joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 23, 2018, 05:01:39 PM |
|
If someone really wants me to provide a special binary build I can be bribed. My tip jar has been empty for a while. I expected some consideration for the hard work implementing 4way, not to mention my quick response to real issues.
I can build all possible combinations of x86_64 architectures from SSE2 to AVX2 with and without AES or SHA if sufficiently motivated. I might even put some more effort into figuring out how to build for the older AMD CPUs.
|
|
|
|
|
Mr.Spider703
Member
Offline
Activity: 392
Merit: 27
http://radio.r41.ru
|
|
January 23, 2018, 09:50:13 PM |
|
thanks for the update, after it the speed on the m7m algorithm grew from 1.5 kh to 2 kh
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 23, 2018, 10:46:06 PM |
|
Testing x16r now, can't argue about fees.
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 24, 2018, 01:48:53 AM |
|
I don't own any AMDs si I don't have any personal experience. Based on the specs it should work well with AES and AVX.
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 24, 2018, 02:15:41 AM |
|
cpuminer-opt-3.8.0 released. https://github.com/JayDDee/cpuminer-opt/releases/tag/v3.8.04way no longer a seperate feature, included in AVX2. Added x16r algo for Ravencoin, anime algo for Animecoin. More 4way optimizations for X13 and up. Tweaked CPU affinity to better support more than 64 CPUs. Fixed compile problem on some old AMD CPUs. This release is a major milestone for cpuminer-opt. It essentially marks the end of 4way phase 1. 15 of the 17 functions in X17 have some form of optimization. Only fugue and whirlpool still use unoptimized code and 4 way seems impossible on these functions using SIMD.
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 24, 2018, 02:45:29 AM |
|
p.s. Add LTC or ETH wallet for donations. BTC fees nowadays are way too high Good idea, in my sig.
|
|
|
|
Mr.Spider703
Member
Offline
Activity: 392
Merit: 27
http://radio.r41.ru
|
|
January 24, 2018, 05:20:15 AM |
|
cpuminer-opt-3.8.0 released. https://github.com/JayDDee/cpuminer-opt/releases/tag/v3.8.04way no longer a seperate feature, included in AVX2. Added x16r algo for Ravencoin, anime algo for Animecoin. More 4way optimizations for X13 and up. Tweaked CPU affinity to better support more than 64 CPUs. Fixed compile problem on some old AMD CPUs. This release is a major milestone for cpuminer-opt. It essentially marks the end of 4way phase 1. 15 of the 17 functions in X17 have some form of optimization. Only fugue and whirlpool still use unoptimized code and 4 way seems impossible on these functions using SIMD. great news, downloaded and went to test, thank you very much for the update
|
|
|
|
nizzuu
Full Member
Offline
Activity: 187
Merit: 100
Cryptocurrency enthusiast
|
|
January 24, 2018, 06:18:02 AM |
|
joblo, please add DOGE donation adress as well.
|
|
|
|
4ward
Member
Offline
Activity: 473
Merit: 18
|
|
January 24, 2018, 11:24:10 AM |
|
cpuminer-opt-3.8.0 released. https://github.com/JayDDee/cpuminer-opt/releases/tag/v3.8.04way no longer a seperate feature, included in AVX2. Added x16r algo for Ravencoin, anime algo for Animecoin. More 4way optimizations for X13 and up. Tweaked CPU affinity to better support more than 64 CPUs. Fixed compile problem on some old AMD CPUs. This release is a major milestone for cpuminer-opt. It essentially marks the end of 4way phase 1. 15 of the 17 functions in X17 have some form of optimization. Only fugue and whirlpool still use unoptimized code and 4 way seems impossible on these functions using SIMD. for some reason X16r shows about 4x the hashrate of the official cpuminer (by Tpuvot), while actually delivering lower hashrate and sharerate on the pool side p.s. dropped a penny in the ETH jar
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 24, 2018, 02:14:26 PM Last edit: January 24, 2018, 02:31:27 PM by joblo |
|
cpuminer-opt-3.8.0 released. https://github.com/JayDDee/cpuminer-opt/releases/tag/v3.8.04way no longer a seperate feature, included in AVX2. Added x16r algo for Ravencoin, anime algo for Animecoin. More 4way optimizations for X13 and up. Tweaked CPU affinity to better support more than 64 CPUs. Fixed compile problem on some old AMD CPUs. This release is a major milestone for cpuminer-opt. It essentially marks the end of 4way phase 1. 15 of the 17 functions in X17 have some form of optimization. Only fugue and whirlpool still use unoptimized code and 4 way seems impossible on these functions using SIMD. for some reason X16r shows about 4x the hashrate of the official cpuminer (by Tpuvot), while actually delivering lower hashrate and sharerate on the pool side p.s. dropped a penny in the ETH jar Thanks for the tip. Your report about hash rate is scary because it's hard to verify. x16r by it's nature has a very volatile hash rate and Suprnova has been notorious for displaying incorrect hash rates. There is no difference in the miner-side hashrate calculation for x16r. I made a change for 4way to account for 4 nonces per iteration but it applies to all 4way algos and seems to be accurate. It'll need more data from other users to determine if there is a problem with hashrate calculation. Edit: hash rate for x16r is more volatile with opt vs multi due to the mix of optimized and unoptimized hash functions. X16r can theoretically run the same function 16 times. If it's an unoptimized function the hash rate with opt wil be the same as multi. But if it's a highly optimized function the hash rate gain will be higher than average.
