[Dev] NeoScrypt GPU Miner - Public Beta Test
-
I am also following Wolf0’s great work. Thanks again. ++ Karma points to Wolf0.
-
try lowering core clock to 1100 and run w 48 i 15 g2 on 280x/7970 getting between 130 and 150 per card
sgminer stable overnigt and still going
clocks same as i used with cgminer
-I 15 not working, some kernel error( too low memory), im using -I 14 w 48 g2 135khs
-
sgminer stable overnigt and still going
clocks same as i used with cgminer
-I 15 not working, some kernel error( too low memory), im using -I 14 w 48 g2 135khs
With SGMiner, you may now set thread concurrency.
-
With SGMiner, you may now set thread concurrency.
YAY!!! More settings!!! lol
-
Finally got it compiled and could test a bit :)
Two things I noticed:
- on a 64Bit compile the A:xxx Value in the first line of the Display reads: A:22323232343463467722132415…
this looks like a 23bit overflow and happened on the first versions of cgminer-neoscryp, too
- Stratum doesn’t work for me, longoll works fine
errormessage: ‘message received: unknown method’
I run against a p2pool, that is working fine for cgminer.
- hashrate slightly increased on my HD 8400 GPU ( I know, I know, it’s just a laptop GPU.
-
Finally got it compiled and could test a bit :)
Two things I noticed:
- on a 64Bit compile the A:xxx Value in the first line of the Display reads: A:22323232343463467722132415…
this looks like a 23bit overflow and happened on the first versions of cgminer-neoscryp, too
- Stratum doesn’t work for me, longoll works fine
errormessage: ‘message received: unknown method’
I run against a p2pool, that is working fine for cgminer.
- hashrate slightly increased on my HD 8400 GPU ( I know, I know, it’s just a laptop GPU.
Use my SGMiner 5 compile for neoscrypt.
-
Use my SGMiner 5 compile for neoscrypt.
earlier was 4.2.2 version, is there another outhere?
-
earlier was 4.2.2 version, is there another outhere?
It’s actually SG5 - the dev branch is misnamed.
-
earlier was 4.2.2 version, is there another outhere?
Wolf0’s compile is a development snapshot of sgminer5. Every development snapshot of sgminer I have used has had the version as 4.2.2 even though it’s really version 5.
Edit: Wolf beat me to it
-
Use my SGMiner 5 compile for neoscrypt.
I used this link: https://github.com/sgminer-dev/sgminer
Do you have forked it?
-
I used this link: https://github.com/sgminer-dev/sgminer
Do you have forked it?
No, that’s the link - use the develop branch.
-
That’s exactly what I did:
git clone --recursive -b develop git(at)github.com:sgminer-dev/sgminer.git
configure
make all / make install
-
That’s exactly what I did:
git clone --recursive -b develop git(at)github.com:sgminer-dev/sgminer.git
configure
make all / make install
Odd - I have none of the issues you described.
-
can you try to mine on pool http://p2pool.neoscrypt.de:19327 ?
User= your wallet address or any other name
Pass= any
If you start the miner in text mode with -D and -P you will see, that the miner switches to long-poll, but does not use stratum.
If you force stratum with stratum+tcp://p2pool.neoscrypt.de:19327 and -fix-protocol you’ll see (or at least I see) a message ’ no pool to get work from’
The only difference I see for the rest of the problems is, that I have a NVidia 750ti, no AMD based card.
-
I’m using SGMiner 5 on yaamp.com right this minute. I’ll check the debug output, though.
Nope - I see gen_stratum_work() and all other kinds of stratum stuff.
-
and how about old cards 5800 6800 6900 series?
-
and how about old cards 5800 6800 6900 series?
It should run. Not gonna be great on them - the kernel code is pretty bad copy+pasted CPU code, except for what I edited, and 6xxx have hardware vectors, so they REALLY could use OpenCL-specific vector types in the kernel. Even better if you vectorize the algorithms; I haven’t thought much about Blake2S, but vectorizing salsa and chacha so that the card can execute 4 integer operations at once would be child’s play.
-
It should run. Not gonna be great on them - the kernel code is pretty bad copy+pasted CPU code, except for what I edited, and 6xxx have hardware vectors, so they REALLY could use OpenCL-specific vector types in the kernel. Even better if you vectorize the algorithms; I haven’t thought much about Blake2S, but vectorizing salsa and chacha so that the card can execute 4 integer operations at once would be child’s play.
What would it take to get you to vectorize the kernel for 6000 series cards? Pls Wolf. Pls. :)
-
What would it take to get you to vectorize the kernel for 6000 series cards? Pls Wolf. Pls. :)
I already gave you guys something; besides, I don’t have a pre-GCN card to test on.
-
It’s ok Wolf, you have already done so much.
insanid, the official team will look into that for you I hope.
Let’s see what we can do.
