GPU Mining Software - NVIDIA/AMD?

ConradJohnson

@Genoil - Awesome. You beat me to it, but I ran benchmarking with the larger memory (8gb) on the current release and it recognized 5 (out of 6) cards and I got 40 Mh/s. - (Ranked #1 in the world with that setup) That's with the E3 chip in there. All processes ended successfully.

I'll try the new version and move the memory and processors around and see what I get and post full results. Will that new version let me run all 6 cards?

edit:
Weird - I ran my test twice and now look: http://gav.ethdev.com:3000/

It said I ranked #1 both times, was that me?

Genoil

@ConradJohnson you should be able to run all 6 with 4GB.

Looked at the stats code briefly. It has a bug where it submits the result * 3. So it's you on #1 and #2.

ConradJohnson

@Genoil - sending you a PM as the output is too long to post here - in short: didn't work out with that new build.

ConradJohnson

@Genoil - from latest build I got some pretty good results. (E3 CPU - NVIDIA 750ti GPU)


with -t=n
n:
1 - 7835192 H/s with 100% of 1 CPU core.
2 - 15990784 H/s with 100% of 2 CPU core.
3 - 23942484 H/s with 100% of 3 CPU core.
4 - 32214584 H/s with 100% of 4 CPU core.
5 - 40253667 H/s with 100% of 5 CPU core. notes: though more thread swapping between cores at this level (not observed in n=1..4) on the E3 but might be the CPU itself optimizing
6 - 40551217 H/s notes: killed in one instance. the system became unusable here with more CPU core swapping and the process was killed. obviously not the +1 increase in H/s and subsequent tests with t=6 or without (default to use all) just gave me worse performance. Also, higher min/mean/max differential (min/mean/max: 28996055/37346072/40370176 H/s)

memory in all cases (8bg didn't get above 40%)


I'll see what the limitations are for the 1840 CPU

ConradJohnson

With the 1840 CPU (NVIDIA 750ti GPU )
with -t=n

n:
1 - 7835192 H/s with 100% of 1 CPU core.
2 - 15961656 H/s with 100% of 2 CPU core.
3 - 22777400 H/s with 100% of 2 CPU core.
4 - 29913542 H/s with 100% of 2 CPU core.
5 - 32360220 H/s with 100% of 2 CPU core.
6 - 33874541 H/s with 100% of 2 CPU core.

Memory never above 40%.

jzen

For kicks, I gave it a whirl on my ancient macbook, unsurprisingly, here are the results,
./eth -G -M --opencl-device 1:

Benchmarking on platform: { "platform": "Apple", "device": "GeForce GT 330M", "version": "OpenCL 1.0 " }
Preparing DAG...
ℹ 06:12:34|main stopWorking for thread gpuminer0
ℹ 06:12:34|main startWorking for thread gpuminer0
ℹ 06:12:34|main Spawning gpuminer0
ℹ 06:12:34|gpuminer0 Thread begins
ℹ 06:12:34|main Waiting until Started...
Warming up...
ℹ 06:12:34|gpuminer0 Trying to set Started: Thread was 0 ; 1
ℹ 06:12:34|gpuminer0 Entering work loop...
Using platform: Apple
Using device: GeForce GT 330M(OpenCL 1.0 )
OpenCL 1.0 is not supported.
✘ 06:12:34|gpuminer0 Error GPU mining. GPU memory fragmentation?
ℹ 06:12:34|gpuminer0 Finishing up worker thread...
ℹ 06:12:34|gpuminer0 State: Stopped: Thread was 1
ℹ 06:12:34|gpuminer0 Waiting until not Stopped...
Trial 1... 0
Trial 2... 0
Trial 3... 0
Trial 4... 0
Trial 5... 0
ℹ 06:12:55|main stopWorking for thread gpuminer0
ℹ 06:12:55|main Stopping gpuminer0
ℹ 06:12:55|main Waiting until Stopped...
ℹ 06:12:55|main Terminating gpuminer0
min/mean/max: 0/0/0 H/s
inner mean: 0 H/s

ConradJohnson

@jzen is this with the latest build? I was getting that with the previous build.

jzen

@ConradJohnson -- Not sure, I'm not at that computer at this moment. I performed a manual build on "git clone https://github.com/ethereum/cpp-ethereum.git" source, pulled this morning at around 6 AM ET.

Genoil

@ConradJohnson is that the latest build on cpp-ethereum? Or mine? I suppose the official one. I wonder what changed since it totally didn't work earlier for you with 6 GPUs. --edit-- oh wait it's the 8GB that makes it work. I would have expected more RAM use with so many GPU's, but apparently it handles that reasonably well.

@jzen your GPU only supports OpenCL 1.0. It's not supported.

ConradJohnson

@genoil - That was the latest develop official. Want me to give yours a go? Would like to see all 6 cards working, but it seems the thread swapping is really holding it back. Any way to designate 1 thread per core without swapping?

jzen

Thanks fellows, I suspected as much, I read that "OpenCL 1.0 is not supported" in the output but I didn't want to believe it...

Genoil

@ConradJohnson no that's not necessary as it wasn't working before. Just surprised to see that the memory requirements aren't as harsh as I had predicted.

I know that on Windows you can assign thread affinity, but that is still no guarantee against swapping.

ConradJohnson

@jzen depending on the level of cuda support the (future) cuda mining version provides, you might still be able to participate. Of course, it's always fun to build your own miner.

AlexNa

It does not build on mac. Any idea?

ConradJohnson

Do you have cuda toolkit installed @alexna ?

ConradJohnson

@genoil - so back on the E3 with 4gb of memory was a no go. On system monitor is went 70 to 80% but still had a hard time with anything above 2 cards. killed processes at 4 and up. Maybe my 6th card problem is a memory issue, like if I go above 8gb on this machine.

