OpenGL Performance

Quake 3: Arena

Once again, the Q3A engine is ubiquitous enough among games that it warrants being included in the benchmarks. On a side note, this benchmark is interesting as there is a change in concavity in the graph of our benchmark data. This gives us an example of a benchmark where the fast get faster and the slow get slower.

Quake III Arena

With our average performance number here being 387.6, the Athlon 64 2800+ has beaten the price/performance curve once again (this time by only 4% of the range).

Jedi Knight: Jedi Academy


Jedi Knight: Jedi Academy

Wolfenstein: Enemy Territory


Wolfenstein: Enemy Territory

In all the OpenGL games, price/performance of the 2800+ is marginally better than linear in the price range and performance of the two 3000+ models.

DirectX 8 Performance DivX 5.1.1 Encoding
Comments Locked

20 Comments

View All Comments

  • Pumpkinierre - Wednesday, April 28, 2004 - link

    Kristopher #18, I presume your correction applies to my post #11 where I quote from your April 16th article. If this is the case then it still doesnt make sense as the 2800+ is only a CG stepping and therefore already IS a Newcastle so doesnt have to "transition". Also you quote:

    "In a nutshell, Clawhammer cores will all be replaced by Newcastle cores (130nm SOI, 512KB L2 cache). "

    Well as the white paper declares CG cpus can be 1MB L2 (coded AR) or 512K (AR and AX). So if your above statement is right it CANT just be the CG review, it must be the AX variant of the CG review. The CG definition is not the stepping. There are three steppings in the CG family. SH7-CG corresponding to AR on the cpu designation, DH7-CG to AX and CH7-CG Christ knows?. see P6 of:

    http://www.amd.com/us-en/assets/content_type/white...

    All steppings have sckt 754 and 939 derivatives which in my book could be classed as steppings in their own right making 6 CG steppings (and CPUID command will return a different stepping No. for each sckt type). In the sckt 754 case, the SH7 stepping (AR) has a CPUID stepping identification return of F4Ah while DH7 (AX) has FC0h (and CH7 has F82h). This corresponds with your model 4 for Clawhammer and C for Newcastle but that doesnt take into account the model "F" ie the Sckt939 DH7-CG stepping (FF0h) soon to be released. My interpretation is that Newcastle is not just any 512K L2 A64 but the one where the L2 hasnt originated from a 1Mb L2 being crippled. So die size will determine true Newcastles (and further down the track Paris cores).
  • TrogdorJW - Wednesday, April 28, 2004 - link

    Looks like this article was started a while back. "Unreal Tournament 2004 is coming down the line very shortly"!? Or maybe they just mean that the release version of UT2K4 has really shitty benchmarking in comparison to UT2K3? Probably. Still, the fact that there was *NO* overclocking information indicates that this is either old stuff just getting finished, and AT didn't want to redo the tests using one of the new 754 mobos that have a working PCI/AGP lock, or else they just pushed it out early.

    Something else to think about, the 1.8 GHz 2800+ compared to the 2.0 GHz 3000+ is a case of bad PR ratings by AMD (again). If the 2.0 GHz is a 3000+ and you reduce it's speed by 10%, then the 1.8 GHz A64 should really be a 2700+. While the charts make the 2800+ look a lot slower in many instances, it is usually almost a perfect scaling with clock speed.

    For example, the page on 3D Studio states, "It is likely that Lightwave is sensitive to memory speed, and memory speed differences could have had a part to play in this dichotomy in 3D rendering performance." So, you have the 1.8 GHz running 10% slower than the 2.0 GHz (or the 2.0 GHz is 11% faster if you prefer that). And the Lightwave scores? 61 seconds for the 2800+ and 55.3 for the 3000+. That makes the 2800+ 10% slower (or the 3000+ is 9% faster - timed tests give odd results). Basically, memory has nothing to do with it. The CPU core speed is the only real factor here (and L2 cache size if you look at the other A64 chips).
  • KristopherKubicki - Wednesday, April 28, 2004 - link

    Pumpkinierre: That sentence should read "new Newcastle core"; meaning the new CG stepping.

