Author

Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 1128. (Read 2347601 times)

sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
Some longer interval tests...
My 970/750ti test rig (linux, quark algo, 0.1 diff) reports ~27M with -g 3 setting now. It was 19M without -g parameter. But yaamp reports lower hashrate! It is about the half of the hasrate reported by ccminer. (13M vs 27M)
So the accepted hasrate decreased by 30% Sad It is a 3 hours average.
The situation is almost the same with -g 2 setting.
But as I recognized acceptance rate is 100%!

If you run quark at nicehash, you will get the average speed of the last 24 hours and also a detailed graph of the performance over time.
Do you get many "does not validate on the cpu" errors?

Try to reduce the cpu-priority
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
@Sp_
Something interesting. Same config -g 3 -i 22 but with cpu-priority to High(4) instead of Realtime(5), no crazy displayed hashrates in the miner window.

This is interesting. My fork uses realtime as default.

I want to test to revert to the original ccminer.cpp stats and threading code (djm34 fork), to see if I can locate invalid hash with multiple thread bug.
newbie
Activity: 14
Merit: 0
Some longer interval tests...

My 970/750ti test rig (linux, quark algo, 0.1 diff) reports ~27M with -g 3 setting now. It was 19M without -g parameter. But yaamp reports lower hashrate! It is about the half of the hasrate reported by ccminer. (13M vs 27M)
So the accepted hasrate decreased by 30% Sad It is a 3 hours average.
The situation is almost the same with -g 2 setting.
But as I recognized acceptance rate is 100%!
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
Thanks for your support

You need to remove the changes in the last submit to github:

https://github.com/sp-hash/ccminer/commit/bf20a3a809aee86520958eb1aad3ad2807878eeb

Then it should be stable.

It ran for 6 hours on my testrig without any issues with -g 3 -i 22
sr. member
Activity: 248
Merit: 250
you need to revert the latest checkin
@Sp Didn't understand clearly, Donation on the way ))) not to much but it's better then nothing )))
Dark  intstanX :  480f058051f5f89797738585f63c0d491149c27f3f2aeacbd42f5f48223a828b-000
sr. member
Activity: 292
Merit: 250
@Sp_

Something interesting. Same config -g 3 -i 22 but with cpu-priority to High(4) instead of Realtime(5), no crazy displayed hashrates in the miner window.
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
you need to revert the latest checkin
sr. member
Activity: 248
Merit: 250
Redownload latest built rom Git, with g 2 or 3 parameter miner crashes within a minute.win 8.1 5x750 Ti . Huh
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
I think the problem is that you are using the 750 (512 shaders, 1gb memory) card. The card is out of resources, and is unable to increase the speed with more threads.
So why miner stats show high hashrate?

try with -d 0,0 . is the total hashrate correct?
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
CPU usage is up significantly when using multiple threads, How much load on the i7?
CPU usage somewhere between 7 to 13% (VARDIFF)
Edit: With d=0.05, constant cpu usage around 13%.
Interesting, Is the load spread across the cpu's? or do you have one cpu that takes most of the load?
It looks to me like most of the load is across 3 threads.


This is as expected. Each gpu thread is assigned a cpu if available.
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
I think the problem is that you are using the 750 (512 shaders, 1gb memory) card. The card is out of resources, and is unable to increase the speed with more threads.
So why miner stats show high hashrate? You should fix it I think. If you will fix the stats you will see real situation with performance.

The stats needs to be fixed, and the cpu validation errors needs to be gone. Working on it..
legendary
Activity: 1510
Merit: 1003

I think the problem is that you are using the 750 (512 shaders, 1gb memory) card. The card is out of resources, and is unable to increase the speed with more threads.

So why miner stats show high hashrate? You should fix it I think. If you will fix the stats you will see real situation with performance.
sr. member
Activity: 292
Merit: 250
CPU usage is up significantly when using multiple threads, How much load on the i7?
CPU usage somewhere between 7 to 13% (VARDIFF)
Edit: With d=0.05, constant cpu usage around 13%.

Interesting, Is the load spread across the cpu's? or do you have one cpu that takes most of the load?

It looks to me like most of the load is across 3 threads.

sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
CPU usage is up significantly when using multiple threads, How much load on the i7?
CPU usage somewhere between 7 to 13% (VARDIFF)
Edit: With d=0.05, constant cpu usage around 13%.

Interesting, Is the load spread across the cpu's? or do you have one cpu that takes most of the load?
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
I think the problem is that you are using the 750 (512 shaders, 1gb memory) card. The card is out of resources, and is unable to increase the speed with more threads.
How is this happening?

I guess the driver will have to wait until a cudacore is available in order to compute the 2nd hash.

The gtx 980x has 2048 cuda cores, but I only use 512 (threads per block). With the -g 4 parameter, all the cudacores will be used. So the highend cards will get the most speedincrease. But with validation errors, the rate on the pool will be lower. (These has to be removed)
sr. member
Activity: 292
Merit: 250
CPU usage is up significantly when using multiple threads, How much load on the i7?

CPU usage somewhere between 7 to 13% (VARDIFF)

Edit: With d=0.05, constant cpu usage around 13%.
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
Also getting loads of errors.


Yes working to fix this. In benchmark mode it works fine and the hashrate is higher. You need to mine with a high difficulty to reduce the errors. for one card set the password to f.ex -p d=0.05
I have confirmed that two threads on the same gpu produce 2 different valid hashes (benchmarkmode) But for some reason the pool connection is causing problems...
sr. member
Activity: 292
Merit: 250
Also getting loads of errors.

hero member
Activity: 672
Merit: 500
I think the problem is that you are using the 750 (512 shaders, 1gb memory) card. The card is out of resources, and is unable to increase the speed with more threads.
How is this happening?
sp_
legendary
Activity: 2926
Merit: 1087
Team Black developer
All this threading is just playing with numbers, pools show real performance drop ...
So far it does improve performance. My 5 min-average went from ~6.5 MHs to ~7.2 MHs. Just not what's reported on ccminer window.
I have also messured a increased performance on the pool when mining quark around 5-10%
What cpu's are you running with? How many cores?
Mne's a i7-4770k. I have the ccminer running in realtime mode.

CPU usage is up significantly when using multiple threads, How much load on the i7?
Jump to: