Quote from: Wolf0 on August 07, 2014, 09:34:22 PM

Quote from: 5w00p on August 07, 2014, 09:24:30 PM

Quote from: PeaMine on August 07, 2014, 04:35:20 AM

Quote from: tgt on August 06, 2014, 06:40:40 PM

It's fully utilizing the cores (minus one for some reason... maybe the scheduler thread is included in the thread limit?) so I can't see a huge gain other than running it on bare metal.

The opteron cores are not as efficient per ghz as the haswells. The only advantage is that you have physical cores without any hyperthreading. I'll try and compile the miner on 12.04 to run it on bare metal and see what/if any advantages are seen vs running it kvm with cpu passthrough.

The server is about $3300, so it's not exactly cost effective. Just fun to see.

edit: just noticed the instructions to install it on 12.04, derp.

http://i.imgur.com/iH7vs4r.png

Is that the 16 core Intel system or the AMD 48 core?

methinks its the Opteron machine.

Wolf0, thanks for sharing. I should not be surprised that the coder of the miner is very effective at managing his rented rigs, and I am not really, but I gotta say that your ~64 h/s/core is quite impressive on the c3.8xlarge instances.

Actually, I can get 70H/s/core on c3.8xlarge, 64H/s/core is just with this miner...

I get 64H/core on my opteron... the speed actually tops out at 26 cores (26*64=1664H/s same as in the screenshot). I guess that's L3 bandwidth limit - It's amusing to see how intel and amd are virtually identical per core due to this similar cache speed limitation. Pushing numactl to force workloads onto physical cores/cache pairs didn't make a difference.