Pages:
Author

Topic: [ANN][YAC] Yacminer GPU miner for Yacoin - page 12. (Read 57517 times)

hero member
Activity: 896
Merit: 1000
There isn't going to be another release of the miner any time soon. The best tuning options have already been implemented. Keep thread concurrency as high as possible and after that increase the lookup gap to allow higher intensities.

You may want to check how many HW errors exactly you're getting. If you're lucky like me, you could be getting only 3% HW errors. I'm still mining with the same settings on my 7790:

Just noticed that all my 7950 and 7970 don't have any errors at I=17 although 5870/5970 get them as soon as I set intensity at 12 or higher. I used various combinations of thread-concurrency, lookup-gap and threads but nothing works past I=11 for 5x70 cards for me (on several versions of Xubuntu Linux).
So this HW errors problem might be specific to <7xxx cards.
Don't think that's true.  I had HW errors on my 7950 until I upped the threadcount.  Now happily running at 123.2 kh/s

Before N increased I used the maximum threadcount my systems supported with lookup-gap = 2 (unless it was faster with a higher threadcount and lookup-gap = 3). I didn't change my configurations after N increased (other than trying to find parameters without HW errors on my 5x70, but results weren't good and I reverted to my previous configurations).

I use threadcount = 41216 on my 7950 and 7970 with lookup-gap = 2, 8000 or 12000 on my 5x70 (lookup-gap = 3 with threadcount = 12000).

I suppose results vary wildly with OS/drivers.
hero member
Activity: 820
Merit: 1000
There isn't going to be another release of the miner any time soon. The best tuning options have already been implemented. Keep thread concurrency as high as possible and after that increase the lookup gap to allow higher intensities.

You may want to check how many HW errors exactly you're getting. If you're lucky like me, you could be getting only 3% HW errors. I'm still mining with the same settings on my 7790:

Just noticed that all my 7950 and 7970 don't have any errors at I=17 although 5870/5970 get them as soon as I set intensity at 12 or higher. I used various combinations of thread-concurrency, lookup-gap and threads but nothing works past I=11 for 5x70 cards for me (on several versions of Xubuntu Linux).
So this HW errors problem might be specific to <7xxx cards.
Don't think that's true.  I had HW errors on my 7950 until I upped the threadcount.  Now happily running at 123.2 kh/s
hero member
Activity: 896
Merit: 1000
There isn't going to be another release of the miner any time soon. The best tuning options have already been implemented. Keep thread concurrency as high as possible and after that increase the lookup gap to allow higher intensities.

You may want to check how many HW errors exactly you're getting. If you're lucky like me, you could be getting only 3% HW errors. I'm still mining with the same settings on my 7790:

Just noticed that all my 7950 and 7970 don't have any errors at I=17 although 5870/5970 get them as soon as I set intensity at 12 or higher. I used various combinations of thread-concurrency, lookup-gap and threads but nothing works past I=11 for 5x70 cards for me (on several versions of Xubuntu Linux).
So this HW errors problem might be specific to <7xxx cards.
legendary
Activity: 1470
Merit: 1030
Well, let's re-iterate this once more. GPU mining is not going to magically stop at N = 8192. My miner should survive it just fine. I'm not going to benchmark it because people always keep finding new settings after N changes. Some people will use a higher lookup gap and others will find ways to use lower values. I might come up with some new tricks but at the moment I'm pretty happy how the code scales with the lookup gap.

Hi Mikael. Firstly, thanks for your releases. Could you make a prediction about the long term viability of GPU mining? Yacoin was originally designed to be a currency that could be mined with CPU efficiently. Is that a busted flush now? Or will continued increments in N lead to a situation where CPU miners compete with, or even outmine GPU for equivalent energy input? I'm sure it's difficult to say for certain, but your guess is likely to be more educated than others.
sr. member
Activity: 301
Merit: 250
There isn't going to be another release of the miner any time soon. The best tuning options have already been implemented. Keep thread concurrency as high as possible and after that increase the lookup gap to allow higher intensities.

You may want to check how many HW errors exactly you're getting. If you're lucky like me, you could be getting only 3% HW errors. I'm still mining with the same settings on my 7790:
Code:
--scrypt -w 128 --lookup-gap 2 -I 17 --thread-concurrency 8000 -g 1

You can quite easily calculate the total number of difficulty-1 shares your GPU is producing with the following formula:
Code:
hashrate * running_time / 65536

For example, if you've been mining for an hour at 59.3 kh/s, the number of difficulty-1 shares would be:
Code:
59300 * 3600 / 65536 = 3257

Now you can go and compare that number with the amount of HW errors you got.
hero member
Activity: 637
Merit: 500
With the new n update, it seems I can't get my 7950s back in the 200 range.  Any ideas?
This is normal, every N incerase hashrate halves for everybody.
The problem now are the HW errors, getting lots of them with the same settings as with Nfactor=10, still trying new settings without any luck.
member
Activity: 103
Merit: 10
With the new n update, it seems I can't get my 7950s back in the 200 range.  Any ideas?
full member
Activity: 196
Merit: 100
Great work Mikaelh ! you made us all some money.. Some donations shoud come your way Wink
sr. member
Activity: 252
Merit: 250
had this (v2) and the previous version crash the video driver on exit a few times now
full member
Activity: 162
Merit: 100
Averaging 245 kh/s with 7950s

Used thread concurrency of 21712, gpu-powertune of 20, core speed of 1100 and mem speed of 1500
sr. member
Activity: 252
Merit: 250
thanks for the update had the network problem today

also found this minor cosmetic thing where i have extra blank lines above and below the bottom ----- line


also not sure if i should be using more than the memory buffer of the card

for anyone wanting some nvidia settings, heres what im using for my 560ti

--scrypt --worksize 128 --lookup-gap 1 -I 11 --thread-concurrency 5376
sr. member
Activity: 301
Merit: 250
Thank you for the update.

Sorry about nagging, but I am still interested in a Ybcoin version of the Yacminer.
Would a small YAC donation and promise of 5% of all mined Ybcoins spark your interest?

Tried and failed at updating the source myself, yes, that is how dumb I am.

Well, the Ybcminer has arrived:
https://bitcointalksearch.org/topic/annybc-ybcminer-gpu-miner-for-ybcoin-249360

Please keep further discussion of Ybcminer or Ybcoin in the new thread.
full member
Activity: 162
Merit: 100
Ok great, I'll try to find some good settings for 7950s
sr. member
Activity: 301
Merit: 250
Can you use all of the same commands as cgminer?

Yup. It's based on cgminer with minimal changes to support Yacoin.
full member
Activity: 162
Merit: 100
Can you use all of the same commands as cgminer?
sr. member
Activity: 301
Merit: 250
Thank you for the update.

Sorry about nagging, but I am still interested in a Ybcoin version of the Yacminer.
Would a small YAC donation and promise of 5% of all mined Ybcoins spark your interest?

Tried and failed at updating the source myself, yes, that is how dumb I am.

Fine, I might make one since there's popular demand. Easy enough to change 4 lines...
full member
Activity: 138
Merit: 100
Benchmarks of my AMD 7970 3GB running at 900MHz core, 1000MHz mem and 0.975V in Afterburner, with yacminer-3.3.1-yac2-windows-x64 in Win 7 64bit on http://yac2.ltcoin.net

150Kh/s --scrypt --thread-concurrency 8192 -I 13 -g 2 -w 256 --lookup-gap 4    (uses 0,8GB RAM)  
236.5Kh/s --scrypt --thread-concurrency 8192 -I 13 -g 2 -w 256 --lookup-gap 2  (uses 1,4GB RAM)  
103Kh/s --scrypt --thread-concurrency 8192 -I 13 -g 2 -w 256 --lookup-gap 1     (uses 2,4GB RAM)  

Benchmarks of my 7970 running at 900MHz core, 1500MHz mem and 0.975V in Afterburner, with yacminer-3.3.1-yac2-windows-x64 on http://yac2.ltcoin.net

151Kh/s --scrypt --thread-concurrency 8192 -I 13 -g 2 -w 256 --lookup-gap 4
237.7Kh/s --scrypt --thread-concurrency 8192 -I 13 -g 2 -w 256 --lookup-gap 2    
181Kh/s --scrypt --thread-concurrency 8192 -I 13 -g 2 -w 256 --lookup-gap 1 (140 for the card running Windows)


Higher thread concurrencies (900/1000/0.975):
229Kh/s --scrypt --thread-concurrency 38000 -I 13 -g 1 -w 256 --lookup-gap 2    (uses 2,6GB RAM)
233Kh/s --scrypt --thread-concurrency 38000 -I 15 -g 1 -w 256 --lookup-gap 2    (uses 2,6GB RAM)
233Kh/s --scrypt --thread-concurrency 38000 -I 17 -g 1 -w 256 --lookup-gap 2    (uses 2,6GB RAM)
Trying -g 2 crashes, increasing lookup gap kills he performance.

EDIT: At 1100/1500MHz I get: 287Khps, temp are getting to high though.
full member
Activity: 138
Merit: 100
Thank you for the update.

Sorry about nagging, but I am still interested in a Ybcoin version of the Yacminer.
Would a small YAC donation and promise of 5% of all mined Ybcoins spark your interest?

Tried and failed at updating the source myself, yes, that is how dumb I am.
sr. member
Activity: 301
Merit: 250
New version is up. -yac2 fixes a problem which has annoyed me greatly. I noticed that the miner always starts experiencing lots of network failures when mining on a pool for a while. I investigated the issue and I think the cause is that the miner does not close long polling connections. -yac2 fixes that by telling the cURL library to limit the number of long polling connections. It's not a perfect fix because the miner is still keeping more than one connection open but at least it shouldn't stall anymore.
sr. member
Activity: 301
Merit: 250
Hey mikaelh, are you looking into the ways around keeping GPU's usable after the Aug N++?  Or is that something hanzac might be doing?  Or neither? Smiley

Well, let's re-iterate this once more. GPU mining is not going to magically stop at N = 8192. My miner should survive it just fine. I'm not going to benchmark it because people always keep finding new settings after N changes. Some people will use a higher lookup gap and others will find ways to use lower values. I might come up with some new tricks but at the moment I'm pretty happy how the code scales with the lookup gap.
My apologies, I thought I understood the way it works but obviously not.  I thought that the RAM on the GPU was the limitation and this couldn't be overcome.  I'm not sure I understand the lookup gap / TC correlation and the use of GPU RAM vs system RAM.

It's pretty hard to discuss the details of lookup gap without explaining the entire scrypt algorithm first. I did write a bit about lookup gap on the other forum:
http://yacointalk.com/index.php/topic,350.msg1511.html#msg1511

The miner allocates the scratchpad from GPU memory. You may need enough system RAM to support that, but that's a driver issue.
Pages:
Jump to: