Anyone noticed that the athlon xp's performance in new games(and probably other apps as well) seemed to take a nosedive as soon as amd basically abandoned the platform? For a long time the athlon xps were neck and neck with the p4s, with the 3200+ at least around the speed of a 2.8ghz or 3ghz p4. But in recent games a it seems like a 3.2ghz p4 can easily have a 50% advantage or more over a 3200+ athlon xp. Did companies stop optimizing/recompiling their code for athlon xps once the 64s came out? Did they all upgrade to amd's latest compiler, which optimizes for athlon 64s and not athlon xps? Am I just seeing a conspiracy where there is none?
Why the sudden drop in athlon xp performance?
- Thread starter Fox5
- Start date
i think it's more a case of modern games/engines pushing things one step further. the athlonXP's "model numbers" are derived from performance in certain programs. those programs are old(er) now, and newer technology is starting to show the weakness of the aging athlon (k7) core.
That's not exactly what I'm seeing. For one, the faster Athlon XP's really were a bit overrated. And if you look at HL2, which I think is a fairly new gameFox5 said:
a XP3200+ competes just fine with a P4 2.8Ghz, http://www.firingsquad.com/hardware/half_life_2_cpu_shootout/page5.asp which is about what I'd have expected.It's possible the AXP doesn't compare too well on some new apps, but that is the case with some old apps/games too (Quake3 for example). Maybe you'll also see typically larger differences with new apps since they'll obviously make more use of advanced features like SSE2 (which really helps the P4 for FPU-intensive apps).
mczak said:That's not exactly what I'm seeing. For one, the faster Athlon XP's really were a bit overrated. And if you look at HL2, which I think is a fairly new gameFox5 said:
It's possible the AXP doesn't compare too well on some new apps, but that is the case with some old apps/games too (Quake3 for example). Maybe you'll also see typically larger differences with new apps since they'll obviously make more use of advanced features like SSE2 (which really helps the P4 for FPU-intensive apps).
Quake 3 sort of supports my point. A little while back someone released a quake 3 exe or something that was recompiled in an athlon xp optimized compiler which put it on par with the p4's performance using the default code, which is believe to already be Intel optimized. And for some contrast I believe the unreal tournament games are more amd optimized, and the athlon xps often perform as well as the p4s or better.(and I think in serious sam the athlon xps beat the p4s by quite a bit, but that's a really old engine now) It seems like whichever cpu a company chooses to support better and optimize for is the cpu that wins the benchmarks.
Accord1999
Newcomer
The same recompiled Q3 DLL worked on a P4 and gave a comparable speed increase.Fox5 said:
The Quake3 DLL's are just compiled game code, not an engine/exe recompile. Normally the game uses an interpreter for the game code, and people have been compiling them to dll's to use for faster speed for years.
What I think is a bigger issue about P4 VS AMD performance in Q3 is that it only uses SSE if the CPU identifies itself as an Intel chip. If you use an AthlonXP the game shows the CPU as "AMD /w 3DNow!" but no mention of SSE anywhere.
Once id releases the full Q3 engine source later this year(they plan to anyway) I'm sure someone will do a simple CPU ID fix and let Athlon chips use the SSE code.
What I think is a bigger issue about P4 VS AMD performance in Q3 is that it only uses SSE if the CPU identifies itself as an Intel chip. If you use an AthlonXP the game shows the CPU as "AMD /w 3DNow!" but no mention of SSE anywhere.
Once id releases the full Q3 engine source later this year(they plan to anyway) I'm sure someone will do a simple CPU ID fix and let Athlon chips use the SSE code.
Reznor007 said:
Hmm, there isn't a program to fool quake 3 into thinking you have a p4 instead of an athlon?(like how there's a program to fool the games into thinking you have a different video card)
BTW, does SSE actually help athlon chips as much as 3dnow? And doom 3 shows similar performance to quake 3, so is it skimping out on SSE for athlon xps too, or maybe the P4s are using SSE2?
Reviewers have always attributed Quake 3's performance on P4s to its superior memory architecture. P4 has a lot lower latency and more bandwidth than Athlon XP.
I've also read that 3DNow on Athlon is sort of redundant since the FPU is powerful enough to approach the same performance. I don't know about SSE though.
I've also read that 3DNow on Athlon is sort of redundant since the FPU is powerful enough to approach the same performance. I don't know about SSE though.
swaaye said:
I think with Sisoft Sandra that SSE can give lower performance.(that may have been for opterons though)
As far as the p4 having more bandwidth and lower latency...well the more bandwidth is true(even without dual channel it can still somewhat take advantage of it), but with the right motherboards, memory, and memory settings an athlon xp can have memory latency as low as a p4. Plus the nforce2 boards have higher latency than via boards but offer better performance.
At 2.4ghz I've had my memory latency as low as 200ns, and at 1.2ghz around 100ns.