AMD Radeon HD 7660D On Linux
Phoronix: AMD Radeon HD 7660D On Linux
In my Linux preview article to the AMD A10-5800K "Trinity" APU I shared some initial benchmarks of the integrated Radeon HD 7660D graphics and compared those results to various NVIDIA and AMD discrete graphics cards. In this article are more Radeon HD 7660D Linux graphics benchmarks. In particular, there are results from the A10-5800K graphics when overclocking the GPU core along with benchmarks comparing this high-end Trinity APU to last year's A8-3870K "Llano" APU with Radeon HD 6550D graphics.
Why do the default Tropics numbers differ from the overclocking baseline?
Ie, page 4, trinity tropics 1280x1024 is at 24fps.
Page 6, at the overclocking baseline 800Mhz, same resolution, 29fps?
You should post more details regarding the RAM in future, especially when talking about APUs.
Faster ram = more fps.
I would like to see how these numbers match up against Win7 marks.
So for the not-particularly-GPU-intensive open source games Trinity had a nice bump in framerates. Then you get to the Unigine benchmarks that actually use real GPU features from more recent versions of OpenGL and Trinity either loses to or barely keeps up with the Llano.... any reason for this? Frankly, I find the Unigine benchmarks to be a much more accurate indicator of the real GPU strength of these chips. Under Windows I think that Trinity wins in these benchmarks vs. Llano, so what is the issue holding up Trinity here (drivers... again?)?
There's something a bit odd with the Unigine numbers (see curaga's post above). Unless I'm missing something, the first Unigine Tropics graph (Llano vs Trinity at 1280x1024) showed the two running at about the same speed...
Llano - 24.41
Trinity - 24.84
... while the second graph (Trinity only, different clocks, 1280x1024) showed 29.64 at default clocks (considerably faster than Llano).
Can't do the same cross-check with Sanctuary, unfortunately, because the screen res is different between the graphs.
I think there are still a lot of issues with the virtual memory code - which i think was turned off by default in the 9.0 release, but left on in git master.
Perhaps that is the issue?
This test was with fglrx.
Originally Posted by smitty3268
Some usefull data, but games.
I would like to see kernel compilation times or something else usefull , but games.