Re: SRBMiner Cryptonight AMD GPU Miner V1.8.2 - native algo switching
V1.8.2
- Upx and litev7 algos on auto settings now have more hashrate or lower power consumption, or both
- Reworked networking code, so probably get rid of a bug or two
- Another try to fix V4 algo crashing
- Maxnosharesent parameter default value is now 45 minutes
- Fixed MarketCash , it was broken in 1.8.1
+ Got some speedup on litev7 and upx algos if you use auto setup, mainly on Vegas and weak Rx550 - 560 cards, also power consumption dropped.
+ Another try to fix/postpone the crash that is happening on V4 algos. I would suggest to create/compile the kernel on 18.6.1 driver, then move to Adrenaline 2019 driver (19.3.2 for example).
Some users reported none/less crashes on newer drivers.
+ Fully rewrote the networking code so now :
- You will get 'Block expired' errors from some pools
- Lot's of rejects on Webchain algo
To explain : Some pools like Nicehash or Nanopool, do not accept stale shares (shares sent after a new job was dispatched). In earlier versions of SRB miner did not send these shares to the pool when it detected that a new job arrived in the mean time. Now these are sent too. I made this decision because i saw some discussions where people say that SRB displays higher hashrates than they really are, or steals hashrate (
). This way user can see all shares - accepted and rejected, and can have a clearer picture of his real speed.
Webchain is a similar story, it rejects stale shares. And oh, there are a lot of those, because it has a short blocktime so new jobs just bam bam bam keep coming.
That's why you have 30% lower hashrate poolside than in miner. Yes i know Loud miner does not have this 'problem' so please use it instead if you have better results.
- Upx and litev7 algos on auto settings now have more hashrate or lower power consumption, or both
- Reworked networking code, so probably get rid of a bug or two
- Another try to fix V4 algo crashing
- Maxnosharesent parameter default value is now 45 minutes
- Fixed MarketCash , it was broken in 1.8.1
+ Got some speedup on litev7 and upx algos if you use auto setup, mainly on Vegas and weak Rx550 - 560 cards, also power consumption dropped.
+ Another try to fix/postpone the crash that is happening on V4 algos. I would suggest to create/compile the kernel on 18.6.1 driver, then move to Adrenaline 2019 driver (19.3.2 for example).
Some users reported none/less crashes on newer drivers.
+ Fully rewrote the networking code so now :
- You will get 'Block expired' errors from some pools
- Lot's of rejects on Webchain algo
To explain : Some pools like Nicehash or Nanopool, do not accept stale shares (shares sent after a new job was dispatched). In earlier versions of SRB miner did not send these shares to the pool when it detected that a new job arrived in the mean time. Now these are sent too. I made this decision because i saw some discussions where people say that SRB displays higher hashrates than they really are, or steals hashrate (
). This way user can see all shares - accepted and rejected, and can have a clearer picture of his real speed.Webchain is a similar story, it rejects stale shares. And oh, there are a lot of those, because it has a short blocktime so new jobs just bam bam bam keep coming.
That's why you have 30% lower hashrate poolside than in miner. Yes i know Loud miner does not have this 'problem' so please use it instead if you have better results.
new version is great
6xvega64 GPU@1447/memory@1100 (-+1300w)
drivers 18.5.1
hashrate ~12,4KH/s (2.5% rejected shares on nanopool)
looks rock solid after 12 hours of running (last version run hardly 4 hours)
great work doc!!!