AlexNa

Just installed CUDA. Still have some errors on build. (see below).

Question: am I correct, that this will be part of geth really soon? So, we do not need to install it separately?

################# Testing C ##################
Scanning dependencies of target ethash
[ 20%] Building C object src/libethash/CMakeFiles/ethash.dir/io.c.o
[ 40%] Building C object src/libethash/CMakeFiles/ethash.dir/internal.c.o
[ 60%] Building C object src/libethash/CMakeFiles/ethash.dir/io_posix.c.o
[ 80%] Building CXX object src/libethash/CMakeFiles/ethash.dir/sha3_cryptopp.cpp.o
Linking CXX static library libethash.a
[ 80%] Built target ethash
Scanning dependencies of target Test
[100%] Building CXX object test/c/CMakeFiles/Test.dir/test.cpp.o
Linking CXX executable Test
[100%] Built target Test
Running 19 test cases...

*** No errors detected
################# Testing Go ##################
../../go/src/github.com/ethereum/go-ethereum/crypto/crypto.go:9:2: cannot find package "code.google.com/p/go.crypto/ripemd160" in any of:
/usr/local/Cellar/go/1.4.2/libexec/src/code.google.com/p/go.crypto/ripemd160 (from $GOROOT)
/Users/alexn/go/src/code.google.com/p/go.crypto/ripemd160 (from $GOPATH)
ethash.go:24:2: cannot find package "github.com/ethereum/go-ethereum/common" in any of:
/usr/local/Cellar/go/1.4.2/libexec/src/github.com/ethereum/go-ethereum/common (from $GOROOT)
/Users/alexn/go/src/github.com/ethereum/go-ethereum/common (from $GOPATH)
ethash.go:27:2: cannot find package "github.com/ethereum/go-ethereum/logger/glog" in any of:
/usr/local/Cellar/go/1.4.2/libexec/src/github.com/ethereum/go-ethereum/logger/glog (from $GOROOT)
/Users/alexn/go/src/github.com/ethereum/go-ethereum/logger/glog (from $GOPATH)
../../go/src/github.com/ethereum/go-ethereum/ethutil/script_unix.go:5:8: cannot find package "github.com/ethereum/serpent-go" in any of:
/usr/local/Cellar/go/1.4.2/libexec/src/github.com/ethereum/serpent-go (from $GOROOT)
/Users/alexn/go/src/github.com/ethereum/serpent-go (from $GOPATH)
../../go/src/github.com/ethereum/go-ethereum/crypto/crypto.go:11:2: cannot find package "github.com/obscuren/ecies" in any of:
/usr/local/Cellar/go/1.4.2/libexec/src/github.com/obscuren/ecies (from $GOROOT)
/Users/alexn/go/src/github.com/obscuren/ecies (from $GOPATH)
../../go/src/github.com/ethereum/go-ethereum/crypto/crypto.go:12:2: cannot find package "github.com/obscuren/secp256k1-go" in any of:
/usr/local/Cellar/go/1.4.2/libexec/src/github.com/obscuren/secp256k1-go (from $GOROOT)
/Users/alexn/go/src/github.com/obscuren/secp256k1-go (from $GOPATH)
../../go/src/github.com/ethereum/go-ethereum/crypto/crypto.go:13:2: cannot find package "github.com/obscuren/sha3" in any of:
/usr/local/Cellar/go/1.4.2/libexec/src/github.com/obscuren/sha3 (from $GOROOT)
/Users/alexn/go/src/github.com/obscuren/sha3 (from $GOPATH)
../../go/src/github.com/ethereum/go-ethereum/ethutil/config.go:8:2: cannot find package "github.com/rakyll/globalconf" in any of:
/usr/local/Cellar/go/1.4.2/libexec/src/github.com/rakyll/globalconf (from $GOROOT)
/Users/alexn/go/src/github.com/rakyll/globalconf (from $GOPATH)
make: *** [test] Error 1

ConradJohnson

@AlexNa so it looks like you built the cpp ethereum and tested the go client. go to this directory: ../cpp-ethereum/build/eth and then run ./eth -G -M

ConradJohnson

Putting this beast together tonight to (eventually) benchmark the AMD cards on the mining. @Genoil

CPU: AMD FX 8350

and 4 mem slots so I can put up to 12gb in this until I acquire more memory.

jzen

My poor old macbook cpu miner can no longer compete!

Genoil

@ConradJohnson please check latest cpp-ethereum:develop again with just 4GB. Should be fine now, regardless amount of GPUs.

ConradJohnson

@Genoil Roger that. I want to get this AMD one up and running first, then I'll switch back to NVIDIA.

Genoil

@ConradJohnson that's fine just stick 4GB in that one

zaxxon

I'm getting:
> admin.miner.hashrate()
888
on a 3.5GHz Xeon E5 Mac Pro running 8 threads and 16GB RAM. Is that KH/s?
Edit: Found a block!

zaxxon

Now if I could only get my GPU's working (I'm running go, apparently GPU not supported except in cpp?)

zaxxon

deleted

acros

@zaxxon, admin.miner.hashrate() is in H/s according to https://github.com/ethereum/go-ethereum/wiki/Mining#gpumining

jzen

However, https://github.com/ethereum/go-ethereum/wiki/JavaScript-Console#adminminerhashrate says the unit is KH/s which I think is the correct one and not H/s.

P.S. -- I submitted an issue ticket (https://github.com/cubedro/eth-netstats/issues/116) as the eth-netstats is misreporting the hash rate.

StephanTual

@acros, @jzen , yes, it's a bug in the way the Go client reports their hashrate. This is going to get fixed soon

michaelaudoux

I'm getting:
> admin.miner.hashrate()
720 on Imac 3.5 GHz Intel Core i7