Pages:
Author

Topic: CUDAminer problem - page 2. (Read 7468 times)

member
Activity: 85
Merit: 10
April 09, 2013, 11:12:44 PM
#16
What kind of performance would I get from a gtx 560 fermi?

Is it possible to allow the gpu to use system memory?
sr. member
Activity: 247
Merit: 250
April 09, 2013, 10:59:11 PM
#15
Running the newer version now. So far, the stales seem to be lower.

However, the hashrate recorded on the server pool end seems to be about half of the speed I'm at client side.

See here: https://i.imgur.com/wc4TRhm.png

Any reason this may be the case?

It is how the server calculates. Let it run for a while it will get close to the hash rate you're seeing on the client.

My client says 165, but the server says 157. latency

the server I am using took about 10 minutes or more before it was reflecting my hash rate close to what I am getting now
newbie
Activity: 35
Merit: 0
April 09, 2013, 10:57:04 PM
#14
Running the newer version now. So far, the stales seem to be lower.

However, the hashrate recorded on the server pool end seems to be about half of the speed I'm at client side.

See here: https://i.imgur.com/wc4TRhm.png

Any reason this may be the case?
newbie
Activity: 35
Merit: 0
April 09, 2013, 10:41:06 PM
#13

Which version are you using? Previous version had high stale/reject rates  the 4-09 version has a very low rate maybe 2%?

If you're not using the newest one..then it would decrease your performance.

Please note the version released today was updated a couple times by Christian.  May want to download it again to make sure you have the newest one

You can also use -D to get the autotuning log and paste it here. It may be selecting an invalid setting for your card.


Thanks for the quick reply. Didn't realize he had updated the version. Re-downloaded and running. Will keep this thread updated if I continue to run into further errors.

Cheers!
sr. member
Activity: 247
Merit: 250
April 09, 2013, 10:16:21 PM
#12
cbuchner1, its good to see you posting on this thread. I was starting to despair about getting to the 4 hour newbie mark.

I've been testing the miner on several different pools. I've been getting anywhere from 50-100% stales, and even invalid shares.

I attach some images here for you to see.

https://i.imgur.com/X6wqWb0.png
https://i.imgur.com/hniLGEd.png

I run a GTX680 card and would also like to know how I can increase performance (newbie user here)

I currently just run the program with the following command in the command prompt:

cudaminer.exe -o http://ltc.kattare.com:9332 -O username.1:password

I really appreciate you taking the time out to answer this. Cheers!

Which version are you using? Previous version had high stale/reject rates  the 4-09 version has a very low rate maybe 2%?

If you're not using the newest one..then it would decrease your performance.

Please note the version released today was updated a couple times by Christian.  May want to download it again to make sure you have the newest one


You can also use -D to get the autotuning log and paste it here. It may be selecting an invalid setting for your card.

newbie
Activity: 35
Merit: 0
April 09, 2013, 10:09:32 PM
#11
cbuchner1, its good to see you posting on this thread. I was starting to despair about getting to the 4 hour newbie mark.

I've been testing the miner on several different pools. I've been getting anywhere from 50-100% stales, and even invalid shares.

I attach some images here for you to see.

https://i.imgur.com/X6wqWb0.png
https://i.imgur.com/hniLGEd.png

I run a GTX680 card and would also like to know how I can increase performance (newbie user here)

I currently just run the program with the following command in the command prompt:

cudaminer.exe -o http://ltc.kattare.com:9332 -O username.1:password

I really appreciate you taking the time out to answer this. Cheers!
sr. member
Activity: 247
Merit: 250
April 09, 2013, 08:58:28 PM
#10
You may want to try running -D switch.

You can then skip autotuning and select the base rate yourself with this output.

If you need help feel free to post the -D output for autotuning here.
sr. member
Activity: 252
Merit: 254
April 09, 2013, 08:53:27 PM
#9
I'm getting roughly 78kh/s on my gtx560se's.  Over 20kh/s higher than cgminer.
newbie
Activity: 19
Merit: 0
April 09, 2013, 08:36:46 PM
#8
I'm getting 137 on my 670...
sr. member
Activity: 247
Merit: 250
April 09, 2013, 06:27:23 PM
#7
There was some bugs on the previous version. Specially with the Autotune feature. It is now fixed as of today. I'm now seeing around 165 on my GTX 570.

Sort note is make sure you've updated to today.

member
Activity: 85
Merit: 10
April 09, 2013, 05:50:55 PM
#6
Keep an eye on this thread if you have a Cuda card.

https://bitcointalk.org/index.php?topic=163750.120

I was able to get it running, but can't post there yet to get some tuning advice.
newbie
Activity: 5
Merit: 0
April 09, 2013, 05:46:19 PM
#5
Based on the parent cudaminer thread, sounds like most of the issues are fixed. Will give it a spin tonight! I have another machine with a radeon 6850 in it, so not sure I really want to have two video cards burning the midnight oil but fun to tinker with none-the-less Smiley
hero member
Activity: 756
Merit: 502
April 09, 2013, 08:00:39 AM
#4
Just a note regarding the 04/08 release.

Don't let the autotuner pick the S1x4 configuration. It's a bad config and autotune shouldn't have used it.

Run with --no-autotune, or manually specify a known good kernel launch configuration with the -l option.

You can also run autotune with the debug (-D) flag, inspect the tables visually and pick a reasonable configuration with high khash values.  S..x4, S..x5, S..x6, S..x7 are bad configurations and must be avoided even when high values appear in the autotune table.  The next release fixes it.

And yes, the 290x2 was tested on a 3GB card. You could try 145x2 instead. Wink

Christian

newbie
Activity: 33
Merit: 0
April 09, 2013, 07:34:21 AM
#3
Newbie w/ some feedback on cudaminer (4/8 release) as well.

It seems the 290x2 660Ti recommendation is for 3GB cards, my 2GB 660ti's wont allocate anywhere close to that blocks*warps.

cpu usage seems high for cuda, which was one of my hopes for cudaminer,
  as rpcminer-cuda gives very low cpu usage compared to cgminer w/ OpenCL for SHA256

Here are some results from some nvidia cards I have in operation right now vs how I had cgminer tuned
9800GT 512MB: 14-16kh/s -l S12x3 , up from 12kh/s
GTX275 896MB: 45-50kh/s -l S30x3 , up from 34kh/s
650Ti 1024MB: 66kh/s -l 88x2 , up from 50kh/s
660Ti 2048MB: 90-100kh/s -l 192x2 , up from 75-85kh/s - probably not tuned optimally in either case.

And of course, typical nvidia display driver crashing on exit, especially on the 600 series cards.
newbie
Activity: 5
Merit: 0
April 08, 2013, 11:38:50 PM
#2
I actually registered to post the exact same question. cudaminer claims my lowly gtx660 is getting 178kh/s (seems awful high), but none of that work seems to be going to the pool. I'm using a stratum proxy to connect to eu.wemineltc.com:3333. Despite the great numbers coming out of cudaminer, I don't think it is actually _doing_ anything. My GPU reports no activity via MSI Afterburner. Being a newbie, I'm sure I'm doing something wrong Smiley

Proxy started with:
mining_proxy.exe -o eu.wemineltc.com -p 3333 -oh localhost -sh localhost -nm

cudaminer started with:
cudaminer -o http://localhost:8332 -l S1x4 --user=myworkername --pass=myworkerpass

(The S1x4 parameter is what resulted after the autotune pass.)
newbie
Activity: 19
Merit: 0
April 08, 2013, 10:41:00 PM
#1
Hello, I installed the new version of CUDAminer and went from 127 khash/s to 180 khash/s. The thing is that my pool doesn't seem to be receiving them. I want to just ask the dev, however I am new to the forum and can't post in his thread yet. Can anyone help?

The CUDAminer thread is here: https://bitcointalksearch.org/topic/ann-cudaminer-ccminer-cuda-based-mining-applications-windowslinuxmacosx-167229
Pages:
Jump to: