Here is a new rough-and-ready GPU computing tool that comes to us from China. GPCBenchmark is an OpenCL based benchmark that evaluates the performance of OpenCL capable devices with a collection tests: global and local memory bandwidth, single and double precision floating point performance, common mathematics operations (256×256 matrix multiplcation – ouch!), image processing and encryption (SHA-1).
Here are some results. My system: Core i7 960 + GT 240 + GTS 250.
SHA-1 test:
– CPU: 34.752 Million loops / sec.
– GTS 250: 122.666 Million loops / sec.
– GT 240: 81.496 Million loops / sec.
Local memory bandwidth test:
– CPU: 182.966 GB / sec.
– GTS 250: 742.682 GB / sec.
– GT 240: 421.841 GB / sec.
On some tests like single and double precision floating point or common mathematics, I got no results after several minutes.
with my 4580 SHA-1 test: 288.000 Million loops / sec.
I got +- 352 million loops / sec with my hd 4870 512mb/256bits 780/1000MHz, using a Q6600 @2.4GHz at the SHA-1 test!
On my stock Core i7 860 ram @ 1600Mhz and stock 5850 I got,
SHA-1 test:
– CPU: 48.500 Million loops / sec.
– Cypress: 585.942 Million loops / sec.
Local memory bandwidth test:
– CPU: 135.664 GB / sec.
– Cypress: 237.277 GB / sec.
I wonder why the GTS250 and GT240 got higher score. The 5850 has more bandwidth, that’s a fact.
CPU
—————————————-
Global Memory: 16.7
Local Memory: 19.4
Int32 Ops: 12.3
Float Ops: 5.9
Double Ops: 23.7
Common Mathematics (Float): 32.0
Common Mathematics (Double): 55.8
Image Processing: 26.4
Cryption: 15.3
Total: 207.5
GeForce GTX 480
—————————————-
Global Memory: 170.2
Local Memory: 129.3
Int32 Ops: 293.5
Float Ops: 154.3
Double Ops: 210.2
Common Mathematics (Float): 626.6
Common Mathematics (Double): 932.6
Image Processing: 615.1
Cryption: 201.9
Total: 3333.9
Is it only me, or the website really has disappeared altogether? (Even the chineese version)