PDA

View Full Version : The Big ForceWare Comparison: 4 Officials and a Beta


Pages : [1] 2

boro
03-12-05, 03:12 PM
This test compares 5 Forceware drivers, being:

61.77, driver date 28/07/2004, official
66.93, driver date 09/11/2004, official
67.66, driver date 08/03/2005, official beta
71.84, driver date 24/02/2005, official
76.10, driver date 24/02/2005, beta

We'll find out what Nvidia's driver team has been up to in the past 6 months ;)

Test Setup

AMD XP Mobile @ 2450MHz
GeForce 6800GT @ 425MHz core and 1150MHz RAM
1GB PC3200
A7N8X deluxe

So the 6800GT is actually running at Ultra OC speeds, and the fact the CPU isn't exactly top of the line anymore means some tests will be CPU limited.

Drivers are set to the default settings, meaning Quality mode.

Testing method

Synthetic benchmarks are run at default settings.

Games are benchmarked with all video options set to highest quality, using timedemo where possible and otherwise Fraps. All games are tested at 2 settings: Speed mode (1024*768, no AA/AF unless it's used by default) and eye-candy mode (1600*1200, 4xAA/8 or 16xAF). AA/AF is set in-game or forced through the driver's control panel when needed.

The scores are not really representative for actual average fps while playing, as most demos are chosen to really stress the gpu.

I won't be comparing image quality between the drivers, but if you ask me, IQ went slightly up with each new driver, with especially 61.77 being poor in some games. Some features like the LOD bias setting added somewhere along the line improved quality as well.

Synthetic Benchmarks

http://photobucket.com/albums/v242/pictarz/3dmark05.jpg

Looks like one day I'll reach 6000 ;)
Large improvements are made both in SM2 and SM3 mode, with the gap between them closing in after 66.93. NVidia has certainly not been neglecting this benchmark.

http://photobucket.com/albums/v242/pictarz/3dmark0103.jpg

In 3DMark2001 61.77 is still king, in '03 we're still on the rise.

http://photobucket.com/albums/v242/pictarz/aquamark.jpg

As we'll see a number of times, 61.77 is a good driver for speed mode, but loses its advantage when the going gets tough. 76.10 looking very promissing with a 1000 point leap over 71.84

http://photobucket.com/albums/v242/pictarz/shadermark.jpg

Shardermark tests SM3 shader performance. 61.77, released prior to this benchmark, wasn't really ready for it. There's also a noticable dip from 76.10 in 1024*768. The scores you see are the results of shader test 19, the combination effect.

First Person Shooters

Let's start with the big three.

http://photobucket.com/albums/v242/pictarz/hl2d3.jpg

Halflife 2 is pretty CPU-limited on my rig, and my timedemo reflects that maybe even a bit too much. No real differences between drivers, nor between 1024*768 and 1600*1200 with everything on. 61.77 scores good but with less IQ to my feeling.

Doom 3 on the other hand can rely on a performance boost with each new (beta) driver. 12 frames the difference in 1024*768 since Doom 3's release last summer.

http://photobucket.com/albums/v242/pictarz/farcry.jpg

71.84 is the fastest of the officials, but once again 76.10 can just top it. With HDR Farcry still runs too slow to play on eye-candy mode, though from experience everything up to 1280*1024 or 1024*768 non-oc'ed plays OK on a 6800GT.

http://photobucket.com/albums/v242/pictarz/riddicket.jpg

Any 75.xx or higher driver comes with OpenGL 2.0 support, which causes a bug with Riddick. I had to force the OpenGL version string to 1.3 with Rivatuner to get the game to run with the 76.10 driver. 71.84 wins on points.

Enemy Territory, another OpenGL game, is fairly CPU limited, except in my demo with lots of airstrikes and fire. :p 76.10 somehow manages to squeek an extra few frames out in 1024*768.

http://photobucket.com/albums/v242/pictarz/pkut.jpg

All drivers perform almost equal in Painkiller.

71.84 looses some terrain in UT2004 to the previous 2 drivers, though nothing spectacular.

http://photobucket.com/albums/v242/pictarz/halo.jpg

I don't really think this is a reliable benchmark. Halo's timedemo function consists of a string of cutscenes, and seems to be taxing the hard disk too much, which causes variable results. Halo doesn't support AA.

