ShaderToyMark + HD 6850 (960 and 1120 shaders processors), source
As you may know, some Radeon HD 6850 have been sent to the press with 1120 scalar processors instead of 960. KitGuru has published a comparative article between HD 6850 with 960 shaders processors and HD 6850 with 1120 shaders processors.
A radeon HD 6850 with 1120 shader processors
1120 shader processors means 160 scalar processors in more or an increase of shader computing power of 16%.
ShaderToyMark is a pure and really intensive pixel shader test. Most of shaders used in ShaderToyMark are math shaders (few or not texture fecthes at all). ShaderToyMark is nearly a GPU computing application!
And guess what? There is exactly an increase of 16% between both versions of the HD 6850 (see the chart above).
ShaderToyMark: HD 6850 with 960 shaders score
ShaderToyMark: HD 6850 with 1120 shaders score
In TessMark 0.2.2 (OpenGL 4 tessellation benchmark), it’s another story. In normal tessellation mode (which correspond to a tessellation amplifation factor of X16) the difference is around 14% which sounds normal. But in moderate tessellation (factor X8), the difference jumps to 36%. Don’t ask me why, I currently have no HD 6800 series in hands so I can’t test TessMark to see how it behaves with HD 6850 / HD 6870.
From TessMark comp. table, here are the score of some Radeon HD 5000 series in moderate tessellation:
– Radeon HD 5770 (MSI R5770 Hawk, 800 shaders, gpu core:875MHz): 20366 points
– Radeon HD 5870 (1600 shaders): 27000 points
I just received from Zardon, the guy behind KitGuru, the TessMark scores for a Radeon HD 6870 (1120 shaders, gpu core: 900MHz):
– Moderate tessellation: 23199 points
– Normal tessellation: 8342 points
The HD 6850 has a GPU core clock of 775MHz. Let do some linear math for the moderate tess level:
- a Radeon HD 6850 1120 shaders with a core clock of 875MHz would have a score of around 21100 points: the score is a bit above the MSI R5770
- a Radeon HD 6850 1120 shaders with a core clock of 900MHz would have a score of around 21700 points: the score is not too far from the HD 6870
In normal tess level, MSI R5770 Hawk gets a score of 7515 points, whereas the HD 6870 scores 8342 points.
- a Radeon HD 6850 1120 shaders with a core clock of 875MHz would have a score of around 7990 points: the score is a bit above the MSI R5770
- a Radeon HD 6850 1120 shaders with a core clock of 900MHz would have a score of around 8220 points: the score is not too far from the HD 6870
At first sight, the scores of the Radeon HD 6850 with 1120 shaders look normal. Actually, seems there’s a problem in the TessMark score of the 6850 HD with 960 shaders. This score should be higher like around 16000 points in moderate tessellation. But I stop here because I don’t have a HD 6850 nor HD 6870. Maybe there’s a bug in TessMark, or in the OpenGL driver of Catalyst 10.10 or in the measure… If someone has a clue, the comment section is there for you!
Can bi BIOS used from 6870, just with some downclocks maybe?
The tessellation difference is because for such synthetic tests like TessMark, the bottleneck can be in the fixed function tessellator component. As according to AMD press release the HD6000 series has double the ammount of tessellation throughput than the HD5000 series, it may mean that for the different shader configs they put different sized tessellator unit.
Well I think I’ll wait for a proper article on this. KG has a reputation for well… ya know. I also think I saw on another site that they were one of the sites that cocked up the 6850 mixup, which is not how they’re portraying it now?!
@ Thud, Check your facts. All of the published KG reviews regarding the 6850 Are using the card variations with the 960 Unified shaders.
Thanks to JeGX For publishing this. It hopefully Will help clear things up.
Thud, you are clearly confusing their sometimes controversial news with their reviews, which are great.
Zardon has been honest with the reviews and has shown that all their reviews are using 960 shader 6850’s, the results in this comparison actually call back to the other reviews and you can see the differences clearly. All looks good to me.
@ Fish and Eric
Facts? Not a word normally associated with Kitguru. Here is how people on
the forums i visit see it.
Kitguru published their Sapphire article, their GPUz showed 1120 shader
processors and the performance of that card. They then became aware of the
shader issue (after launch as there was no mention in the original article)
and photoshopped the GPUz uploading a new pic.
Problem is that the performance in the article was wrong. Look at
overclocking for example, clearly done on the Sapphire 1120 card as it shows
the 6850 OC beating the 6870 which only happens on the enhanced model. On
top of that, the GPUz on that page shows the 1120 shaders still (oops!).
What part about this mentions the shader error?: “These increases are huge
as the figures show – with the clocks at 950MHz core and 4560MHz effective
on the memory, the HD6850 is outscoring the reference clocked HD6870.”
What makes it worse is that the results were clearly copied over into the
XFX 6850 review where the “960” card matches the “1120” for performance. i.e
past the 6870 level.
Face it, Kitguru messed up and then tried to cover their tracks, badly. But
it’s not the first time as most enthusiasts know. A real laughing stock that
site and they are not fooling anyone. Maybe if they actually spent less time
faking review comments and tested all the products they received people
would have more respect.
Here you can see the GPUz overlayed, and how they positioned it just out of skew with where it would have been, and also forgot to change the filrate.
http://i.picasion.com/pic32/a796ab1361b9f09299590c95ae9f31c8.gif
Original: http://www.kitguru.net/wp-content/uploads/2010/10/gpuz4.png
Photoshopped: http://www.kitguru.net/wp-content/uploads/2010/10/gpuz6.png
Catalyst 10.10a MLAA enabled + XFX HD5770
121 points 20fps 60seconds
960×540
no AA
no fullscreen
it has some thing missed with the performance of HD6850?
se it guys,http://img264.imageshack.us/i/hd5770.jpg/
thanks jeGX
Too many conspiracy theories for my liking.
I always rely on my own testing for accuracy.
I bought a HD6850 on friday, and it arrived yesterday. I ran my own tests (its an XFX board with 960 shaders, sadly!).
My 3dmark vantage score with a 920 core i7 overclocked was 12578 which looks identical to this score http://www.kitguru.net/components/graphic-cards/zardon/the-amd-hd6850-1120960-shader-debate-comparisons/3/
which also looks identical to this one http://www.kitguru.net/components/graphic-cards/zardon/sapphire-radeon-hd6850-and-hd6870-review/6/
I also ran heaven benchmark scores and got 29.4 fps average which looks identical to this http://www.kitguru.net/components/graphic-cards/zardon/the-amd-hd6850-1120960-shader-debate-comparisons/4/ which also looks identical to this http://www.kitguru.net/components/graphic-cards/zardon/sapphire-radeon-hd6850-and-hd6870-review/5/
As I said, ill rely on my own testing, but ive read all those articles and they look fine to me. I could go on, but whats the point 🙂
Hi jeGX,
I hope this helps.
My HD6850 arrived and my results match exactly what KitGuru posted for Tessmark and Shader Toy.
I also get the same for 3dmark Vantage which is a shame as I see the feature tests are much higher with a 1120 core HD6850 🙁
What a shock, the gpuz screenshot disappears from Kitguru and then, completely randomly, some users post here who just happen to have bought the card, happen to have read this article and happen to have the same scores (and specs?) as Kitguru… you couldnt make it up. Actually you could because they just did!
Thud,
What? we aren’t allowed to buy a HD6850? the 3dmark Vantage scores are almost universal, its pretty easy to work out and other sites have posted the same feature test scores with various core I7’s. Seems really simple to me 🙂
I guess its ok to just spout hatred around the net, not sure what your deal is. relax man, its only technology.
The joys of the internet are that there are always tons of options to read other sites, but I guess some people get banned and go on personal vendettas!
Guys, to CLOSE THE DEBATE, I post Zardon’s position about the story:
“The 6850 cards used in the kitguru reviews all had 960 shaders, this is why the results between our sapphire and Xfx reviews are identical. If one of the cards had 1120 shaders the results would be different between Reviews.
We alerted sapphire to this shader issue and we tested with another card for performance results. We have several reference boards and other partner boards so this was not an issue. Systems between reviews are identical.
The gpuz shots were altered in a rush to stop other sites and readers emailing us informing us that we had tested a card with the wrong shader count when in fact we hadn’t. In hindsight we could have put a massive disclaimer saying that we tested another board with games and synthetic testing in the sapphire review but we felt this would have caused more confusion at the time. This was an error on our part, but is nothing unethical in regards to testing – obviously as this is the Internet a few troublemakers latch onto these things and make a meal out of it. Incorrect gpuz shots do not negate results of a weeks solid testing.”
Pingback: 結局、ビデオカードはGeForceとRadeonどっちがいいんだよ | ツンダオワタ情報