Quake 3 was not optimized for Athlon CPU?

[Brians256]

I first saw this on the Inquirer (http://www.theinquirer.net/?article=11451), which pointed me to AMDZone (http://www.amdzone.com/) so that I finally got to the actual website here (http://speedycpu.dyndns.org/opt/).

Apparently, he enabled SSE for the Quake 3 engine on Athlon chips. If I recall, this is an old hack, but it makes me question the benchmarketing seen around the web about how the P4 is tremendously faster than the Athlon for Quake 3. Well duh! If the DLLs are optimized for the Intel architecture but NOT optimized for the AMD architecture...

The oddest statement I saw asserted that John Carmack refused to allow AMD to demo Quake 3 with these patches to optimize for the Athlon CPU. That just doesn't make sense!

What the heck is going on here?

[pHaestus]

I'm sure Carmack will comment on this before too long in a .plan, but I seem to remember way back in the day that there were optimized dlls that were released by third parties. I think carmack was opposed to using architecture-specific dlls at all for running Q3 but I honestly don't remember why. It's been too long for me but you might wanna search the shacknews archived comments. I believe there was some discussion of this a few years ago when the dlls first were released.

[Brians256]

I'm sure that there was a reason that Carmack didn't want to include these patches. I just have a hard time understanding why AMD wouldn't make a stink about how their CPU really is just as fast as the P4 in all these game benchmarks.

After all, the Quake3 benches put P4's WAY over the top of an Athlon. This just explains why.

As for not including arch specific optimizations.... didn't Carmack put SSE in there?

[Brians256]

pHaestus found this link (http://www.shacknews.com/onearticle.x/3992/) on ShackNews (http://www.shacknews.com/) that helps explain some history.

If ID had just gone back and altered the code to allow SSE on AthlonXP chips.... scores would be much better.

Thanks for the websearch, pH.

[ChemMan]

Don't forget that when Q3 first came out the current Athlon, then based on the Thunderbird core, did not have support for SSE. AFAIK. So SSE was in there for the Pentiums, but just not enabled for the Athlon. Why this wasn't rectified when the Palamino core came out, I don't know.:confused:

[pHaestus]

Yea my memory was pretty close to right. Carmack doesn't want the game running on dlls because they are insecure. So you can't use these optimizations and play on Pure servers. But the general Quake VM isnt quite as optimized for any one specific PC as compiling it would be for that architecture.

It looks to me like id just didn't feel like going back and redoing the SSE detection from AMD=3dnow and Intel = SSE to AMD K6,Athlon, Tbird=3dnow, AMD XP, TBred, Barton and Intel=SSE


Can't say as I blame them as by the time SSE-enabled AMD chips came out then Quake3 performance wasnt a concern anyway (it's already running at 200fps+)

[Brians256]

Well then, if websites would just stop using Q3 to tout Intel superiority over AMD, I'd be happy.

You're right. 200+ fps really doesn't justify any effort to make things even faster.

[pHaestus]

Do they still use Q3? I thought it was more DVD encoding nowadays to show Intel giving AMD "the old in and out". Hammer seems to reverse things somewhat though with it coming out noticably better in games than Intel.

It seems like most games can always take advantage of a higher memory bandwidth. That's what the P4s have usually had over the AMD chips, and now that's what Hammers have on P4s.

[Joe]

All depends on the DVD app. I have not seen the same as you just stated real world compressions, since I have compressed more DVD's for home video watching than most a anyone I know ( hows close to 200 DVD's sound so far) all with TMPGEnc (Plus). AMD clock for clock is = or better than Intel I have found. a 1.7Ghz TbredB core was compressing at the same speed as my 2ghz P4 with same ram. My Dual 1.4 Ghz Tbird compressed close to 75% faster than my 2Ghz P4.

But I can do certain options to favor a P4 and others to favor a AMD.. it all depends how you want to compress and what bias you have. I think DVD compression benchmarks are bull shit overall since media source, and very very minor changes in the software can change compression time greatly.

[Bignuts]

The real question is whether I can use these Dll's with the other Quake 3 games, such as MoHAA

[joeyd]

3dnow is broken in all quake 3 based games. Which is said because the palmino (sp?) and throughbred(sp?) chips see a %-10% improvement and about the same from sse. So while i don't think the athlons will over take the p4s i think it would get alot closer to them.

[Brians256]

I hope that the next ID game(s) do not suffer from such an unequal treatment. This is particularly important since I own an AMD rig. :D

[joeyd]

Originally posted by Brians256
I hope that the next ID game(s) do not suffer from such an unequal treatment. This is particularly important since I own an AMD rig. :D Yes it does. He spent massive hours tweaking it for the nv30. So now it runs in less quality than the radeons. Which is bs in my book

[Brians256]

I saw that. Carmack optimized for the lesser quality pipeline of the nV GPU's. Perhaps it would not have looked any better with full 96-bit precision, but I would argue that it would have looked better if he'd relaxed his mental filter on the allowed visual effects. By requiring higher precision from the GPU, more effects could have been included in the base engine thus achieving higher realism.

Blah blah blah. I'm sure that Carmack doesn't give spit for my opinion. By choosing the lower quality route, he allows for more people to enjoy his game and he gets more money for his Ferrari collection. Nothing wrong with capitalism, but I wish he'd gone the high road this time.

[joeyd]

The problem is so many people will buy the nvidia card based on this game. Which as we see does not equal the best card for dx 9 games. This is a huge problem.