What GPU specification is crucial to the mining speed? I'm asking because I changed my AMD A8-7670K APU to an AMD A10-7870K now. A10 has 8 instead of 6 integrated R7 GPU cores, thus I expected distict shorter reading times. But it is still the same (about 40s for 152 TB, ~950 MB/s). Some specs:
8 vs. 6 R7 GPU cores
512 vs. 384 Shaders
866 vs. 757 MHz GPU Clock
887 vs. 581 GFLOPS
Looks like that GPU is not the bottleneck. I did not change anything else, still the same setup (19 drives, 8 GB RAM > 2 GB for R7 cores)
AMD A88X chipset is capable of 2 GB/s between SATA and APU and in addition 8 drives are connected by two USB 3.0 controllers (two times full USB 3.0 bandwidth) over PCIe.
Changes to 'chunkPartNonces=960000' make no difference.
I will try a dedicted GPU next days to see what happens to the bandwidth.
Found another tool to monitor CPU/GPU load: AMD system monitor is especially made for APUs to show the load distribution between CPU and GPU cores: http://support.amd.com/en-us/kb-articles/Pages/AMDSystemMonitor.aspx
This tool tells me, that jminer utilizes the CPU cores. WTF?!
This is either true and all other monitor tools are wrong or vice versa.
It somehow makes sense, that there is no performance gain from A8 to A10. -> There is allmost no difference in CPU performance between A8 and A10, the main difference between those two APUs is the GPU performance.
Will try to underclock, that should give me a true indication what is going on.
EDIT: Yep, AMD system monitor is wrong. Underclock from 3.90 GHz to 1.40 GHz without changing the GPU clock gives me something around 55s instead of 40s. There is some impact of the CPU clock to the mining performance but mining is done by GPU cores. Changing device-ID to CPU cores gives me muuuuch longer read times without load at GPU cores (GPU-z).
To sum this up: It's still not clear to me, why there is no performance gain with additional 2 GPU cores.










