Quote from: zzz.to.the.moon on September 27, 2017, 02:36:58 AM

Hello @dstm

I really like your miner, it's really stable compared with EBWF's, and when it does crash (my hardware issue), it does so softly, allowing me to do a restart of the system. With EBWF's I couldn't kill it not even with -9. I had to power cycle the hardware.
However I do have the following issues:
- When using more than 10 GPUs the CPU is overloaded.
  * if it helps I can live with a longer time between refreshes, say 15 - 30 seconds
  * when that happens most of the share get rejected as invalid, see pool screenshot
  * do you think you can improve the CPU load performance, if not do you have any suggestion on what upgrade I should focus on? Cores, speed?
   + by my calculations I would need to use an i5 or i7 to be able to scale to 19 GPUs
- an API would be immensely appreciated
* No more having to stay logged in to monitor when one GPU died (and I could track down which one is the culprit)
* Get better statistics for optimization purposes.

Hardware:
- Asus B250 Mining Expert
- 10 x GTX 1070
- Intel Celeron G3900
- 2 GB DDR4 2133
- nvdia-smi -pl 90

On the left 12 x GTX 1070 GPUs, on the right 10 x GTX 1070 GPUs.

Hi zzz.to.the.moon thx for reporting, this is really welcome since it speedups development.

Quote

When using more than 10 GPUs the CPU is overloaded.

Judging from you htop output: the miner is using about 38% cpu on your system, most of your cpu load is on the kernel side, it's most likely caused by the drivers. If some of your GPUs are crashing you might get unpredictable behavior.

Quote

when that happens most of the share get rejected as invalid, see pool screenshot

This happens most likely because the shares arrive the pool too late because of the cpu load issues you're having.

Quote

do you think you can improve the CPU load performance, if not do you have any suggestion on what upgrade I should focus on? Cores, speed?

I'm constantly improving CPU and GPU side, however it's pretty well optimized now.
A high single-core performance is not required, the miner is written such that it scales well with the amount of available cores.

This is what I'm getting on a 10GPU system (6*1070 + 4*1080Ti) with an i3-7300T (a 35W CPU).

Code:

1 [|||||||||| 18.2%] Tasks: 112, 279 thr; 2 running
2 [|||||||||| 16.6%] Load average: 0.80 0.57 0.27
3 [|||||||||||||||||||||||||||| 51.0%] Uptime: 00:53:42
4 [||||||||| 15.3%]

Quote

by my calculations I would need to use an i5 or i7 to be able to scale to 19 GPUs

It depends on the speed of your GPUs - the overall speed of all GPUs on your system determine the CPU load. Judging from the htop results I'm getting on an i3-7300T an i3 should be able to drive 19GPUs, however I'm not sure if there are differences in the number of PCIE-lanes between the CPUs.

Quote

an API would be immensely appreciated

Will be there