ET article confirms ATi/Nvidia cheats
- Thread starter John Reynolds
- Start date
Doomtrooper
Veteran
Interesting, they claim that ATI is 'replacing shader code', while ATI states just re-ordering.
The quote above is expected, the Red Phone has been ringing off the hook at Nvidia..call in the Damage Control Team.
The quote above is expected, the Red Phone has been ringing off the hook at Nvidia..call in the Damage Control Team.
Myrmecophagavir
Newcomer
They found a more noticeable change for the 9800 in game tests 2 & 3 than I did, I tested with different res, AF/AA and CPU though; however I didn't think those tests were considered suspect for ATI ATM?
I don't think ATI intended to imply they were applying a general re-ordering algorithm, since that would still work when FM changed the code trivially, rather that they re-jigged the code by hand and then detect FM's exact shader and replace it with their own as Nvidia did.Doomtrooper said:
ET is testing at high resolutions, and not the 3DMark default. The net result of this is that the parts that are Pixel Shader optimised will have a greater difference, but the buffer clears will have a lesser difference (the more frames being drawn the more not clearing buffers gains you).
Myrmecophagavir
Newcomer
I thought the preliminary accusation against ATI didn't include buffer clearing, just replacement shaders in GT4 (I suspect GT1 too)? Then it would scale for resolution & AA/AF, agreed, but not in tests 2 & 3 which is where ET noticed a reasonable difference. I mean I saw literally no change in 2, and 0.3% in 3.
Looks like Extremetech's screwed up 9800 Pro numbers with 330 patch...I bet he has something like aniso on for 330 and off for 320...My testing here shows that the overall 3dmark03 number does not vary more than 3% between 330 and 320 at all resolutions. Not sure how ET got over 10% variance.
Not the first time that ET screws up the numbers. The least any website that portrays itself and makes money as a hardware review site should do is to make sure that the numbers they print are accurate. I had an impression before that ET was usually a reliable site...looks like they are getting old and jittery
Not the first time that ET screws up the numbers. The least any website that portrays itself and makes money as a hardware review site should do is to make sure that the numbers they print are accurate. I had an impression before that ET was usually a reliable site...looks like they are getting old and jittery
That'd only be a bad thing if NVIDIA didn't suck so much right now.
Oh well, can always hope NV40 will Magically Fix Everything. And that they'll fire their idiot PR department.
Oh well, can always hope NV40 will Magically Fix Everything. And that they'll fire their idiot PR department.
I just downloaded the patch and ran some before and after tests. I didn't want to bench all night so I only used 1024x768 and 1600x1200. This is on a 9700pro with cat 3.4 drivers.
Here are my numbers:
1024x768
320 | 330
test 1. 95.6|95.6
test 2. 32.9|33.0
test 3. 28.4|28.4
test 4. 30.9|28.5
1600x1200
320 | 330
Test 1. 78.2|77.9
Test 2. 16.0|16.0
Test 3. 15.2|15.2
Test 4. 21.4|19.0
Nothing like extreme techs numbers. The only tests that changed besides the Nature numbers were the test 2 in default rez went up by .1fps and the game test 1 at 1600x1200 went down by .3fps. I'd say aside from the nature scores the other tests seem to be within the proper range.
Not sure what ET did but their results don't match mine at all.
Here are my numbers:
1024x768
320 | 330
test 1. 95.6|95.6
test 2. 32.9|33.0
test 3. 28.4|28.4
test 4. 30.9|28.5
1600x1200
320 | 330
Test 1. 78.2|77.9
Test 2. 16.0|16.0
Test 3. 15.2|15.2
Test 4. 21.4|19.0
Nothing like extreme techs numbers. The only tests that changed besides the Nature numbers were the test 2 in default rez went up by .1fps and the game test 1 at 1600x1200 went down by .3fps. I'd say aside from the nature scores the other tests seem to be within the proper range.
Not sure what ET did but their results don't match mine at all.
, but it's not the same specs.