    Kristopher
  • thebluesgnr - Tuesday, April 27, 2004 - link

    I'd like to point out that the "The test..." page is missing the motherboard used for the Athlon XP system.

    I'm guessing it's the ASUS A7N8X Deluxe 2.0?
  • AtaStrumf - Tuesday, April 27, 2004 - link

    They're just calling them Newcastles for simplisity sake. Of course they're just Hammers with half of the L2 Cache disabled. As a matter of fact all Opterons/FX/64 up to now are one and the same chip with different packaging.

    When a real new core comes out, you can be sure AT staff will know about it before you do :)

    This new CG revision is probably a good sign we'll be seing real Newcastles and Socket 939 chips real soon.

    But they again, why get excited. I for one am not gonna pay throug my teeth for few percentige points difference in some weird bechmark that works word/excel/powerpoint/... 10x faster than I could ever dream to. What's the point people? Bragging rights, I'll give you that, but is there anything else.
  • Pumpkinierre - Tuesday, April 27, 2004 - link

    On the Newcastle naming if you look at the AMD white paper (p16-10):

    http://www.amd.com/us-en/assets/content_type/white...

    Then there are two revisions C0 (old one) and CG (new one) for the A64. The CO has one part number AP and CG two part numbers AR and AX. In the ensuing a64 core specs in the white paper, AP and AR have both 1MB and 512K L2s while AX is ONLY 512K L2 caches. Power and thermal specs are the same for all parts. So Newcastle could be AX (if K.Kubicki's comments about Newcastle having only 512K are correct) or it could be CG - the newer review. Or it is dual channel Sckt939 yet to be released. Take your pick.

    The 3000+ is released under all part numbers (AP,AR and AX) while for the 2800+ there is only one: AX and hence may be the first true Newcastle. So, my bet is that the 3000+ AP and AR designations are crippled 1Mb L2 cache 512K variants and AX are smaller die size pure 512K L2 cored cpus.

  • fatdog6 - Tuesday, April 27, 2004 - link

    well i build quite a few systems and I can not agree that it will have a much longer shelf
    (i'm gonna read that as usable)life IMHO in 2 years
    we will have a working 64bit computing enviroment and will be on like socket 1257 mobo's using enhanced ddr v8.3 and a solid state hdd so either way it's gonna be a whole new system. In my 10+ years of computing I have learned that you can not build a future proof computer.
    Peace..
  • Myrandex - Tuesday, April 27, 2004 - link

    fatdog, I agree as they do sell 3000+'s with 200MHx. FSB's. But I do have to say that a setup based on this will have a much longer shelf life then the regular AthlonXP if one was to buy one today, because they do have the possibility to upgrade to a different CPU at a later date without buying a new motherboard, instead of being forced to buy a whole new setup.
  • fatdog6 - Tuesday, April 27, 2004 - link

    What really bothers me about this review is why in the world did they use an xp 3000+ running a 166mhz fsb ??? I think I can safly say most of us are running xp's that are overclocked to at least a 200fsb like my mobile 2600+ @ 2.6ghz 200x13 throw that in the mix and lets see who comes out as price/preformace leader
  • Pumpkinierre - Tuesday, April 27, 2004 - link

    On Newcastle cores from Kris. Kubicki's april 16th article (http://www.anandtech.com/cpu/showdoc.html?i=2027):

    "All original Athlon 64s were based off the Clawhammer core; 130nm SOI with 1MB L2 cache. The recent Athlon 64 3000+ was also based on the ClawHammer core, but with half the L2 cache disabled"

    and further:

    "We did not find any mention of the Athlon 64 2800+ transitioning to the Newcastle core,"

    Maybe AT needs to get together and have a group hug on this one. My understanding is that Newcastles are the new 512K L2 dual mem.bank Sckt939 cpus now being released in may.

Log in

Don't have an account? Sign up now