AMD A64 S754 Memory timings explored on nForce3 150

Memory by piotke @ 2004-04-15

On an nForce2 based platform it was proven that that CL2 2-2-5 was not the fastest setting for your ram, in fact CL2 2-2-11 was! But what about the new A64 S754 , does it show the same anomality when combined with the nForce3 150? Is the famous CL2 2-2-5 back? Does Cas Latency influence performance a lot? Let?s find out

Introduction

Short intro

Rumours.

They make me write these articles. Not that long ago we had the CL2 2-2-11 timing “issue” on the Nforce2,and now I heard that Cas Latency 2 has no benefit when compared to 2.5. It is time to spend some hours booting, rebooting and benching.

To keep me occupied during those “booting-sessions” I started making figures with the spare spare memory I've got, more models to follow later (I also take requests :-) )

Madshrimps (c)


Now that memory prices are going up again everybody wants to buy his memory now and as cheap as possible. What memory to get? Some normal TwinMos or maybe that nice Mushkin, which costs you an arm and a leg. I hope this review can help you make a decision to choose wisely.

In case you missed my nForce2 memory timings article then this link will take you there.


Don't know what all these terms mean? Ras / Cas / Tras ? More and detailed information concerning this can be found here.

Test setup & settings

The test setup :


Test setup
CPU AMD Athlon 64 3000+
Mainboard Shuttle AN50R
Memory 1 * 256 Mb pc3500 Corsair BH-5
VGARadeon 9700 Pro


Both CPU and Radeon were running at stock speeds. Memory was always sync with the HTT speed @ 200 Mhz.

I also wanted to test some Cas 3 settings but couldn't find any memory that wanted to run at those setting on my Shuttle. Although the sticks I tested were capable of running CL3 on other mainboards.

The different memory timings in this test:

  • CL2 2-2-5
  • CL2.5 2-2-5

  • CL2 2-2-11
  • CL2.5 2-2-11

  • CL2 2-4-11
  • CL2.5 3-3-7
  • CL2.5 4-4-8

    I haven't taken a snapshot from CPU-Z with every possible setting, but to give you an idea of what these numbers above represent, behold this screenshot:

    Madshrimps (c)

  • Sisoft Sandra & SuperPi

    Sisoft Sandra - Memory Bandwidth Benchmark

    Benchmark used: Sisoft Sandra 2004

    You can't make any decision, based on this test... The difference between the two edge values isn't even 100 Mb/sec.
    But, CAS 2 seems to be always faster then CAS 2.5 and 2-2-2-11 isn't the fasted here. But the differences are too small to name a clear winner.

    Madshrimps (c)



    SuperPi

    Benchmark used: SuperPi

    It's very simple. CAS 2 is always the fasted. And so is 2-2-2-5.

    Madshrimps (c)

    3DM01 & Particle Fury

    3DMark 2001Se

    Benchmark used: Futuremarks’ 3DMark2001SE

    Only a meagre score difference of 450-500 points, I honestly expected more, but at there is already a noticeable increase of performance.

    Madshrimps (c)


    Particle Fury

    Benchmark used: Particle Fury


    Despite the alpha-blended OpenGL goodness, Particle Fury is actually NOT fillrate limited! CPU (particularly FPU) and memory bandwidth and latency are the bottlenecks with modern-era 3D cards. Only when I go below 5000 particles onscreen do I get fillrate limited.


    This part is getting boring, once again CAS 2 and 2-2-2-5...

    Madshrimps (c)


    Madshrimps (c)

    Q3A & Conclusion

    Quake 3 Arena

    Benchmark used: Quake 3 Arena

    A difference of about 3 or 4 frames, while using CAS 2 instead of 2.5.

    Madshrimps (c)



    Conclusion

    There is a difference between CAS 2 and 2.5 not nothing to shout home about.

    If you're a 3d Mark enthusiast in search for the top spot or just somebody who wants the best of the best, get decent ram that can handle the lowest timings. But those Winbond BH-5/BH-6 chips are no longer made so getting a new pair of those sticks (if you find them) will cost you quite a lot. The meagre performance increase doesn’t justify the asking price at all.

    Unfortunately I couldn't test with any CAS 3 memory, but hey, who has such memory anyway in an AMD rig? :-)

    Lowest timings possible seem to be the best, but an average user will be satisfied by normal, not specially for overclocking designed memory. The A64 system has already enough power (for now).


    Questions/Comments: Forum thead

    /Piotke out.
      翻译: