6800's and Fastwrites

Discussion in '3D Hardware, Software & Output Devices' started by mustrum, Oct 26, 2004.

  1. mustrum

    Regular

    Joined:
    Dec 26, 2002
    Messages:
    288
    Likes Received:
    0
    I don't have my 6800 GT yet so i can't test it out by myself.
    I read on a german hardware forum that FW have an incredible impact on the 6800's performance. I was tellign someone to disable FW to get stability andd this person showed me why this is no good idea.

    His results: (6800 ULTRA watercooled i guess)
    Aquamark 3 470/1300: 72800 without FW
    Aquamark 3 430/1200 75000 with FW

    Can someone here confirm this? I am looking for performance not for the biggest mhz score! If this is true disabling FW would be more than redicilous!
    How's the impact of FW in games and 3dmark05?
    I owned a modded VIOV so far and FW didn't impact on performance at all.
    The difference was within 50 points in 3dmark03 wich is nothing.
     
  2. wireframe

    Veteran

    Joined:
    Jul 14, 2004
    Messages:
    1,347
    Likes Received:
    33
    Before you go any further, you may want to get your data from a source where only one variable is being changed. In your example you are changing the FW status and the GPU/memory clocks. What is to say that the more heavily overclocked scenario is not throttling back due to overheating or insufficient power?

    I tried disabling Fastwrites on my GF6800 Ultra, but I didn't notice any major differences. Nothing scientific or even for the purposes of discovering a performance difference. I did it to test if it would resolve other issues. I noticed no difference in gameplay with FW on or off. Neither did it solve my problem.

    EDIT:

    Furthermore, you will want to consider driver/ForceWare revisions in regards to fastwrites. Some versions may have problems or fix problems with the AGP interface and certain chipsets. A beta driver that may fix an issue on one chipset may break compatibility on another.

    All in all I think it is safe to say that disabling fastwrites should not cause a major impact on performance. It may improve performance marginally in certain situations, but I think there is a reason why ATI, for example, often recommends disabling this feature: Any performance/feature gains it may offer is minimal and you run the risk of running into compatibility problems with certain AGP/GART implementations.
     
  3. Nick Spolec

    Newcomer

    Joined:
    Apr 7, 2004
    Messages:
    199
    Likes Received:
    0
    Unless you use MAME and emulators in general, which benefit greatly from the direct access between the GPU and CPU (since emulators, especially mame, do a lot of writing from the CPU to the video).
     
  4. mustrum

    Regular

    Joined:
    Dec 26, 2002
    Messages:
    288
    Likes Received:
    0
    Hmm i find more and more people backing up that theory though. People loose a lot performance by disabling FW.
    This is strange since FW did almost nothing on my x800.
    Can't wait to test this out by myself.
    Maybe it's really chipset/motherboard/bios related.
     
  5. DeathKnight

    Regular

    Joined:
    Jun 19, 2002
    Messages:
    744
    Likes Received:
    4
    Location:
    Cincinnati, OH
    It's always been best to have FW disabled with ATI's cards. With Nvidia cards it's kind of a toss-up I guess.
     
  6. Nick Spolec

    Newcomer

    Joined:
    Apr 7, 2004
    Messages:
    199
    Likes Received:
    0
    I have FW's enabled with my 9500 Pro and I've had no problems.. I think it is more the combination of the video card and motherboard..

    Right now I'm using an Albatron KX18D (NForce2 Ultra). Previously, I was using a Shuttle AK37 (KT400). When I had FW's enabled on that board using my 9500 Pro, I would get crashing galore in some apps (mainly Wolfenstein ET).
     
  7. ChrisRay

    ChrisRay <span style="color: rgb(124, 197, 0)">R.I.P. 1983-
    Veteran

    Joined:
    Nov 25, 2002
    Messages:
    2,234
    Likes Received:
    26
    I have been able to enable fast writes on my Nforce 2 board with just about any graphic chipset there is, From 9500 pro/9800 Pro to FX 5900/6800NU/6800GT without a problem thus far.
     
  8. Guden Oden

    Guden Oden Senior Member
    Legend

    Joined:
    Dec 20, 2003
    Messages:
    6,201
    Likes Received:
    91
    I have no problems at all with fastwrites in my Dell, i850 chipset (I guess Intel builds their mobos), either with a GF3 or 9800Pro. Sidebanding is a different story though! :lol: That feature is borked in GF3, results in near-instantaneous crasssssh if enabled. Works fine in the 9800. :)

    Haven't bothered checking if there's any performance differences or not, it's such a hassle with all the reboots required. :)
     
  9. Bludd

    Bludd Experiencing A Significant Gravitas Shortfall
    Veteran

    Joined:
    Oct 26, 2003
    Messages:
    3,654
    Likes Received:
    1,267
    Location:
    Funny, It Worked Last Time...
    My GF3 on an A7V333-X ran flawlessly with SBA enabled, and my R9800 on an P4C800-E Deluxe works flawlessly with both FW and SBA.
     
  10. wireframe

    Veteran

    Joined:
    Jul 14, 2004
    Messages:
    1,347
    Likes Received:
    33
    Can you show some numbers? The same exact setup with only fastwrites being enabled/disabled should do. Are you proposing Aquamark as the definitive test or do you have others in mind? I would not mind providing numbers as long as we discuss what the deciding benchmarks should be. However, I am quite confident that disabling fastwrites will not significantly alter any benchmarking results. However, my confidence shall not dissuade science from prevailing. Choose your benchmarks and let's do this the old fashioned way...by burning a witch!

    ...or through scientific benchmarking discovery.

    Some thoughts on choosing benchmarks:

    1. We probably want 3dMark 2001/2003, right?
    2. How about some actual game?
    3. Do we know of a benchmark that will beat up on the AGP so we can see the impact of fastwrites when AGP is being used to shuffle data rather than everything running from the video card's local memory?
     
  11. wireframe

    Veteran

    Joined:
    Jul 14, 2004
    Messages:
    1,347
    Likes Received:
    33
    Here are some completely unscientific number for Aquamark just to contradict what I said in the previous post about approaching this with method and purpose (single runs, but ran more tests that are not included here to quickly test impact of other tweaks in Rivatuner. Those tests returned similar results. Take that as you will):

    I used Rivatuner 2.0 rc 15.2 to disable fastwrites. I did not reboot when applying this switch, but I don't think it is needed.

    Aquamark was running using the settings: 1024x768, 0x AA, 4x AF, Details: V.High (HyperThreading enabled)

    Geforce 6800 Ultra (425/1200) + ForceWare 61.77
    Pentium 4 2.6C on Intel 875P chipset using AGP x8, SBA, and FastWrites available.

    Results with Fastwrites ON:

    GFX : 8,673
    CPU : 8,756
    Total Score: 58,016 (58.02 fps)

    Results with Fastwrites OFF:

    GFX : 8,658
    CPU : 8,756
    Total Score: 57,946 (57.95 fps)
     
  12. mustrum

    Regular

    Joined:
    Dec 26, 2002
    Messages:
    288
    Likes Received:
    0
    I got my 6800 GT now ad enabbling FW is no problem.
    I see a performance increase in the region like the above poster. In other words: NOTHING.
    Half a FPS isn't worth risking any instability or stutter to me.
    It's disabled. FW didn't affect my OC though.
    Seems like nforce2 boards have no problem with FW but also no real gain.
     
  13. BlackPearl

    Newcomer

    Joined:
    Oct 30, 2004
    Messages:
    3
    Likes Received:
    0
    Location:
    Texas
    I got no appreciable performance change in games and no reduction in stablility by enabling FW on my VIA chipset system, but playback of all video formats was significantly improved (sharp reduction in CPU utilization). I wish that I understood the reason for that improvement.
     
  14. Colourless

    Colourless Monochrome wench
    Veteran

    Joined:
    Feb 6, 2002
    Messages:
    1,274
    Likes Received:
    30
    Location:
    Somewhere in outback South Australia
    Writing to a video memory buffer is significantly improved when fast writes are enabled. So it takes less time for the CPU to write the data meaning the CPU utilization goes down.
     
  15. Guden Oden

    Guden Oden Senior Member
    Legend

    Joined:
    Dec 20, 2003
    Messages:
    6,201
    Likes Received:
    91
    ...Which means you never actually disabled it, since it cannot be toggled on-the-fly. It changes registry keys that are only read on bootup, so it's not strange you didn't see a difference outside the error margin.

    Anyway, I don't see why you think it's dangerous to have fastwrites enabled. Turn em on, regardless if you actually see a difference or not. It's stupid to have it disabled. You could just as well decide to disable write combining and CPU caches too if you're going to have that attitude.
     
  16. wireframe

    Veteran

    Joined:
    Jul 14, 2004
    Messages:
    1,347
    Likes Received:
    33
    Good catch. Strange that Rivatuner does not give a reboot and warning dialogue when using this option. It was quite tricky to disable for some reason. Using the Nvidia control center proved futile this time around, but I am almost 100% certain that it worked for disabling FW before.

    Anyways, I used Rivatuner to save the settings, rebooted, and ran a battery of tests. Without getting into details I can confirm that switching fastwrites on or off has almost no impact on my system. I tested with Quake 3 timedemos, 3DMark 2001 and 2003, and Aquamark 3. All gave the same results regardless of the fastwrite status. I am too lazy to provide the numbers right now so you will just have to take my word for it. So much for my 'scientific approach' huh? Well, it would not be a very interesting proof for fastwrites impact on performance with only one person providing numbers anyway.

    If you want the actual benchmarking data, just ask and I will paste it in here when I am feeling less lazy and more numeric.
     
  17. Guden Oden

    Guden Oden Senior Member
    Legend

    Joined:
    Dec 20, 2003
    Messages:
    6,201
    Likes Received:
    91
    Of course, fastwrites only makes a difference if the driver is coded to use it. If it builds display lists in main memory and the card fetches them itself via DMA we'll never see any impact at all from the feature regardless of if it's enabled or disabled. :)
     
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...