System and Memory Benchmarks

SiSoft Sandra 2004 64-Bit

The 64-bit version of Sandra 2004 has been available for a while, but we did not have an Operating System to reliably run with the 64-bit version.  Sandra 64-bit runs fine on the XP64 preview.  While Sandra is a Synthetic Benchmark, we were curious to see if there would be any performance difference in memory, CPU Arithmetic, and Multimedia benchmarks between the 32-bit and 64-bit versions.  Everything was kept the same; we even used nVidia drivers close to the same version number.  The only difference is Sandra 2004 tests were run on XP Pro, while Sandra 2004 64-bit tests were run on XP 64-bit Preview Edition.

SiSoft Sandra 2004 - Athlon 64 FX51 Performance

 

32-Bit

(Windows XP SP1)

64-Bit

(XP64 Preview Edition)

% Change

32 to 64-bit

Sandra 2004 Standard

Buffered

INT 5722

FLT 5660

INT 5910

FLT 5831

+3.2%

Sandra 2004 UNBuffered

INT 2588

FLT 2682

INT 2811

FLT 2791

+6.3%

Sandra 2004 CPU Arithmetic

9161 mips

3470/4534 mflops

10121 mips

3881/4105 mflops

+10.5% mips

-0.2% mflops

Sandra 2004 CPU Multimedia

INT 16404

FLOAT 21642

INT 16598

FLOAT 22869

+1% INT

+5.7% FLOAT

The 32-bit vs. 64-bit results in Sandra are very interesting.  Even in this pre-release version of XP64, the Athlon 64 CPU and Memory Performance is higher than in 32-bit Windows XP.  Mips, which is based on ALU tests, is more than 10% faster, and Integer and Float tests in the Sandra 2004 Multimedia benchmark is 1% to 6% faster.  The only area without increased performance in 64-bit is the mflops component of the Arithmetic benchmark.  If we look closer, this benchmark is a combination FPU performance and iSSE2 performance.  While Floating Point increases some 11.6% in the move from XP to XP64 Preview, the Intel SSE2 results decrease by about the same amount.   The net result is virtually no change in the composite mflops.  We do not know if this is because Intel SSE2 is penalized by 64-bit operation or whether XP64 and/or Sandra 2004 64-bit benchmark require some optimizations for 64-bit performance.    

Super Pi

Super PI is very simple - it calculates the value of pi.  In the benchmark you can select the number of placed for calculation, and we used 2 million places as used in memory tests at AnandTech.   

Super Pi - Athlon 64 FX51 Performance

 

32-Bit

(Windows XP SP1)

64-Bit

(XP64 Preview Edition)

% Change

32 to 64-bit

Super Pi

2M Places

88 seconds

88 seconds

0%

As you can see, Super Pi was exactly the same result in both 32 and 64-bit.

Performance Test Configuration Media Encoding and Gaming Benchmarks
Comments Locked

42 Comments

View All Comments

  • gmenfan - Sunday, May 16, 2004 - link

    I would like to see performance benchmarks with Windows XP 64-bit on a socket 754 platform. I am curious to see if the dual memory controller makes a difference in a 64-bit environment.
  • DallasTexas - Friday, February 13, 2004 - link

    I sure see lots of apologies here for the initial crappy gaming performance shown. But anyway, I'd like to add to the chorus anywahy "yippeee, 64 bits is here ".
  • TrogdorJW - Thursday, February 12, 2004 - link

    What really makes me laugh at this whole situation is the fact that people have been saying Windows XP-64 was ready to be released months ago and that MS was just delaying it so Intel could create their own 64-bit chip. Where are all those stupid comments now, people? As I said on numerous occasions, gettings drivers and the critical OS paths ported from x86 assembly to AMD64 assembly is a difficult task at best. However, it's a task that must be done for the OS to reach its potential.

    WOW-64 would most likely not account for the 20% performance decrease in games that are witnessed here. It is a type of emulation (actually, just an extra layer that system calls have to go through), so it would definitely be slower than running the same application without that layer. However, if the 64-bit drivers and everything else are tuned properly, the extra work done in WOW could be negated by the potential performance boost that 64-bit drivers and such would experience.

    I would be curious to know what version of Nvidia's drivers were being used in testing. They said they used a version number that was "close" to the XP-64 driver version on the XP system, but that doesn't tell us a whole lot. If the 64-bit Nvidia driver is three months old or more, it might be back in the 45.23 driver era, which might explain the performance difference better.

    Still, the thing that is illustrated best by this article is that XP-64 is *not* ready for prime time right now. Fool around with it if you want to, but until all of the drivers are fully ported to AMD64, you're better off running XP or a 64-bit version of Linux. Even once all the drivers are available, I wouldn't be surprised to see companies like ATI and Nvidia releasing new versions that contain better 64-bit optimizations for the next year or two at least.
  • DallasTexas - Wednesday, February 11, 2004 - link

    "...Performance of current 32-bit games under SP64, however, was below expectations..."

    BELOW EXPECTATIONS ? WOW, the understatement of the year and it's only february.
  • CrossfireAction - Tuesday, February 10, 2004 - link

    20% Performance loss in games shood realy be a driver problem as you can check out here:

    http://www.easy-mod.de/emcontent/content.php?id=17...

    These guys testet the new Beta XP 64 with an ATI Radeon 9700 non Pro with Microsoft 64 bit drivers based on OEM Catalyst 3.6 from August 2003
    As you can see the gameing performance is increasing about 4,8% to 8% !!!
  • ZapZilla - Tuesday, February 10, 2004 - link

    Even if the gaming performance of 32bit games stays poor, 1 year from now, when faster CPUs/GPUs are out and the 64 bit OS is mature, it won't matter, because:

    1) new games will be 64 bit or DX9++ and require new hardware anyway to get best performance or

    2) 32 bit games (then called 32 bit legacy games) will benefit from faster CPUs/GPUs and get higher FPS to make up for it.

    How many FPS does Quake n get on a modern system these days compared to when the game was released?

    So why all the fuss? Its all good.
  • sipc660 - Monday, February 9, 2004 - link

    some say its emulated others say its not
    some say its emulated others say its not

    i agree with no.33
    just shut up everyone and wait till the guys from anandtech bring us the revised performance with some decent drivers and real 64-bit benchmarks

    hail anandtech

    go amd
  • FalcomPSX - Monday, February 9, 2004 - link

    I plan on doing some benchmarks using sandra soon, using the 32-bit sandra and 64-bit as this should show a good raw increase in performance, albeit an early example. But this is the best way to guage performance based on what we have now. 32-32, 64-32 and 64-64. Once this is done, we will have a early estimate in the gain 64-bit provides and the hit( eumlating 32 bit in the 64bit os provides.
  • Phiro - Monday, February 9, 2004 - link

    This article was _not_ a waste of time and I applaude Anandtech for spending the hours neccessary to put it together for us.

    The lack of intelligence and sheer stupidity (and trolling) in the comments for this article on the otherhand is overwhelming.

    Comments like "this just makes Prescott look better and better", "This is an AMD only version of Windoze what will Intel do haw haw haw", "a 15% increase in number crunching but a big hit on game playing - I don't think consumers will swallow that" and the hair splitting going on between the definition of "subsystem" and the definition of "emulation" make me want to put a gun in my mouth and just pull the trigger.

    Holy gosh people, please give me intelligent comments to read while I eat my lunch, the drivel coming out of most of you is enough to spontaneously create another CRAMITPAL.
  • araczynski - Monday, February 9, 2004 - link

    this is akin to putting in a new engine into a car and taking it out on the track before you put on its tires. the results are obvious and the whole excercise is a waste of time. you might as well start benchmarks on prescott's successor and discuss the results with a straight face.