Program Failure after connect

[nichtdu]

Hey,

GPU 0: unspecified launch failure C:/Users/MAIN/Documents/Visual Studio 2015/Projects/xmr-stak-nvidia/xmr-stak-nvidia/nvcc_code/cuda_extra.cu line 245

[miziel]

Hey, Probably wrong parameters in config file.

[fireice-uk]

Most likely. If you still have a problem with this, please post the config and the hardware that you have.

[nichtdu]

i got a 970ti

"gpu_threads_conf" : [ { "index" : 0, "threads" : 17, "blocks" : 60, "bfactor" : 0, "bsleep" : 0, "affine_to_cpu" : false}, ],

it works now but with whooping 50/Hashes

[nichtdu]

aeh the 8 and 100 ofcourse - copied from wrong config

[fireice-uk]

the default threads and blocks are set for 1070 - which is a more powerful card, try dialling them down to something like 17x30 and see how it does then

[psychocrypt]

You should use this settings:

"gpu_threads_conf" : [
{ "index" : 0, "threads" : 24, "blocks" : 39, "bfactor" : 6, "bsleep" : 25, "affine_to_cpu" : false},
],

to get a good performance.

[fireice-uk]

@psychocrypt Interesting - what is the reason for the odd 39 block count?

[psychocrypt]

970ti has 13 SMX (multiprozessors). It is a Maxwell GPU what menas that the RAM usage must be limited to 2GiB. Each SMX gets 3 blocks thus we calculate simultaneously 13x3x24 = 936 Hashes (1872GB RAM) at the same time. See also this short description.

[fireice-uk]

@psychocrypt it seems you snipped the link that you wanted to put in. Do you still have it around (browser history)?

[psychocrypt]

I updated the link in my last comment.

[fireice-uk]

@psychocrypt Just out of curiosity, is it a hard per card limit or some kind of restriction in the firmware? Can it be gamed using multiple processes?

[psychocrypt]

The memory limit is a hard hardware limit depending on the compute architecture. For fermi (sm_2x), kepler(sm_3x) and maxwell(sm_5x) the limit is 2GB (in the most cases, some gpus have a 1GiB limit) and for Pascal (sm_6x) the limit should be 16GB. That is no limit that avoid to use more than 2GB but if you do random memory access on more than 2GB the performance goes done by factor of 6. The reason for the limit is the small TLB and the 512KiB pages. With pascal the pages size is increased to 2MiB. Multiple processes not helps the avoid this limit. I talked a month ago with a NVIDIA dev if there is any workaround available but it is not possible to avoid the TLB misses or change the page size for the gpu.

Maybe I can increase the performance a little bit in the feature by using the fact that some gpus have disabled SMX(multiprocessors), this could be used to increase the number of blocks for those SMX. Depending on the hardware 3-4 SMX shares one TLB and the limit of virtual memory addresses which can be translated differs for each SMX group.

[psychocrypt]

@nichtdu Is your issue solved with the provided config options?

[reppolice]

I get the OP's error on my 940MX, which is Maxwell but is rejected by another miner that usually accepts Maxwell, right now it escapes me which miner :) The card has 2GB RAM. I am using the config suggested by xmr

"gpu_threads_conf" : [ { "index" : 0, "threads" : 64, "blocks" : 9, "bfactor" : 6, "bsleep" : 25, "affine_to_cpu" : false, }, ],

[psychocrypt]

@reppolice Do you downloaded thw latest version? Please post the full.command line output.

[TrojanZA]

Hi, I get a similar issue to the OP

GPU 0: unspecified launch failure C:/Users/MAIN/Documents/Visual Studio 2015/Projects/xmr-stak-nvidia/xmr-stak-nvidia/nvcc_code/cuda_extra.cu line 246

I am running Windows 10, NVIDIA 940MX. I downloaded on the 24 July 2017, but I am not sure how to confirm the version number.

[psychocrypt]

@TrojanZA could you please readuce the number of threads maybe by a factor of two. I think your gou runs out of memory but I am not sure.

[TrojanZA]

That's done it, thanks!

[psychocrypt]

@TrojanZA You should slowly increase the threads to find a optimum without crashes. Thx for the information that the reduction is solving the issue. Now I can think about a solution to avoid such unexpected crashes.

[TrojanZA]

Sorry - bit late in the evening. Was looking at CPU miner. Reducing threads did not resolve.

[TrojanZA]

44 threads seems to be stable. Thanks. Hash rate is really low 64.9. Any tips to get that higher?

[pwrpt]

im using a Asus strix gtx 1060 oc and Asus strix gtx 1070 oc, and im getting this error:

GPU 0: unspecified launch failure C:/Users/MAIN/Documents/Visual Studio 2015/Projects/xmr-stak-nvidia/xmr-stak-nvidia/nvcc_code/cuda_extra.cu line 246 GPU 1: unspecified launch failure C:/Users/MAIN/Documents/Visual Studio 2015/Projects/xmr-stak-nvidia/xmr-stak-nvidia/nvcc_code/cuda_extra.cu line 246

And my config is:

"gpu_threads_conf" : [ { "index" : 0, "threads" : 64, "blocks" : 45, "bfactor" : 6, "bsleep" : 25, "affine_to_cpu" : false, }, { "index" : 1, "threads" : 64, "blocks" : 30, "bfactor" : 6, "bsleep" : 25, "affine_to_cpu" : false, }, ],

Thank you,

[faithsloan]

I changed mine by dialing down my threads from 32 to 8 and changing bfactor to 0 and it no longer closes abruptly. "gpu_threads_conf" : [ { "index" : 0, "threads" : 8, "blocks" : 15, "bfactor" : 0, "bsleep" : 25, "affine_to_cpu" : false, }, ],