SiSoft Sandra 2009
- GPGPU Processing |
Source: Sandra |
|
Sisoft's
Sandra's GPGPU (general purpose graphics processing unit) benchmark analyzes the
ability of the GPU to perform Mandelbrot equations to generate fractals. These
same types of equations are used in tasks like image processing, video encoding
and scientific simulations, giving an index of how well a GPU can perform in
this test.
SiSoft Sandra 2009 ver
15.72 - GPGPU Processing |
Native Float Shaders: (Mpixel/s) |
Points |
Ranking |
Gigabyte
GV-R587SO-1GD |
1000 |
|
ASUS EAH5850 DirectCu Top/2DIS/1GD5 |
744.74 |
|
ASUS ENGTX480 2DI/1536MD5 |
744.35 |
|
Native Double Shaders: (Mpixel/s) |
Points |
Ranking |
Gigabyte
GV-R587SO-1GD |
556.2 |
|
ASUS EAH5850 DirectCu Top/2DIS/1GD5 |
398.71 |
|
ASUS ENGTX480 2DI/1536MD5 |
151.8 |
| |
The Gigabyte GVR587SO-1GD videocard has what it takes to
make it as a GPGPU device, with better scores overall for both its
floating-point shader performance and its double precision shaders performance
as well. Given that this is some very new technology its likely that NVIDIA's
lack of performance here is due to a less than optimal test configuration,
rather than a failure of its CUDA technology.
SiSoft Sandra 2009
- GPGPU Memory Bandwidth |
Source: Sandra |
|
The
second part of Sisoft's Sandra's GPGPU measures the videocard's memory
bandwidth, which will directly affect the overall performance of GPGPU
applications.
SiSoft Sandra 2009 ver
15.72 - GPGPU Memory Bandwidth |
Internal Memory Bandwidth: (GB/s) |
Points |
Ranking |
Gigabyte
GV-R587SO-1GD |
111 |
|
ASUS EAH5850 DirectCu Top/2DIS/1GD5 |
87.63 |
|
ASUS ENGTX480 2DI/1536MD5 |
114 |
|
Data Transfer Bandwidth: (GB/s) |
Points |
Ranking |
Gigabyte
GV-R587SO-1GD |
4.73 |
|
ASUS EAH5850 DirectCu Top/2DIS/1GD5 |
4.72 |
|
ASUS ENGTX480 2DI/1536MD5 |
5 |
| |
The numbers here are much more in favor of the Geforce
GTX 480, which edges out the Gigabyte GVR587SO-1GD's overclocked memory
performance. The Gigabyte GVR587SO-1GD has faster GDDR5 memory (5GHz vs.
3.7GHz), but a narrower memory bus interface (256-bit vs. 384-bit), which tips
the scales in NVIDIA's favor.