
Just wanted to let you know that I appreciate your efforts on the Combatsim site and find your work truly useful to the sim enthusiast. I also have a question. :)
I just read on Gamepen's Hardware Therapy (I write the Flight-Sim therapy myself, please have a look if you aren't familiar with the column) the following:
"Judging from the newsgroup chatter, many feel that Intel�s recent investment in 3Dfx may turn into a very cozy arrangement, with Intel perhaps getting the license to use Glide with their new 3D video chips and Voodoo 2 chips getting "secret" info on Intel Pentium 2 processors and enhancing the Voodoo 2 to run better on them. Read about the P2 enhancements here. This development may be bad news for Cyrix and AMD owners who were thinking of upgrading to Voodoo 2.
Gary Tarolli of 3Dfx has this to say about the whole mini-controversy:
'We specifically modified/designed Voodoo2 for P2 speculative IO. As it turns out, other processors benefit from this as well. But it's a big improvement over Voodoo1 on a P2. I can't tell you how we optimized it, that's a secret :-)' "
I was wondering if you had any comments or observations on this. Based on what you wrote, I was going to upgrade via the 100MHz board/AMD route, now I am finding myself reconsidering. How much weight should Tarolli's comment on other processors be given? Is it a valid point to go P2 only if one can afford it based on this new info?
Gordon Berg
Gordon,
First of all, I've been a big fan of your column for quite some time. I'm not too ashamed to admit that the only reason I go to GamePen every few days is to see if you've got the latest Flight Sim Therapy online yet. This one month wait between Therapies has got to go... I almost had to get therapy from Therapy withdrawal =0
Now on to your question, which BTW, gets harder to answer with each passing day. The reason is the divergence of the CPU architectures and by the end of 1998, system architectures that have resulted from Intel forcing the competition along alternate paths and therefore causing the underlying technology and also the market to fragment.
I suppose the fairest way to answer your question would be for you to take a serious accounting of yourself and decide into which camp you fall:
1) The "Money is No Object" camp 2) The "Performance/Value" camp
For people who are not willing to give up even a single frame of performance, I'd say that you're probably bound, at least for the next year, to the Intel roadmap (and price tag). But, if you're the type of person who can live with a better Value vs. Performance proposition, then you would look at alternate choices.
I am an avid flight simmer. I cut my teeth on Red Baron, the Aces Series and Falcon 3.0 and have never looked back since. However, I find that even if I buy the latest and most expensive that Intel has to offer, it will again be old news within 12 months and then I'm forced to upgrade again. I have looked at this, as I'm sure many flight sim enthusiasts have, and have come to the following conclusions:
1) The CPU is the SINGLE most expensive piece of hardware you can buy. Constantly relying on the CPU to ride the performance curve gets very expensive, very quickly.
2) The advent of hardware 3D like Voodoo2 makes FPU performance irrelevant. While Intel may lead the performance curve, AMD and Cyrix are not far behind. The value that they bring is tremendous since I may trade 5%-10% performance (which only comes down to a difference of a few frames) but pay for a system that could be as much as 1/3 less than a comparably equipped Intel based system.
3) The next generation of motherboards will diverge. Intel's roadmap shows 100MHz host bus for motherboards. AMD's K7 will utilize the Alpha EV6 protocol bus which will be capable of 200-300MHz host bus speeds. However, AMD's solution will probably still cost less than Intel while offering, for the first time, superior system performance.
Now, I don't have any numbers to show how much better the Voodoo2 Speculative I/O designed for Pentium II works for Intel vs. AMD/Cyrix. However, I suspect that the difference wouldn't be something like 20% or more, which is the type of additional performance I'd have to get to make purchasing an Intel solution worthwhile to me.
The long and the short of it is that it all comes down to personal choice, as well as the size of one's pocketbook. I stopped buying Intel about 2 years ago and have never turned back. Don't get me wrong, I can afford to buy Intel's best... but it's become a matter of getting the best value for the money that I spend.
Have I had any regrets? Only a few, but certainly they were not of the proportion to encourage me to go back to paying Intel's outrageously high prices. I have been, on the whole, very pleased with my non-Intel solutions. The money that I have saved has allowed me to increase my system memory, add a speedy 3D card, and still have a few extra dollars left over for the software which is truly what all this fussing over hardware is all about.
SIncerely, Shui-Che Lim
Letter 4: From Craig
Will my Intel 233MMX CPU work on the Super7 motherboard and if it does, will I see a performance increase in flight sims?
AMD's Super7 is a truly Open Architecture in that it does not discriminate against CPUs from other vendors. This is not true of "standards" that Intel tries to force on the market like Slot-1.
Super7 motherboards will continue to work at 66/75/83MHz just like the old Socket 7 motherboards now on the market. In addition, it will offer the newer 100MHz host bus speed. Your 233MMX CPU will continue to work on Super7 motherboards, however, since it was originally designed for 66MHz host bus speed, it is highly doubtful that it will run at 100MHz. This would require 50% tolerance which is far in excess of most semiconductor device design tolerances.
You will probably see some performance improvement from other areas, however. The newer Super7 chipsets will support more than 512K of onboard cache. If you select a motherboard with 1MB or even 2MB of cache, you will see incremental improvement in performance. You will also gain AGP for your Pentium 233MMX, which you would not have otherwise have gotten without an upgrade to Pentium II. As AGP cards come out and drivers mature, I'm sure it will eventually become a "must have" item.
Finally, Super7 will allow you to upgrade to a 100MHz host bus capable CPU such as AMD's K6-3D or K6+3D or Cyrix's Cayenne when you're ready. This would give you performance on a par with the best that Intel has to offer without spending the big bucks for a Pentium II system.
Sincerely, Shui-Che Lim
Email to Shui-Che..Shui-Che Lim