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 24, 2018, 03:42:48 PM Last edit: January 24, 2018, 04:17:07 PM by joblo |
|
joblo, please add DOGE donation adress as well.
This is a joke, right? DOGE is a joke coin. I don't have a ton of wallets installed, I try to keep everything in BTC but with fees as they are it's becoming difficult to transact. I chose ETH as a backup because of it's market cap and trading volume. I could add BCH and BTG (I installed wallets to claim coins from BTC fork) or I could dust off my old Darkcoin wallet (it was still Darkcoin last time I used it). I could also add LTC or XMR. But DOGE??? Seriously??? Edit: I added LTC, BCH and BTG donation addresses to OP.
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 24, 2018, 04:09:18 PM |
|
The plan forward.
With 4way phase 1 complete it's time to look at phase 2.
Phase 1 focussed on the SHA3 family of chained algos such as x11 (even though they are now ASIC dominated)
This was a relatively easy path that allowed me to sharpen my SIMD coding skills. Now things get more complicated.
There is still some improvement available in the same algos by converting some of the SSE2 functions to use 4way, such as Luffa and Simd. They won't produce the same level of improvement due to their existing SIMD code but they set the stage for AVX512 and 8way. (AVX512 is likely phase 3 or 4). IT also allows me to continue to learn new things before tackling the bigger challenges.
Those bigger challenges include some algos that are more CPU friendly but they will require a complete rewrite. I've done some exploratory work with one unnamed algo. There are some data dependent memory accesses (a killer for SIMD instructions (not the algo)) but they don't seem to be a dominant part of the algo. We shall see how it goes but it will take time.
|
|
|
|
nizzuu
Full Member
Offline
Activity: 187
Merit: 100
Cryptocurrency enthusiast
|
|
January 24, 2018, 07:31:57 PM |
|
I could also add LTC or XMR. But DOGE??? Seriously???
Yep, seriously As for XMR, those fees are not good, too. I just offered though, you may accept or not Many people do not donate because of the high fees, withdraw limits etc, that's where such coins like DOT, DOGE, XVG, DGB and similar have no limits. Read "add some actively traded high-supply coins wallets together with hodl coin wallets".
|
|
|
|
Nana99
Newbie
Offline
Activity: 126
Merit: 0
|
|
January 24, 2018, 07:52:39 PM |
|
Configure command line?
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 24, 2018, 09:09:25 PM |
|
Configure command line?
RTFM
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 24, 2018, 09:18:22 PM |
|
Read "add some actively traded high-supply coins wallets together with hodl coin wallets".
I don't understand.
|
|
|
|
zazaxxx
Newbie
Offline
Activity: 3
Merit: 0
|
|
January 24, 2018, 09:48:52 PM |
|
Any plans to add 4way to yescrypt?
|
|
|
|
Mr.Spider703
Member
Offline
Activity: 392
Merit: 27
http://radio.r41.ru
|
|
January 24, 2018, 10:00:17 PM |
|
joblo, please add DOGE donation adress as well.
This is a joke, right? DOGE is a joke coin. I don't have a ton of wallets installed, I try to keep everything in BTC but with fees as they are it's becoming difficult to transact. I chose ETH as a backup because of it's market cap and trading volume. I could add BCH and BTG (I installed wallets to claim coins from BTC fork) or I could dust off my old Darkcoin wallet (it was still Darkcoin last time I used it). I could also add LTC or XMR. But DOGE??? Seriously??? Edit: I added LTC, BCH and BTG donation addresses to OP. you can create a purse in https: //ru.cryptonator.com it is multi, there is all of the above, put auto-exchange on ETH and then you can output to your wallet. P.S. do not forget to turn on 2FA
|
|
|
|
joblo (OP)
Legendary
Offline
Activity: 1470
Merit: 1114
|
|
January 24, 2018, 10:25:06 PM |
|
Any plans to add 4way to yescrypt?
I deliberately did not mention the algos that I'm looking at because I have no idea if I can do anything with them yet.
|
|
|
|
|