updated R350 preview (new driver optmization tests)

[Ante P]

http://www.nordichardware.com/reviews/graphiccard/2003/Radeon9800Pro/index.php?ez=14

 

[Mintmaster]

Is the driver the cause of the really good 9800 vs. 9700 results in UT2003 4xAA/8xAF on page 13 of your review?

 

[Luminescent]

Were those tests made with the 9800pro downclocked to 325/310?

 

[Althornin]

Well, the next page shows the default clocked 9800 scoring 5400 3dmark03's
So i'd say yes, that is the 9800 downclocked.

 

[mboeller]

Why have you done all tests at 1024x768 only. 1024x768 seems to be CPU-limited most of the time.

How about further tests at 1600 x 1200 with 2x AA; 4x AA and / or 16xAF

Then we would really see if the R9800 is better then the R9700 MHz for MHz

 

[ET]

Have you tested to see if there are any graphical issues caused by the hack? It may be that the 9800 simply fixes some paths that don't work perfectly on the 9700 and are disabled in the drivers for this reason. It may not even be all that apparent (for example, lesser anisotropic filtering, or whatever).

 

[Luminescent]

How about putting in some 3DMark03 pixel shader results and some shadermarks. This would help determine the differences between the 9700pro and the 9800pro during shading.

 

[Luminescent]

How about putting in some 3DMark03 pixel shader results and some shadermarks. This would help determine the differences between the 9700pro and the 9800pro when shading.

 

[rubank]

From NordicHardware

The tests show us that the architectural differences were smaller than we thought at first...

Hmmm, are you sure? Would the result be the same if you compared 9800 at normal clocks to a 9700 overclocked to the 9800 speed?

Memory timings, latencies ....

 

[Ante P]

Is the driver the cause of the really good 9800 vs. 9700 results in UT2003 4xAA/8xAF on page 13 of your review?

Nope, the drivers didn't effect UT2003 strangely enough. (Not at all from what I could tell.)

 

[Ante P]

Were those tests made with the 9800pro downclocked to 325/310?

Actually it was clocked to 326/620 because whenever I clocked the core to 325 I would get instant corruption.
(I overclocked the 9700 to 326, of course that didn't matter much)

 

[Ante P]

Why have you done all tests at 1024x768 only. 1024x768 seems to be CPU-limited most of the time.

How about further tests at 1600 x 1200 with 2x AA; 4x AA and / or 16xAF

Then we would really see if the R9800 is better then the R9700 MHz for MHz

The tests on that page are not CPU-limited. None of them are.
If they were I wouldn't have used them.

 

[Ante P]

Have you tested to see if there are any graphical issues caused by the hack? It may be that the 9800 simply fixes some paths that don't work perfectly on the 9700 and are disabled in the drivers for this reason. It may not even be all that apparent (for example, lesser anisotropic filtering, or whatever).

Haven't seen anything yet. I'll keep looking.
I'm now using the hacked driver on my primary rig.

 

[Ante P]

How about putting in some 3DMark03 pixel shader results and some shadermarks. This would help determine the differences between the 9700pro and the 9800pro during shading.

I f'ed up and lost the detailed results from the 9800 Pro somwhow.

 

[Dave Baumann]

I'm not sure those tests are actually very reflective, my review showed the performance to actually decrease in most gaming situations. However, compare and contrast the differences in my review and this in fablemark - they are withing the same ballpark (although my performance is lower overall).

Try some tests such as PS1.1 (Pixel Sahder in 3DM2001) and 4x FSAA.

 

[Ante P]

From NordicHardware

The tests show us that the architectural differences were smaller than we thought at first...

Hmmm, are you sure? Would the result be the same if you compared 9800 at normal clocks to a 9700 overclocked to the 9800 speed?

Memory timings, latencies ....

In any case it does alliviate some of the differences between the two card so no matter how you look at it it does make the performance differences smaller than we first thought.
And also I tried the tests with the overclocked 9700 Pro too.

 

[GreenBeret]

AnteP

One thing I noticed when using Catalyst 2.5~3.1 (I have a R9700 Pro) is that some HyperZ features are disabled by default (I check with Rage3D Tweaker). Could that be that Cat 3.2 has them on by default ?

 

[Ante P]

I'm not sure those tests are actually very reflective, my review showed the performance to actually decrease in most gaming situations.

Could you be more specific.
When did the performance decrease?

As for running more tests I didn't get to keep the R350 so that's a no go.

 

[Ante P]

AnteP

One thing I noticed when using Catalyst 2.5~3.1 (I have a R9700 Pro) is that some HyperZ features are disabled by default (I check with Rage3D Tweaker). Could that be that Cat 3.2 has them on by default ?

You mean that it has them disabled without the hack?
Well in any case Heirachial Z is disabled in OGL according to Rage3D Tweaker.
Also Z compression is set at "default" (goes from 0-3) in D3D.

But it's been like that in all the later drivers I've tried out.

 

[GreenBeret]

You mean that it has them disabled without the hack?
Well in any case Heirachial Z is disabled in OGL according to Rage3D Tweaker.
Also Z compression is set at "default" (goes from 0-3) in D3D.

But it's been like that in all the later drivers I've tried out.

Yep, disabled without the hack, and it's not just Hierrachical Z. If you switch Rage3D tweaker to Advanced mode, you will see heaps more in both OGL and D3D (I don't have it installed atm so can't give the details).