Other Games

http://photobucket.com/albums/v242/pictarz/dtmflatout.jpg

DTM Racedriver 2 (aka. TOCA Racedriver 2 and V8 Supercars 2) doesn't officially support antialiasing, though it does sort of half work. This might be the reason for the very poor performance of 61.77 with AA.

Flat Out seems limited to 100 fps, but in eye-candy mode there's a one-fps gain with 76.10. Great. ;)

http://photobucket.com/albums/v242/pictarz/romedawn.jpg

Rome:Total War can stand right next to Halflife 2 when it comes to being CPU-limited. 61.77 sure is fast in speed mode.

Also little differences in Warhammer 40000: Dawn of War.

Conclusion

So what happenend in the last 6 months in terms of performance ? Some improvements, especially when stressing the videocard. The biggest leaps were from 61.77 to 66.93 and from the new 71.84 official to current betas.

Does 71.84 bring a new level of performance ? No, it doesn't, at least not on my rig. What it does bring is the same or better IQ without performance loss, and some speed gains here and there. Some bugs are fixed, some bugs are added, like the idle temperature bug (which I don't suffer from as I overclock my card). Some nice new features and finally fully working video acceleration? (have yet to test).

Judging from the 2 leaked release 75 drivers (75.90 and 76.10), the future promises good things for us NVidia users. But the list of bugs with these drivers is long, most annoying one for me is the fallback to low-power 3D clock speeds in some cases when I start or exit a game, or save/load, which seems to cause small pauses.

For now, I'm staying with 71.84 as I don't have a single problem with them.

Roliath
03-12-05, 03:37 PM
wow! Thanks, maybe i'll give 71.84 a try i'm still using the 67.03

radial
03-12-05, 07:00 PM
Well what can I say. Impressive work there boro, much appreciated. How do you even find the time to compile something like that? Anyway, well done!

radial
03-12-05, 07:01 PM
wow! Thanks, maybe i'll give 71.84 a try i'm still using the 67.03

Imho it is time to scrap those 67.xx drivers and move on to the next step :afro:

Acid Rain
03-12-05, 07:38 PM
Imho it is time to scrap those 67.xx drivers and move on to the next step :afro:
If you are into image quality, keep the 67.66s. All the others I've tried don't render as crisply, and the specular highlights and such don't come through as cleanly, so I'd recommend keeping the 67.66s.

Unless you aren't an IQ guy. If not, by all means, move on and snag that extra 0-3% performance increase. :p

Q
03-12-05, 08:25 PM
Excellent work. You are an asset to the forums.

Keep up the good work!

Vapor Trail
03-12-05, 08:27 PM
Nice job Boro!!!

:clap: :clap: :clap:

Wow, for a minute there I forgot I was reading a forum post and thought I had drifted off to a review site somewhere. Nice work! ;)


From what I can see there isn't really a big difference performance wise with all the last few sets. Better to go with the set that causes the least problems for what you play.

I'm sticking with the 71.84's at least until a new set comes along that fixes the temp bug issue.

OWA
03-12-05, 08:53 PM
Yep, nice job and thanks for taking the time to do it.

macatak
03-13-05, 01:10 AM
Nice job Boro!!!

:clap: :clap: :clap:

Wow, for a minute there I forgot I was reading a forum post and thought I had drifted off to a review site somewhere. Nice work! ;)
Ditto :)

michaelj
03-13-05, 10:22 AM
But what is the best one suit for my Geforce 6200??

rflair
03-13-05, 11:44 AM
Very good comparison, just two things, Flatout seems to limiting itself to whatever Hz your monitor allows as maxinum for said resolution, and also UT2K4 seems to be CPU limited.

Otherwise great review, and I hate to say it but it seems Nvidia is pandering to benchmarks more then anything, seeing as 3Dmark05 is seeing huge increases but nothing else really is.

Good job Boro.

GT-Force
03-13-05, 09:38 PM
This is simply a great job. Pro level testing & comparison. Thanks a ton.

boro
03-14-05, 04:59 AM
Thanks for the comments guys :)

JoKeRr
03-14-05, 03:06 PM
ahh excellent!! wish they could improve farcry and HL2 a bit further in performance though, those 2 games seem constantly owned by X800xtz

Waffles
03-14-05, 06:02 PM
well you have a 400mhz core competing with a 500mhz core and a VERY mature Shadermodel 2 pipeline.

I have no problem admitting that ATI is going to be 5 FPS faster under those circumstances. :)

|MaguS|
03-14-05, 06:36 PM
Nice Job.

EverGreen
03-14-05, 07:23 PM
Applause! :clap::clap::clap::clap: u beaten macatak :D

Karrde
03-18-05, 06:31 PM
Thanks, was looking for something like this.

ofelas
03-20-05, 10:28 PM
Nice work boro!

boro
03-21-05, 05:12 AM
does 76.10 work on riddick?
does 76.10 water in hl2 yellow?

yes, but you need to force the openGL version string to 1.3 or 1.5 in rivatuner
Yellow water in HL2 was a temporary problem with 71.xx but has been fixed, in 71.84 as well.

admiral1313
03-21-05, 05:14 AM
The water isn't yellow,but I think that the riddick bug is still there.There is a workaround for it somewhere in this forum.I really like the 76.10 drivers and I recommend you give them a try.

Darkoz
03-21-05, 07:09 AM
The water isn't yellow,but I think that the riddick bug is still there.There is a workaround for it somewhere in this forum.I really like the 76.10 drivers and I recommend you give them a try.

To get Riddick to work with the 76.10 driver and accept openGL 2.0 you can edit the gliConfig.ini file in the game's System\Win32_x86_SSE2 folder, or the corresponding folder depending on which CPU you have.
here's a copy of the edited file.......



LogEnabled = False;
LogFlush = False;
LogFileName = "gliInterceptLog"
LogFormat = XML;
XMLFormat
{
XSLFile = gliIntercept_DHTML2.xsl;
BaseDir = "C:\Program Files\GLIntercept0_41\XSL";
}

LogPerFrame
{
Enabled = False;
FrameStartKeys = (ctrl,shift,f);
OneFrameOnly = True;
}

GLFunctionDefines = "C:\Program Files\GLIntercept0_41\GLFunctions\gliIncludes.h"

GLErrorChecking = False;
ThreadChecking = False;
BreakOnError = False;
LogOnError = False;
ExtendedErrorLog = False;
DebuggerErrorLog = False;

ImageLog
{
LogEnabled = False;
RenderCallStateLog = False;
SaveFormats = PNG;
FlipXAxis = False;
TileCubeMaps= True;
SaveGLTypes = (1D,2D,3D,CUBE);
SavePbufferTex = True;
ImageIcon
{
Enabled=true;
Size = 40;
}
}

ShaderLog
{
LogEnabled = False;
RenderCallStateLog = True;
AttachLogState = True;
ValidatePreRender = False;
UniformLogPreRender = False;
}

DisplayListLog
{
LogEnabled = False;
}

FrameLog
{
LogEnabled = False;
SaveFormat = JPG;
ColorBufferLog = (pre,post,diff);

StencilColors = (0,0xFF000000, //Black
1,0xFFFF0000, //Blue
2,0xFFFFFF00, //Light Blue
3,0xFF0000FF, //Red
4,0xFF00FFFF, //Yellow
5,0xFFFF00FF, //Purple
6,0xFF80FFFF, //Bright Yellow
7,0xFFFFFFFF); //White (Note green is not used as that is the "diff" color)
}

TimerLog
{
LogEnabled = False;
LogCutoff = 20;
}

PluginData
{
BaseDir = "C:\Program Files\GLIntercept0_41\Plugins";
Plugins
{
ExtensionOverride = ("GLExtOverride/GLExtOverride.dll")
{
VendorString = "NVIDIA Corporation";
RendererString = "GeForce 6800 GT/AGP/SSE2";
VersionString = "1.5.2";
}
}
}

Burnt_Ram
03-22-05, 09:01 AM
nice comparison. i was thinknig about trying the 76.10's now i will ;)

|MaguS|
03-22-05, 09:26 AM
Anyway of throwing EQ2 into the benchmark comparison? Many people play that and are looking for the optimal set since we gotta wait for Nvidia to fix the issues...

subbo
03-22-05, 11:06 AM
Great comparison. Now I have proof for my feelings that the official 71.84 is a pretty solid driver. I will install these into my friends rig this week when he gets his 6800GT.

IQ in many games and even 3dmark05 seems to be much approved, especially the water monster no longer has black blotches flying all over, and my previous score of 5135 changed to 5226.