Open kenshin6353 opened 6 years ago
The same for CUDA 10!
Depends of hardware, GPUs with compute capability >= 6.0 works well with same settings as before, hashrate drop is few %.
compute capability 5.x required different settings threads: X, blocks: 8 * SMX
where X is some magic value, eg 8 or 12. Autoconfig NOT handle it. Hashrate drop is higher than modern GPUs.
Older GPUs with compute capability less 5.0 no good news.
hello !
i have a gtx970m and drop hash when you say 8 * SMX what is SMX please ?
THX !
what is SMX please ?
SMX can be seen when starting the miner.
I have got a few 650Ti cards and hashrates have dropped badly. With a 650Ti Boost I was getting over 220H/s previously, now it's 150H/s. I've tried compiling on Linux with GCC and on Windows with MSVC 2017 and both CUDA 9 and CUDA 10 and the results are the same. Is there anything that can be done to improve performance on these older cards?
550Ti, CUDA 8, Ubuntu 16.04, hashrate drop: 40% with autodetect. Autodetect set Threads: 44, blocks 8. As per xmrig suggested, I've played a little with Threads and blocks to Threads: 12, blocks 32 and the drop was diminished to 20% Also, I had to recompile and upgrade CUDA because previous build was using CUDA 7.5 succesfully, but latest build gave me errors so maybe some of the hashrate change is due to cn/2 and some is due to CUDA7.5 -> CUDA8
hashrate drop 60% for cuda 8? like seriously? lol
same too
2.8.3=>2.8.4 hashrate drop again (20%) on some old gpus or hashrate not changed on others gpus
@miningmeister please show all information, GPU type and config (first lines from miner is ok) and hashrate for all cases, without this information I can say nothing. Thank you.
$ xmrig-nvidia --version
XMRig-NVIDIA 2.8.3
built on Oct 19 2018 with GCC 7.3.1
features: 64-bit AES
libuv/1.23.2
OpenSSL/1.1.1
$ xmrig-nvidia -o stratum+tcp://xmr.pool.minergate.com:45700 -u $USER -p x --cuda-max-threads=24
# hashrate is around 215H/s
$ xmrig-nvidia --version
XMRig-NVIDIA 2.8.4
built on Oct 29 2018 with GCC 7.3.1
features: 64-bit AES
libuv/1.23.2
OpenSSL/1.1.1
$ xmrig-nvidia -o stratum+tcp://xmr.pool.minergate.com:45700 -u $USER -p x --cuda-max-threads=24
# hashrate is around 140H/s
$ nvidia-smi
Mon Oct 29 16:01:16 2018
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 410.66 Driver Version: 410.66 CUDA Version: 10.0 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce GTX 960M On | 00000000:01:00.0 Off | N/A |
| N/A 40C P8 N/A / N/A | 0MiB / 2004MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
GeForce GT 440 arch:21 SMX:2 before hardfork 83h/s ("threads": 64, "blocks": 4, bfactor 6) - best settings after hf 2.8.1 53h/s ("threads": 16, "blocks": 20, bfactor 8) - best settings 2.8.4 ~41h/s ("threads": 16, "blocks": 20, bfactor 8)
GeForce GTX 550ti arch:21 SMX:4 before hardfork 177h/s ("threads": 48, "blocks": 8, bfactor 6) - best settings after hf 2.8.1 96h/s hh("threads": 32, "blocks": 14, bfactor 8) - best settings 2.8.4 ~103h/s ("threads": 32, "blocks": 14, bfactor 8)
gtx660 and higher is ok
@RBlaine95 I meant miner output right after start, it contains all useful information including threads and blocks, autoconfig is different, v2.8.4 use just only 4 threads (safe, but not best) and much more blocks (8 * SMX), you should increase threads count to higher values eg 8.
@miningmeister arch:21 is too slow for cn/2
but did you try much lower threads count?
Thank you.
@xmrig my bad, here's the output:
$ xmrig-nvidia --donate-level=0 -o stratum+tcp://xmr.pool.minergate.com:45700 -u $USER -p x --cuda-max-threads=24
* ABOUT XMRig-NVIDIA/2.8.3 gcc/7.3.1
* LIBS libuv/1.23.2 CUDA/10.0 OpenSSL/1.1.1
* CPU Intel(R) Core(TM) i7-6700HQ CPU @ 2.60GHz x64 AES
* GPU #0 PCI:0000:01:00 GeForce GTX 960M @ 1176/2505 MHz 24x15 0x0 arch:50 SMX:5
* ALGO cryptonight, donate=0%
* POOL #1 stratum+tcp://xmr.pool.minergate.com:45700 variant 2
* COMMANDS hashrate, health, pause, resume
#---
$ xmrig-nvidia --donate-level=0 -o stratum+tcp://xmr.pool.minergate.com:45700 -u $USER -p x --cuda-max-threads=24
* ABOUT XMRig-NVIDIA/2.8.4 gcc/7.3.1
* LIBS libuv/1.23.2 CUDA/10.0 OpenSSL/1.1.1
* CPU Intel(R) Core(TM) i7-6700HQ CPU @ 2.60GHz x64 AES
* GPU #0 PCI:0000:01:00 GeForce GTX 960M @ 1176/2505 MHz 4x40 0x0 arch:50 SMX:5
* ALGO cryptonight, donate=0%
* POOL #1 stratum+tcp://xmr.pool.minergate.com:45700 variant 2
* COMMANDS hashrate, health, pause, resume
Edit: (to prevent spam)
Using --cuda-launch=24x20
instead of --cuda-max-threads=24
(which seems to be ignored in 2.8.4
) fixes my hash rate and brings it back to normal (±10 extra H/s)
Hello guys !
I have built a second little RIG with 3 nvidia cards (MSI GTX 670, 760 and 770). I began with Minergate for some stupid reasons... but 2KH/s it was correct. So, I try with XMRig and surprise... 280H/s Cuda 9 installed, not the last driver nvidia....I try to change the Thread and Block and have some reboot ^^ I use the pool of minergate to finish and have my payment, Is this a pbm ?
Some idees ? THX !
https://www.flickr.com/gp/165769524@N07/mW725G
A little inproove with Thread modification at 8 and block at 8xSMX So 429H/s now.... Can i hope best results ?
@fred62mail 2KH/s is very high HR for your rig. I suppose you can squeeze out near 0.8-1KH/s. Try to play with threads/blocks. Blocks will be 8xSMX, you're right, threads can be something between 8 and num of cuda cores divided by num of blocks (8xSMX). And double check your config file!
@miningmeister arch:21 is too slow for
cn/2
but did you try much lower threads count? Thank you.
yes, i try. threads": 16, "blocks": 20 - best hashrate
hashrate drop 60% for cuda 8? like seriously? lol