That chart's pretty weird though - look at the OpenCL score for the i5. The 4670S has the same HD 4600 (GT2) graphics as the 4770K mentioned in the article I linked, so the iGPU should have peak SP performance of 400 GFLOPS, more than half of the 737 GFLOPS for the 7850K, yet it performs MUCH slower in OpenCL than software-only. Clearly that benchmark isn't using the iGPU in the Intel chip at all! Which is consistent with Intel only giving access to the GPU with Direct3D (although I thought there was a Windows Intel OpenCL driver now?).
Nevertheless, it does look like HSA is necessary to get close to peak performance from the iGPU, as the speedup from "software" to "HSA" is approx the same as the ratio of CPU GFLOPS (118.4) to CPU+GPU GFLOPS (855.7). Plain OpenCL only gives it a small speed up.
Can't wait for more HSA benchmarks!
Edit: lol here are some I missed two weeks ago! http://www.techpowerup.com/forums/threads/amd-kaveri-review-focused-on-hsa-features.197158/ Nice to see it killing the i5.
Nevertheless, it does look like HSA is necessary to get close to peak performance from the iGPU, as the speedup from "software" to "HSA" is approx the same as the ratio of CPU GFLOPS (118.4) to CPU+GPU GFLOPS (855.7). Plain OpenCL only gives it a small speed up.
Can't wait for more HSA benchmarks!
Edit: lol here are some I missed two weeks ago! http://www.techpowerup.com/forums/threads/amd-kaveri-review-focused-on-hsa-features.197158/ Nice to see it killing the i5.
Last edited: