[ANN] cudaMiner & ccMiner CUDA based mining applications [Windows/Linux/MacOSX] - page 1131.

rethaw

sr. member

Activity: 378

Merit: 255

I tried this out on Amazon's EC2 and saw a marked improvement over cgminer's with scrypt. The auto-tune found 223x2, which I presume is aiming for half of the CUDA cores (224x2).

https://bitcointalksearch.org/topic/mining-on-amazon-ec2-scrypt-or-btc-169377

datguyian

sr. member

Activity: 840

Merit: 251

So I was thinking cool, I might be able to get another 50 kh/s from that workstation that's been sitting in my office for the last few months. WRONG! I'm averaging almost 400 kh/s right now from a quadro 600... if this keeps up and doesn't burn my office down by the morning (i'm not physically at the location atm, so hoping nothing is smoking right now), expect a decent donation to this project soon.

omo

full member

Activity: 147

Merit: 100

after upgrading to the latest driver,I made my Quadro FX 1800M running:

Code:

.......
.......
[2013-04-10 12:28:56] GPU #0: 14.12 khash/s with configuration S10x3
[2013-04-10 12:28:56] GPU #0: using launch configuration S10x3
[2013-04-10 12:28:56] GPU #0: Quadro FX 2800M, 960 hashes, 0.00 khash/s
[2013-04-10 12:28:56] GPU #0: Quadro FX 2800M, 960 hashes, 6.91 khash/s
[2013-04-10 12:28:56] LONGPOLL detected new block
[2013-04-10 12:28:56] DEBUG: got new work
[2013-04-10 12:28:56] GPU #0: Quadro FX 2800M, 7680 hashes, 12.41 khash/s
[2013-04-10 12:29:06] DEBUG: hash <= target
Hash: 00005f1b32735cc067794562b715b6dc1b42133abb3416db26e127ca7e8d59de

and GTX 680:

Code:

[2013-04-10 14:06:19] GPU #0: 143.56 khash/s with configuration 166x2
[2013-04-10 14:06:19] GPU #0: using launch configuration 166x2
[2013-04-10 14:06:20] GPU #0: GeForce GTX 680, 10624 hashes, 0.06 khash/s
[2013-04-10 14:06:20] DEBUG: got new work in 647 ms
[2013-04-10 14:06:20] GPU #0: GeForce GTX 680, 10624 hashes, 69.53 khash/s
[2013-04-10 14:06:23] DEBUG: hash <= target

omo

full member

Activity: 147

Merit: 100

Quote from: Lacan82 on April 09, 2013, 10:11:42 PM

I saw your edit on your previous post. Is your computer in another language? your card name isn't even showing. So maybe a language issue?

Drives are up to date?

yeah, my win7 is Chinese version.
I shall check the driver.
thank you

Lacan82

sr. member

Activity: 247

Merit: 250

I saw your edit on your previous post. Is your computer in another language? your card name isn't even showing. So maybe a language issue?

Drives are up to date?

omo

full member

Activity: 147

Merit: 100

Quote from: Lacan82 on April 09, 2013, 09:58:48 PM

Quote from: omo on April 09, 2013, 09:50:56 PM

I tried the latest version on a Quadro FX 2800M card, it crashed after giving the following message:

Code:

E:\cudaminer-2013-04-09>cudaminer.exe --url http://litecoinpool.org:9332/ --user user --pass x --thread 1
*** CudaMiner for nVidia GPUs by Christian Buchner ***
This is version 2013-04-09 (alpha)
based on pooler-cpuminer 2.2.3 (c) 2010 Jeff Garzik, 2012 pooler
Cuda additions Copyright 2013 Christian Buchner
My donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm

[2013-04-10 10:43:36] 1 miner threads started, using 'scrypt' algorithm.
[2013-04-10 10:43:37] Long-polling activated for http://litecoinpool.org:9332/LP
[2013-04-10 10:43:37] GPU #0: with compute capability 0.47446600
[2013-04-10 10:43:37] GPU #0: Performing auto-tuning (Patience...)
[2013-04-10 10:43:37] GPU #0: 0.00 khash/s with configuration 0x0
[2013-04-10 10:43:37] GPU #0: using launch configuration 0x0

What happens if you don't specify the thread? same thing? you can also use -D to see what Autotuning does

thank you for your reply.
if I did'nt specify the thread option, it won't go below the donation line.
I specified -D and got a slightly different messages, but still crashed:

Code:

E:\cudaminer-2013-04-09>cudaminer.exe --url http://litecoinpool.org:9332/ --user user --pass x --thread 1 -D

*** CudaMiner for nVidia GPUs by Christian Buchner ***
This is version 2013-04-09 (alpha)
based on pooler-cpuminer 2.2.3 (c) 2010 Jeff Garzik, 2012 pooler
Cuda additions Copyright 2013 Christian Buchner
My donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm

[2013-04-10 11:07:21] 1 miner threads started, using 'scrypt' algorithm.
[2013-04-10 11:07:22] Long-polling activated for http://litecoinpool.org:9332/LP
[2013-04-10 11:07:22] DEBUG: got new work in 1019 ms
[2013-04-10 11:07:22] GPU #0: with compute capability 532.0
[2013-04-10 11:07:22] GPU #0: Performing auto-tuning (Patience...)
[2013-04-10 11:07:22] GPU #0: 0.00 khash/s with configuration 0x0
[2013-04-10 11:07:22] GPU #0: using launch configuration 0x0

Lacan82

sr. member

Activity: 247

Merit: 250

Quote from: omo on April 09, 2013, 09:50:56 PM

I tried the latest version on a Quadro FX 2800M card, it crashed after giving the following message:

Code:

E:\cudaminer-2013-04-09>cudaminer.exe --url http://litecoinpool.org:9332/ --user user --pass x --thread 1
   *** CudaMiner for nVidia GPUs by Christian Buchner ***
   This is version 2013-04-09 (alpha)
   based on pooler-cpuminer 2.2.3 (c) 2010 Jeff Garzik, 2012 pooler
   Cuda additions Copyright 2013 Christian Buchner
   My donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm

[2013-04-10 10:43:36] 1 miner threads started, using 'scrypt' algorithm.
[2013-04-10 10:43:37] Long-polling activated for http://litecoinpool.org:9332/LP
[2013-04-10 10:43:37] GPU #0: with compute capability 0.47446600
[2013-04-10 10:43:37] GPU #0: Performing auto-tuning (Patience...)
[2013-04-10 10:43:37] GPU #0: 0.00 khash/s with configuration 0x0
[2013-04-10 10:43:37] GPU #0: using launch configuration 0x0

What happens if you don't specify the thread? same thing? you can also use -D to see what Autotuning does

Nevermind. it isn't even seeing your cards properly. Language issue maybe? I see Japanese character on the one print out.

omo

full member

Activity: 147

Merit: 100

I tried the latest version on a Quadro FX 2800M card, it crashed after giving the following message:

Code:

E:\cudaminer-2013-04-09>cudaminer.exe --url http://litecoinpool.org:9332/ --user user --pass x --thread 1
   *** CudaMiner for nVidia GPUs by Christian Buchner ***
   This is version 2013-04-09 (alpha)
   based on pooler-cpuminer 2.2.3 (c) 2010 Jeff Garzik, 2012 pooler
   Cuda additions Copyright 2013 Christian Buchner
   My donation address: LKS1WDKGED647msBQfLBHV3Ls8sveGncnm

[2013-04-10 10:43:36] 1 miner threads started, using 'scrypt' algorithm.
[2013-04-10 10:43:37] Long-polling activated for http://litecoinpool.org:9332/LP
[2013-04-10 10:43:37] GPU #0: with compute capability 0.47446600
[2013-04-10 10:43:37] GPU #0: Performing auto-tuning (Patience...)
[2013-04-10 10:43:37] GPU #0: 0.00 khash/s with configuration 0x0
[2013-04-10 10:43:37] GPU #0: using launch configuration 0x0

edit:
I got similar messages on a GTX680 box(desktop remote accessing), all are win7/64 os:

Code:

[2013-04-10 11:01:22] 1 miner threads started, using 'scrypt' algorithm.
[2013-04-10 11:01:23] Long-polling activated for http://litecoinpool.org:9332/LP

[2013-04-10 11:01:23] GPU #0: ,跸?羨 with compute capability 10002000.10002000
[2013-04-10 11:01:23] GPU #0: Performing auto-tuning (Patience...)
[2013-04-10 11:01:23] GPU #0: 0.00 khash/s with configuration 0x0
[2013-04-10 11:01:23] GPU #0: using launch configuration 0x0

gchil0

newbie

Activity: 59

Merit: 0

Quote from: cbuchner1 on April 09, 2013, 07:46:16 PM

Not nearly as impressive as I thought it would be. Is ECC turned off?
Performance is not scaling up with core count as much as I would have hoped for.
Nice to know the code does not barf on 64bit and that shares are accepted.

Yes, ECC is disabled. The older version only got ~110 khash/sec, so the Titan version is nearly 70% faster on the K20.

gchil0

newbie

Activity: 59

Merit: 0

And here's a GTX 480 autotune:

http://pastebin.com/Wm2GnZ26

Lacan82

sr. member

Activity: 247

Merit: 250

Thank god! I can finally post! No longer have to email Christian Directly

\o/

GTX 570 GPU 822 MHZ receiving 165 KH
650M receiving 29 KH

On the list: GTX 9800 to be tested.

Pool is reporting tho I am mining at 190 KH, and 41 KHs

wndrbr3d

hero member

Activity: 914

Merit: 500

Quote from: Testarossa on April 09, 2013, 08:29:07 PM

Hello,

First of all, good work Christian. Applause!

I have a GTX 295 (2 GPU's). The first version of the cudaminer performs a bit better than the latest(09/04) one. Between 1 a 2 Khash/s better.
Without overclocking I get +-36 Khash/s on each GPU.

Tested for bitcoin mining (not a cudaminer), and there it went to +-52 Mhash/s on each GPU.
Should I be able to get more or less, litecoin whise mining?

What do I have to overclock/underclock to be efficient?

Bitcoin uses SHA-256 where as Litecoin uses scrypt for proof of work. Scrypt is much more memory intensive, so it's a world of difference.

Testarossa

newbie

Activity: 11

Merit: 0

Hello,

First of all, good work Christian. Applause!

I have a GTX 295 (2 GPU's). The first version of the cudaminer performs a bit better than the latest(09/04) one. Between 1 a 2 Khash/s better.
Without overclocking I get +-36 Khash/s on each GPU.

Tested for bitcoin mining (not a cudaminer), and there it went to +-52 Mhash/s on each GPU.
Should I be able to get more or less, litecoin whise mining?

What do I have to overclock/underclock to be efficient?

cbuchner1

hero member

Activity: 756

Merit: 502

Quote from: gchil0 on April 09, 2013, 07:38:59 PM

I couldn't get a 32-bit compile in Ubuntu 12.04 because of libcurl issues. For the 64-bit compile, I had to add -fpermissive to CXXFLAGS to get the compiler to accept a cast related to jansson. After that, I have a binary.

Here's the autotune for a K20.
http://pastebin.com/s9Cyb8yA

Not nearly as impressive as I thought it would be. Is ECC turned off?
Performance is not scaling up with core count as much as I would have hoped for.
Nice to know the code does not barf on 64bit and that shares are accepted.

gchil0

newbie

Activity: 59

Merit: 0

I couldn't get a 32-bit compile in Ubuntu 12.04 because of libcurl issues. For the 64-bit compile, I had to add -fpermissive to CXXFLAGS to get the compiler to accept a cast related to jansson. After that, I have a binary.

Here's the autotune for a K20.
http://pastebin.com/s9Cyb8yA

Number6

sr. member

Activity: 490

Merit: 254

Quote from: cbuchner1 on April 09, 2013, 06:40:00 PM

I am still lacking data points on compute 1.0 and 1.1 devices (8800GTX, etc...), and also compute 1.2. Really ancient hardware Wink

I actually have a old box with an 8800GTX in it! I will run it for awhile and report my results.

Update:
Autotune picked 1x4
Hashrate of 2.4 kH/sec

didn't bother which much more testing due to such a low rate

Number6

sr. member

Activity: 490

Merit: 254

Quote from: cbuchner1 on April 09, 2013, 06:38:06 PM

Memory size can be a limiting factor if you have lots of SP (streamprocessors or CUDA cores) on the cards.

e.g. on a 660Ti with 3GB I can go for 290x2 (which consumes some 2.5 GB of RAM), whereas on smaller cards we can choose e.g. 148x2.

That explains why mine didn't work on 290x2 then and picked 148x2. I thought I had a 3 GB 660TI, but then looked at the box and indeed it was only a 2 GB model.

Would you mind explaining the math a bit more? From what you posted it seems as if I could go higher than 148x2, but not to 290x2 which you mention consumes 2.5 GB.

For other's reference:

GTX 660Ti 2GB
314.22 WHLQ Driver
GPU Core OC to 1320 MHz
autotune launch config settings 148x2
130-140 khash/sec with 98.8% average accepted share rate - via console
130.8 kH/sec reported by Pool (average over 10 minutes)

cbuchner1

hero member

Activity: 756

Merit: 502

Quote from: TiWu on April 09, 2013, 06:18:54 PM

What is the autotune using to determine the outcome? Should I try to find an suitable config and then fix than in the startup params? Or is it better if I let it autotune on each startup?

Due to measurement errors there is always a bit of randomness involved. Occasionally it will even find a surprising new configuration that suddenly beats the one you were using previously.

It may depend on current overclocking settings as well (memory vs. core clock), that's why it's called "tuning".

Thanks to everybody who is posting their launch configs here. It allowed me to improve the heuristics code for people who do not have the patience to wait through an autotune session.

I am still lacking data points on compute 1.0 and 1.1 devices (8800GTX, etc...), and also compute 1.2. Really ancient hardware Wink

And of course, the TITAN!

Due to good benchmark results, I just ordered myself a GTX 570 (Club 3D also, 149 Euros)
Anyone got a GTX 590? This dual GPU card could do 300 kHash/s I guess.

Also I posted one last binary+source code update for today, which includes the improved heuristics for --no-autotune. I found and fixed one more bug appearing on Linux in my 9600M GPU. Most autotune measurements were showing 0 kHash suddenly.

Christian

cbuchner1

hero member

Activity: 756

Merit: 502

Quote from: wndrbr3d on April 09, 2013, 06:30:44 PM

Is it the SP's available or card memory available that is a limiting factor for the CUDA miner? I ask because I have a 670 with 4GB (vs. 2GB standard) and was curious if I should pop that in and give it a try.

Memory size can be a limiting factor if you have lots of SP (streamprocessors or CUDA cores) on the cards.

e.g. on a 660Ti with 3GB I can go for 290x2 (which consumes some 2.5 GB of RAM), whereas on smaller cards we can choose e.g. 148x2.

The more CUDA cores a card has, the larger the grid x block value we need to throw at it to keep the SPs busy - but also the memory requirement grows.

I have now improved my heuristics (the --no-autotune case) for such cases on compute 3.0 cards.

I have no experience with the 670 or 680 cards yet. But 4GB vs 2GB can make a difference in the autotune results.

Christian

wndrbr3d

hero member

Activity: 914

Merit: 500

Is it the SP's available or card memory available that is a limiting factor for the CUDA miner? I ask because I have a 670 with 4GB (vs. 2GB standard) and was curious if I should pop that in and give it a try.

Topic: [ANN] cudaMiner & ccMiner CUDA based mining applications [Windows/Linux/MacOSX] - page 1131. (Read 3426947 times)