Beginner question,
I could install Tromp solvers to my 16 core server and it seem like mining now with it.
How can I benchmark the performance of mining? Thank you for make this open sauce and your effort.
1 Like
Following a conversation with judge solardiz in which I was explaining the use of two-stage bucketing, I realized my stage-two implementation could be optimized by linking, rather than listing, xor-able slots together.
This gives a nice speed boost in all my solvers, as commited in the repo now.
ALSO, I was able to implement the AVX2 4-way parallel blake2b using intrinsics rather than assembly. Although not as fast, this is a more portable approach that gets about 2/3 of the speed improvement.
I’ve split out the make targets for avx2. As shown in my github repo’s README, we now have
equi1: 4.6 Sol/s
eqavx21: 5.9 Sol/s
dev1: 6.5 Sol/s
equi -t 8: 13.7 Sol/s
eqavx2 -t 8: 16.7 Sol/s
dev -t 8: 17.2 Sol/s
8 x eqavx21: 20.3 Sol/s
8 x dev1: 20.6 Sol/s (note shrinking assembly advantage)
eqcuda: 23.6 Sol/s
And for (144,5) parameters:
eq1445 -t 8: 1.0 Sol/s
eq1445avx2 -t 8: 1.2 Sol/s
eqcuda1445: 2.2 Sol/s
6 Likes
Compliments ! Your results are really greats 
Thanks. Now all that’s left to do is improve code comments…
Will this last solver version be implemented in mainnet Zcashd miner ? Until now I always solo mined on testnet with Zcashd and your solver activated, I’m looking forward to test the “advanced” version ! 
1 Like
I’m sure it will eventually, hopefully within a week.
But probably not in time for launch.
1 Like
@tromp: seems like @nicecashdev is not going to release the source of the AVX2 enabled miner after all. Can you see with @str4d if he has the time to add your last optimized CPU solver to his standalone miner?
We will release the source. We are obligated to do so
@nicehashdev, the question isn’t really if you are going to release the source, but rather when. It’s been like 5 days now you released that windows binary, and we still don’t have anything for Linux. As you know the network is launching in 6h or so, and in current situation, Windows folks have a 100% advantage over Linux folks, in spite of the fact most “serious” miners are all using Linux. Obviously that’s not a fair and healthy situation.
Integration of AMD support had priority and we still don’t have linux compilable solution yet.
I understand, it’s totally your call when you want to release your stuff, that’s why I’m asking @tromp if he can give his stuff to @str4d so that we can at least have an AVX2 enabled CPU miner available for launch…
Tromp’s stuff is all public and everyone can integrate it into a miner
@str4d has been doing the integration so far, and has the next most usable CPU miner available, so it makes sense to ask him. If that doesn’t happen someone else will step in.
Latest commit adds an option
-x <140_byte_header_nonce_in_hex>
as an alternative to the
-h <ascii_header> -n <32-bit_nonce>
options, in order to allow for solving arbitrary binary headers.
Really great project!
Did you have any plans to integrate my latest solvers into zcashd, @sarath-hotspot ?
1 Like
Hey Tromp, is there a way to allocate each miner a certain amount of RAM for better speeds?
Speed is already optimal at only 144MB per instance…
Tromp, have you seen the guy who previously coded branches of lbc miner now reporting he’s developed an equihash nvidia miner getting 120 sol/s (with screen shots) on gtx 1070?
Someone else also mentioned
"I’m getting 100sol/s on a run of the mill quad core laptop because I cleaned up the code in the miners and rebuilt with better flags. "
Might be outright lies but just mentioning it if that helps.