Posts: 6.069
Threads: 68
Joined: May 2010
Reputation:
167
Location: Grenoble, France
I didn't say anything. Multithread is a very complex topic. GSdx MT is implemented with C++11 atomic and a ring buffer from boost (it used to uses windows primitive and STL). It yields a big increase of perf in linux. And the same performance level on Windows.
There is likely a bad interaction between VS2015 / GSdx MT / Skylake CPU. Maybe there is an option to generate fast code, instead of debug/safe code. Maybe there is a bug in skylake. Maybe my code isn't good. I don't know.
Posts: 1.689
Threads: 7
Joined: Apr 2013
Reputation:
66
A Skylake specific problem that happens when using hyperthreading? That sounds a lot like the Prime95 Skylake bug. Has anyone tried updating the BIOS - PC's bios, not ps2's - and check if that fixed it?
Posts: 5.076
Threads: 18
Joined: Oct 2010
Reputation:
154
I think it's the same bug. Did they already provide bios updates? I though they were at best still working on it.
Btw I think as in the Prime95 test it only manifests in AVX code. SSE4.1 could work without problems - the same if you disable hyper threading. I think Dr_hycodan validated that using his i7-6700.
Posts: 6.069
Threads: 68
Joined: May 2010
Reputation:
167
Location: Grenoble, France
04-05-2016, 04:50 PM
(This post was last modified: 04-05-2016, 05:34 PM by gregory.)
I think some people tested a new bios without success. Are you sure that it is working fine with SSE ?
Posts: 5.076
Threads: 18
Joined: Oct 2010
Reputation:
154
04-05-2016, 05:52 PM
(This post was last modified: 04-05-2016, 05:53 PM by willkuer.)
Maybe I am confusing it with GIMPS. I couldn't find the reference I had in mind from dr_hycodan. For Prime95 it is only observable using AVX and not using their FMA build.
@sandman332 can you please test if this problem happens in gsdx SSE2/3.1/4 as well? I think AVX and AVX2 behave similar (as they are probably both compiled with the /avx flag).