Optimising performance AMD 6200 for Monero

[deluxeuk]

Hi All. Need some advice about performance optimisation.

I have a server I'm using to mine Monero. It has 4x AMD 6200 CPUs, each are 16 core and 256GB RAM.

I'm using the xmrig docker image and geting about 1200 KH/s

How many threads should I be running and should I just have a single docker instance or several?

Any help appreciated

Thanks

 

[pyro_]

It is 2mb of cache per thread. I believe that those chips have 16mb of cache so 8 threads per chip

 

[deluxeuk]

It is 2mb of cache per thread. I believe that those chips have 16mb of cache so 8 threads per chip

Ok great. I'll try with 32 threads

Thanks

 

[deluxeuk]

Would you do 1 docker instance with 32 threads or 4x instances with 8 threads each?

 

[fishtacos]

Would you do 1 docker instance with 32 threads or 4x instances with 8 threads each?

You could try it with 1x 8 thread instance and look at the results, multiplied by 4. That's the way I'd go about it. You should too.

 

[deluxeuk]

You could try it with 1x 8 thread instance and look at the results, multiplied by 4. That's the way I'd go about it. You should too.

So using 4x docker containers with 8 threads each I'm getting about 430 H/s per container so about 1720 H/s overall. That seems ok to me. Would that be about correct?

 

[jims2321]

what amd 6200 version are you running deluxe?

 

[cafcwest]

The 16 core chips are the 627x and 628x chips. These chips have 16MB L2 + 12MB L3. Four processors, 112MB total /2 = 56 cores for mining.

I have a 4P system with 6276s and I get 2050H/s with the latest XMR-Stak.

 

[jims2321]

From my reading it not the total of all of the caches L2+L3 that determines the number of threads to run (cryptonight algorithm) but which is the largest that will meet the requirements . So if L2 will meet the requirements then it will. Not sure it will spill over to the L3 if you run more than intended (my non scientific testing seems to support this observation). So in your case 16MB x4 cpu's = 64MB / 2 MB/thread = 32 Threads, should be optimal. Further tweaking you can probably get some minor improvements.

 

[Joel]

The 16 core chips are the 627x and 628x chips. These chips have 16MB L2 + 12MB L3. Four processors, 112MB total /2 = 56 cores for mining.

I have a 4P system with 6276s and I get 2050H/s with the latest XMR-Stak.

That's pretty good... Do you know what the power draw is?