Skip to content

Fix CUDA8 crashing on 20; add true 21 support; add memory size display in Summary#255

Merged
xmrig merged 1 commit into
xmrig:devfrom
Spudz76:dev-fixCUDA8
Mar 21, 2019
Merged

Fix CUDA8 crashing on 20; add true 21 support; add memory size display in Summary#255
xmrig merged 1 commit into
xmrig:devfrom
Spudz76:dev-fixCUDA8

Conversation

@Spudz76

@Spudz76 Spudz76 commented Mar 20, 2019

Copy link
Copy Markdown
Contributor

CUDA8 would crash on init randomly but constantly

moving the memory-size-check to before the cudaGetDeviceProperties fixes it

Bonus, now the Summary line shows GPU memory sizes at startup

I have not seen another startup crash since applying this on my running branch, NVS5200M arch 21.

Also adds true 2.1 support which was at least 0.6H/s (2%) faster on CN-GPU when it's the only one included via -DCUDA_ARCH=21 - even slightly faster on easier variants like RWZ

Moved the deprecated targets silence flag to a proper location (it had been adding it once per arch within the loop, wrongly)

@Spudz76
CUDA8 would crash on init randomly but constantly: moving the memory-…
4570e5d 
…size-check to before the cudaGetDeviceProperties fixes it; Bonus, now the Summary line shows GPU memory sizes at startup
@Spudz76 Spudz76 changed the title CUDA8 would crash on init randomly but constantly: moving the memory-… Fix CUDA8 crashing on 20; add true 21 support; add memory size display in Summary Mar 20, 2019
@Spudz76

Spudz76 commented Mar 20, 2019

Copy link
Copy Markdown
Contributor Author

Essentially bisected to when xmr-stak had the same init problems, it was never real clear what fixed it but it was right when they added the memory display in their summary (which required moving the section ahead of the properties call, which fixed it like magic for no reason anyone can find)

I simply duplicated that move and then it works perfectly. Definitely not documented, and doesn't seem to matter anymore in 9.0-10.1, so it was an undocumented bug that had apparently been fixed but not in the final 8.0

@Spudz76

Spudz76 commented Mar 20, 2019

Copy link
Copy Markdown
Contributor Author

It seems like I tested a 2.1 build before and it made no difference but the trick may be to compile ONLY 2.1 by itself with this patch (so that 2.0 generic layer is not added whatsoever) and then I see the above gains.

This was referenced Mar 20, 2019
Open
Open
@Spudz76

Spudz76 commented Mar 21, 2019

Copy link
Copy Markdown
Contributor Author

Rotated this build into my other GTX970 (5.2) rig, works great there too although nothing was previously broken on this type of GPU

Here's a successful jump from CN-R over to CN-GPU with no crashes and the memory info correct

2019-03-20 18:49:14] speed 10s/60s/15m 388.5 386.7 n/a H/s max 390.6 H/s
2019-03-20 18:49:14]  * GPU #0: 1278/3947 MHz 111W 72C FAN 31%
2019-03-20 18:49:23] accepted (18/0) diff 13288 (61 ms)
>> Starting miner './xmrig-nvidia101 --config=config-gpu.json' to process new cn/gpu algo
>> Stopping './xmrig-nvidia101 --config=config-gpu.json' miner
!! Miner socket error
!! Miner './xmrig-nvidia101 --config=config-r.json' exited with nonzero code 1
>> Starting miner: ./xmrig-nvidia101 --config=config-gpu.json
* ABOUT        XMRig-NVIDIA/2.14.2-dev MSVC/2017
* LIBS         libuv/1.23.0 CUDA/10.10 OpenSSL/1.1.1 microhttpd/0.9.59
* CPU                  Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz x64 AES
* GPU #0       PCI:0000:01:00 GeForce GTX 970 @ 1177/3945 MHz 26x39 6x100 arch:52 SMX:13 MEM:3914/4096 MiB
* ALGO         cryptonight, donate=5%
* POOL #1      127.0.0.1:3334 variant=gpu
* API BIND     [::]:10081
* COMMANDS     'h' hashrate, 'e' health, 'p' pause, 'r' resume
>> Miner server on 127.0.0.1:3334 port connected from 127.0.0.1
>> Pool (gulf.moneroocean.stream:ssl20008) <-> miner link was established due to new miner connection
2019-03-20 18:49:44] use pool 127.0.0.1:3334  127.0.0.1
2019-03-20 18:49:44] new job from 127.0.0.1:3334 diff 16557 algo cn/gpu height 240667
2019-03-20 18:50:39] accepted (1/0) diff 16557 (191 ms)
2019-03-20 18:50:50] speed 10s/60s/15m 486.2 487.3 n/a H/s max 488.6 H/s
2019-03-20 18:50:50]  * GPU #0: 1164/3947 MHz 112W 76C FAN 34%

@xmrig xmrig merged commit 4570e5d into xmrig:dev Mar 21, 2019
@xmrig

xmrig commented Mar 21, 2019

Copy link
Copy Markdown
Owner

I revert static_cast it was absolutely fine + other minor/not code related changes 8913809
Thank you.

@Spudz76

Spudz76 commented Mar 24, 2019

Copy link
Copy Markdown
Contributor Author

idk, CLion introspector told me dynamic_cast was correct given the inheritance or whatever
it's all greek to me I just prefer when my editor isn't highlighting and complaining :}
and it seems it does suggest good ideas often
but mostly I have no idea unless one way doesn't run

I'll leave them lit up like an airport instead, maybe I can figure out how to disable that code suggestion rule item

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@Spudz76 @xmrig