search

fancyIX / sgminer-phi2-branch

A branch of sgminer optimized with GCN cross lane instructions on AMD (ethash, phi2, lyra2Z[z], allium, x25x, lyra2REv2/v3, argon2d, yescrypt, neoscrypt, 0x10)
GNU General Public License v3.0
54 stars 27 forks source link

Investigate argon2d #158

Closed fancyIX closed 5 years ago

fancyIX commented 5 years ago

could implement argon2d250 and argon2d500 for amd gpu

this person was able to do it based off of ccminer-opt by jayddee.

https://bitcointalk.org/index.php?topic=5053255.0

https://github.com/bogdanadnan/multiminer

yuzi-co commented 5 years ago

even better one: https://github.com/duality-solutions/Dynamic-GPU-Miner-AMD

fancyIX commented 5 years ago

@yuzi-co thanks for the info. Seems like there are lots of shuffling. I can apply GCN cross lane instructions to speed it up. (only if I have time to work on it...)

yuzi-co commented 5 years ago

And if you can also add the 250 and 4096 variants ;) I tried modifying it for 250, but failed, beyond basic algo parameters, there must be something else

alucard20724 commented 5 years ago

i was just about to point out, but @yuzi-co got it for Dynamic. Dynamic is better. vega64 @ 155 khash vega56 @ 140 khash RX580 @ 90 khash RX570 @ 80 khash RX560 @ 35 khash each were just set with -g 2 parameter.

hopefully i'll have numbers for the new VII when they are available.

fancyIX commented 5 years ago

Not able to speed up by cross lane instructions, at least for now

fancyIX commented 5 years ago

@alucard20724 I modified Dynamic-GPU-Miner-AMD a bit by altering LDS shuffling with GCN cross lane instruction. My version of sgminer may be a little bit faster (or slower). Welcome to test

alucard20724 commented 5 years ago

@fancyIX speed on vega64 are impressive. @180 khash. I'll check the rest of the cards this weekend when i have more time

yuzi-co commented 5 years ago

@fancyIX, Can you add the 250 variant(crds coin)?

alucard20724 commented 5 years ago

@fancyIX for the rx580 i get @80khash and with the rx560, i get @32khash... all were done with just the -g 2 parameter.

Definately a big boost with the vegas. I'll test the vega56 later.... need to head to bed.

ok..more info. for the vega64. i have the mem clock at 1000mhz. over clocking the mem on the vega appear to help.

yuzi-co commented 5 years ago

I get a slight speed improvement on RX480 (94 to 98k)

fancyIX commented 5 years ago

@alucard20724 for 580 you need to wait a few minutes. It may ramp up to 90+