NVIDIA GF100 & Friends speculation

GTX480 benchmark:
25qw2lc.jpg

If you omit the Unigine benchmark the average performance difference is 26%!

Quite different than that 49%!
 
If you omit the Unigine benchmark the average performance difference is 26%!

Quite different than that 49%!
There are a lot more irrelevant numbers like Crysis being twice as fast as 7fps :p Looking at those numbers, the delta looks just like the previous generation.
 
The _prev instructions can be used only in the x and y lanes. The x lane uses the result from the z ALU and the y lane the one from w ALU.

One has basically paired ALUs x and z is one pair, y and w the other one.

Edit:
I just see that this wouldn't fit the description of the interpolation instructions. Guess I have to try to trick the compiler into generating those instructions and see what comes out. With all those small errors in the ISA docs I don't really trust what stands there. It wouldn't make much sense in my view to create just another datapath for the interpolation instructions which is basically already there. A swizzle of the xyzw designations in the interpolation description would align the functionality quite well with the _prev instructions.
Aha, I was just about to congratulate you on unravelling this, before I saw your edit. I guess the only thing to do is to see if the compiler will do these...

I find the MUL after MUL limitation peculiar and apart from how this meshes with DOT4 I also wonder how this meshes with double-precision. Oh well...

Jawed
 
And quite a bit higher than 5-10% :LOL:

True, but if you continue with the omissions, taking out those 2560X1600 scores (quite possibly framebuffer limited) you end up with even lower performance difference.

And no, I am not trying to make ATI look better, just paying attention to the resolution I game at (1920X1080).

I mean its just stupid to draw these kind of conclusions from three different resolutions. Each resolution should have its own average!

This is what I do in my benchmarks at least but I use one resolution anyway!:p
 
True, but if you continue with the omissions, taking out those 2560X1600 scores (quite possibly framebuffer limited) you end up with even lower performance difference.

False! Go see Hexus review of Shappire HD5870 Toxic 2GB. There is virtually no effect of one more Gigabyte on the card...
 
The last 2 numbers are idle and load power of the whole computer ? Because than percents are quite misleading.

Crap, I totally missed that! If the GTX 480 and the 5870 have 45W difference and since we know the 5870 idles at 27W then the GTX 480 idles at 72W?:oops:
 
25qw2lc.jpg
[/QUOTE]

20091030082631511.png


wtf?

Quakewar

1680x1050 4xAA 16xAF 9.10 drv: 171.0fps 10.3drv 163.0fps 480-168.0fps

1920x1200 4xAA 16xAF 9.10 drv: 146.0fps 10.3drv 139.0fps 480-140.0fps

2560x1600 4xAA 16xAF 9.10 drv: 97.0fps 10.3drv 95.0fps 480-94.0fps

Crysis

1680x1050 4xAA 16xAF 9.10 drv: 32.0fps 10.3drv 31.9fps 480-30.0fps

1920x1200 4xAA 16xAF 9.10 drv: 28.0fps 10.3drv 28.0fps 480-27.5.0fps

2560x1600 4xAA 16xAF 9.10 drv: 14.1fps 10.3drv 8.2fps 480-15.4fps

Farcry2

1680x1050 4xAA 16xAF 9.10 drv: 77.8fps 10.3drv 72.4fps 480-105.6fps

1920x1200 4xAA 16xAF 9.10 drv: 67.9fps 10.3drv 63.9fps 480-90.4fps

2560x1600 4xAA 16xAF 9.10 drv: 48.3fps 10.3drv 46.3fps 480-89.1fps

UT3

1680x1050 4xAA 16xAF 9.10 drv: 158fps 10.3drv 156fps 480-253fps

1920x1200 4xAA 16xAF 9.10 drv: 132fps 10.3drv 131fps 480-207fps

2560x1600 4xAA 16xAF 9.10 drv: 85fps 10.3drv 85fps 480-133fps

AMD should fire ATI Cat Team according to these two tests with the same
hardware configuration, lol:down:
 
And the cherrypicking is on a leak that could be fake (and also the other leaks with not-so-high difference between 480 and 5870 could be fake as well).
Only a few hours are missing, so I think it's a waste of force starting to argue now...
 
Weird number. Here's a Stalker Review using 4x AA at 1920x1200. http://www.legitreviews.com/article/1253/8/ The 5870 is getting 54.3 FPS where as in the leaked numbers, we're seeing 32.6? The diff being SSAO?

ffs, atleast bench games people are actually playing. BF:BC2, MW2, etc.
CoP Benchmark has 4 different modes, the last one, Sunshafts, is much much slower than the others:

http://www.pcgameshardware.com/aid,...hmarks-from-DirectX-9-to-DirectX-11/Practice/

Jawed
 
Back
Top