I’ve decided to keep the AVX2/assembly dependent code in dev_miner.h,
while keeping equi_miner.h portable.
So if you have a Haswell or later intel CPU, and have fasm installed, try
make dev1
time ./dev1 -n 255 -r 100
to see the speed boost for yourself.
On a 4Ghz i7, I’m seeing times dip below 31s;
that’s over 6 Sol/s!
I just had a run of 30.093s, or 6.2473 Sol/s 
Again, a big thanks to xenoncat who made this speedup possible!
Impressive, very good progress.
very great tromp and xenoncat, do you think someone could merge into standalone-miner before 28 launch?
The solver prior to these optimizations is already merged into this standalone miner GitHub - str4d/zcash at standalone-miner (and I assume if these changes are merged into master before launch they will also be in the standalone miner).
The progress here is awesome!
There will at least be a 13.5H/s GPU miner PR made to zcashd as well for a gpu mining flag before launch. If we are lucky we might be more around 28H/s at launch. Thanks to tromp and xenoncat for making it possible.
Im having trouble to run the miner on linux, after clone and run “make all” what i need to do?
im running bash comand line on windows10
Someone can help me?
I whant to try the cuda and the CPU
Can someone expalin the connection between H/s and Sol/s ?
I’m a bit lost 
1 H/s is the same as 1 Sol/s (just be aware that some may be using a different metric for H/s which is where the confusion is coming from). But in this thread you can consider 1 H/s = 1 Sol/s.
@tromp did you run 8 devequi instances simultaneously?
Hi! That’s awesome. Will the zcashd w/ gpu mining flag on work with both AMD and Nvidia GPUs? If it is OpenCL, I was under the impression it would be a poor performer on Nvidia as compared with a CUDA implementation. If that turns out to be true, is there a full open source mining client that uses @tromp’s CUDA implementation? He has released solvers only, is there a client that uses them/will use them?
Sorry for the many questions, thank you for all your work.
Somebody implemented the solvers for OpenCL and Cuda, Windows and Linux
equidev is currently non functional.
did you mean running 8 dev1 simultaneously?
I just tried that and it runs at 25.6 Sol/s on the 4Ghz i7
Seems to be a very slow implementation though compared to the closed source ones.
Will the Tromp Cuda solver make it to the 1st release ?
btw, sterling effort to for getting the new-improved CPU solver into rc2 and many thanks !
Do your CPU optimizations in any way apply to GPU miners or is it CPU bound only?
I haven’t tried porting the latest micro-optimizations to CUDA yet,
but it’s not clear they would be of benefit there anyway.
Also there’s no need for faster blake hashing on CUDA as it
already spends a small percentage of time on it (12ms).
Having worked on both the CPU and GPU miner, what is your opinion on the speed difference between the latest gen. CPU vs latest gen GPU ? 2 weeks ago Toomim Bros claimed a 100x diff, has anything changed ?
also this is an OpenCL implementation only, there is no CUDA implimentation. Nvidia can run OpenCL but its optimized for CUDA; AMD gpus cannot run CUDA but tend to run OpenCL better than Nvidia gpus (to the best of my knowledge).
I think two weeks ago, toomim was refeering to the original standart miner. Thing have improved since then and the difference seems to be *2 or *3. Maybe *5 at most from what I red here and there.
Actually someone on bitcointalk whos really switched on says that 1 sol/s = .5h/s