-
Notifications
You must be signed in to change notification settings - Fork 3.4k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[v4.1.0-beta] Xmrig hangs after a minute #1199
Comments
Please show miner output from both versions. |
I also couln't get it to work with my miners, all are RX570 and RX580.. it compiles for each GPU but it doesn't actually start mining on RandomX(rx) with benchmark pool. |
Compile may take very long time, depends of CPU, many minutes, once it completed miner should cache compilation result for future use. |
here is a screenshot of 8x rx580 rig.. after all the compilation completed, from time to time it mines with 12.3 hashes/s, this rig with xmrig-amd does 7300 hashes per second.. it looks like only one of the cards actually tries to mine.. driver is version 19.5.2 Also it crashes randomly, and takes whole rig with it, and needs hard reset |
Please show config file, you mention you try mine RandomX, but 7300 H/s not possible on 8 rx580. |
{ |
It not looks good, miner should fill |
{ |
On GPUs with 4 GB of memory (where Another option, run only one 1 thread on that GPUs I mark this issue as bug, due workaround of another AMD bug, miner use more memory that expected by autoconfig and 4 GB GPUs running out of memory. |
@komatom can you confirm suggestions works for you or not. |
@xmrig I have done several tests with the settings with your recommendations it turns out 1 thread but higher intesity is the best performing, each test was around 5 min long. And they run without need of the dataset_host enabled in that case..
|
Fixed in evo branch, now original configuration with 2 threads should work, however hashrate seems ok, according earlier reports SChernykh/RandomX_OpenCL#5 |
Thank you very much.. Can you point me out to the workaround issue that caused 4GB VRAM to not be enough on rx580 carads? |
6bc217e but this commit part of much larger refactoring f60118e AMD driver not release memory buffers bellow 1 GB, it some kind of optimization and not issue when use similar algorithms, memory not actually leak in this case, but if switch between very different algorithms in runtime (cn <-> rx <-> cn) it make GPU out of memory. intesity 448 it about 896 MB of memory, but miner was increase buffer size to 1024 MB, so 1024 + 1024 + 2080 (RandomX dataset) is larger that 4 GB, now memory buffer shared between threads and this is already larger than 1 GB ((448 + 448) * 2 = ~1792 MB), so no extra unexpected memory required. |
@xmrig so I have built the latest evo branch. Should I expect autoconfig to set dataset_host to true or the workaround above was changed so 4GB can run. GPUs still can't run out of the box.. I also rely on the command options instead of the config file, so each rig autoconfig can set the appropriate settings, without me having to enter a separate config. it doesn't work on rigs with 8GB too. here is the command line xmrig.exe --opencl --no-cpu --http-port 80 -o randomx-benchmark.xmrig.com:7777 -u BENCHMARK.TEST -p x -k I am not sure if we have applied a solution for this.. Thanks |
| OPENCL # | AFFINITY | 10s H/s | 60s H/s | 15m H/s | |
To continue on this issue, I have tested on the other rigs too, only 8GB GPUs work.. definetely 4GB VRAM is not enough |
I have some RX580 with 4Gb and some with 8Gb. |
@bizlevel more on the same issue is on #1336 rx580 gets around ~400/420 hash/s |
CPU: i3-9100F
GPU: RX570-4Gb
OS: Win10 1903
Running from cli:
xmrig.exe --donate-level 4 --no-cpu --opencl -a cn-lite/1 -o ... -u ... -p x --print-time=5
Jobs are received from pool, but no hashrate prints (all nulls) for about a minute, then driver hangs and gets restored.
AMD driver v. 19.9.2
Xmrig-AMD v2.14.6 works perfectly! Of course, w/o specific cli keys.
I checked the launch with config file, and xmrig-amd works perfectly with the same 2-thread-per-gpu config.
The text was updated successfully, but these errors were encountered: