DirectX 9 Performance

As we have seen before, processor speed has much less impact on DirectX 9 games than does video card power. Most of the differences in these numbers are negligible, but Aquamark shows some interesting numbers when comparing the performance of Athlon 64 based processors and the Athlon XP.

Aquamark


Aquamark

As we can see, most of the frames per second scores are too close together to really mean anything, except where the Athlon XP 3000+ is concerned. In this benchmark, the 3000+ lags its nearest neighbor (the Athlon 64 2800+) by about 8.3%. The 2800+ is much closer to the 512kB L2 based 3000+ in this benchmark.

Aquamark

Aquamark tries to isolate the impact of the CPU and GPU on overall performance by giving different scores for both. We can't really be sure what this score is based on, but it is interesting to look at when trying to understand the effect of different CPUs on the benchmark. We can see once again that the Barton 3000+ lags its 64 bit brother by quite a good margin.

Halo


Halo

GunMetal


GunMetal

SYSmark 2004 DirectX 8 Performance
Comments Locked

20 Comments

View All Comments

  • Pollock - Tuesday, April 27, 2004 - link

    I thought Unreal Tournament 2004 was already released?

    I was under the impression that some of the 3000+s were indeed Clawhammers with half the cache disabled, and that some were true Newcastles manufactured at those specifications. I think the 2800+ is like that, too, in that some of them are ones that failed at 3000+ speeds and that others are manufactured at those specifications. But I'm not completely sure.

    I agree with AtaStrumf, I don't see why people are making such a big deal over socket 939. The performance difference with dual channel memory has already been demonstrated to be negligible with socket 940 processors, so that's no reason to get 939 over 754. And upgradeability, in my eyes if you have the money to buy a new processor every few months or so, you shouldn't be concerned with upgradeability. People get socket 754 for value, so you only want to buy one processor every 18-24 months anyway, like which AtaStrumF said, a new socket should be out.

    And speaking of little performance loss, I, too, would have liked to see overclocking results with the 2800+...after all, that's the whole point of it, really.

    Rant over.
  • SKiller - Tuesday, April 27, 2004 - link

    those are OEM prices btw
  • SKiller - Tuesday, April 27, 2004 - link

    The current street prices for those are currently about what's listed below.

    XP 3000+ $120
    64 2800+ $170
    64 3000+ $210
  • Cybercat - Tuesday, April 27, 2004 - link

    Most of the benchmarks are VPU limited. It would have been better to use older games with low resolutions in order to eliminate the graphics card from the equation. It's hard to tell which is better in gaming with this method, since with most of them it's so close it's within margin of error.
  • AtaStrumf - Tuesday, April 27, 2004 - link

    I'll just stick to my 2500+@3200+ on NF2Ultra400/SoundStorm mobo, thank you very much.

    All these new CPUs look like a total waste of money, when there is such a cheap alternative out there (AthlonXP), that is not that much slower. At least not enough for anyone who doesn't run benchmarks all day long to notice. And that's most of the people, alhough not most AT readers :)

    As for upgreadability that everybody is so concerned about. Forget about it! Just make sure that what you buy will last you 12-18 months, after that you can be preety sure you'll have to change at least the mobo with whatever CPU you will be getting then. Intel and AMD seem to be on top of that one.

    And one last thing. Most of your readers like to OC their chips, so next time you do a review, try to OC it and include the results.
  • skiboysteve - Tuesday, April 27, 2004 - link

    yeah all these graphs are wrong. the newcastle is a completly different core than what is tested here.
  • wassup4u2 - Tuesday, April 27, 2004 - link

    The 3000+ is a ClawHammer with 1/2 of the cache disabled. Starting soon, however AMD will transition the A64 from ClawHammer to Newcastle. I think this actually does start with the 2800+.
  • iCeVbLaSt - Tuesday, April 27, 2004 - link

    actually AMD have confirmed that the 3000+ is a Clawhammer with disabled chache (that failked in chache test with 1mb so half was disabled) ... also if you'll check those chips manufacturing and stepping code and cheeck AMD charts you'll discover its a clawhammer .... I dunno why anandtech keep claiming its a Newcastle since AMD sayd by themself that the Newcastles production will start only around the 2 quater of this year and the 3000+ were produced for a long time (mine is week 50 of 2003)
  • Jeff7181 - Tuesday, April 27, 2004 - link

    I keep wondering if these are in fact completely different cores, or if they're a regular "hammer" core with half the cache disabled. In other words, is an A64 3000+ just an A64 3200+ with half the cache disabled, similar to a Thorton? Or is it in fact a completely separate core with a different transistor count to reflect the smaller cache?

    I was under the impression that the Newcastle core was made differently... if all that means is that it's a Hammer with half the cache disabled, then fine. But reading this makes it sound like the difference between a Thoroughbred B and a Barton core.
  • f11 - Tuesday, April 27, 2004 - link

    you can buy a barton 2500 and oc to 3200 (a simple fsb change) and that costs like 1/3rd the price of this. no point. great article though, it least it clarifies the (lack of) speed advantage over alternatives.

Log in

Don't have an account? Sign up now