Similarity Forum

General Category => General => Topic started by: JoeeoJ on August 09, 2013, 16:26:32

Title: Open CL 1 kernel ?
Post by: JoeeoJ on August 09, 2013, 16:26:32
Hi there,

did today my benchmarking with different work group sizes and am now a bit confused:

1. workgroup size has absolutly no impact  on the cpu's result: it's always at 624ms
2. Although the benchmark reports that only 1 core should be used, I can see in the taks manager that all my 6 cores are working at 100%

The GPU's benchmark results differ due to work group size, but the best result I got with work group size 2 : 1965 ms, the worst with 12433ms - 1024 workgroup size

I have an AMD Phenom IIX6 and a AMD Radeon 6800 GPU on an Asus M5A99x R2.0 Mobo

So how should I interpret this result ? Is the Gpu better because the CPU's result was created by all 6 cores or ?

TIA, Joe
Title: Re: Open CL 1 kernel ?
Post by: Admin on August 09, 2013, 20:20:47
Our OpenCL algorithms doesn't designed for CPU's, they more optimized for video cards, and we don't checked it on CPU OpenCL drivers. Similarity main algorithms already have very huge CPU optimization without OpenCL, we us Intel Performance Primitives Library in places requiring speed.