search

xmrig / xmrig-cuda

NVIDIA CUDA plugin for XMRig miner
GNU General Public License v3.0
358 stars 155 forks source link

[BUG] Out of memory with new CUDA Plugin v6.15.0 / Xmrig v6.15.0 #118

Closed RainbowMiner closed 2 years ago

RainbowMiner commented 2 years ago

The new CUDA plugin v6.15.0 + Xmrig v6.15.0 seems to have memory calculation problems on Windows 10, Nvidia Driver v466.74. This happens for many algorithms, including AstroBWT, RandomX, RandomXGraft, KawPow. I have been using the current CUDA 11.3 download from the current release. All cryptonight algorithms plus RandomArq mining worked good. For testing, I have replaced the CUDA plugin with the prior v6.12.0 + Xmrig v6.15.0 and the memory problems don't appear any more.

 * ABOUT        XMRig/6.15.0 MSVC/2019
 * LIBS         libuv/1.42.0 OpenSSL/1.1.1l hwloc/2.5.0
 * HUGE PAGES   permission granted
 * 1GB PAGES    unavailable
 * CPU          Intel(R) Pentium(R) CPU G4560 @ 3.50GHz (1) 64-bit AES
                L2:0.5 MB L3:3.0 MB 2C/4T NUMA:1
 * MEMORY       4.8/7.9 GB (61%)
                ChannelA-DIMM0: <empty>
                DIMM_A1: 4 GB DDR4 @ 2133 MHz KHX2133C14D4/4G
                ChannelB-DIMM0: <empty>
                DIMM_B1: 4 GB DDR4 @ 2133 MHz KHX2133C14D4/4G
 * MOTHERBOARD  MSI - Z170A GAMING PRO CARBON (MS-7A12)
 * DONATE       1%
 * ASSEMBLY     auto:intel
 * POOL #1      stratum+tcp://kawpow.mine.zergpool.com:3638 algo kawpow
 * COMMANDS     hashrate, pause, resume, results, connection
 * HTTP API     127.0.0.1:35006
 * OPENCL       disabled
 * CUDA         11.3/11.3/6.15.0
 * NVML         disabled
 * CUDA GPU     #6 09:00.0 NVIDIA GeForce RTX 3070 1905/7001 MHz smx:46 arch:86 mem:6686/8192 MB
[2021-08-31 11:26:53.874]  net      use pool kawpow.mine.zergpool.com:3638  51.210.180.98
[2021-08-31 11:26:53.875]  net      new job from kawpow.mine.zergpool.com:3638 diff 2155M algo kawpow height 1909397
[2021-08-31 11:26:53.875]  nvidia   use profile  kawpow  (1 thread) scratchpad 32 KB
|  # | GPU |  BUS ID | INTENSITY | THREADS | BLOCKS | BF |  BS | MEMORY | NAME
|  0 |   6 | 09:00.0 |  24117248 |     256 |  94208 |  6 |  25 |   3056 | NVIDIA GeForce RTX 3070
[2021-08-31 11:26:55.077]  nvidia   thread #0 failed with error <cryptonight_extra_cpu_init>:379 "out of memory"
[2021-08-31 11:26:55.172]  nvidia   thread #0 self-test failed
[2021-08-31 11:26:55.172]  nvidia   disabled (failed to start threads)
RainbowMiner commented 2 years ago

Here is another one, trying to mine RandomX

 * CUDA GPU     #6 09:00.0 NVIDIA GeForce RTX 3070 1905/7001 MHz smx:46 arch:86 mem:6686/8192 MB
[2021-08-31 11:41:29.694]  net      use pool rx-eu.unmineable.com:3333  159.65.30.104
[2021-08-31 11:41:29.694]  net      new job from rx-eu.unmineable.com:3333 diff 100001 algo rx/0 height 2439047 (42 tx)
[2021-08-31 11:41:29.695]  cpu      use argon2 implementation SSSE3
[2021-08-31 11:41:29.750]  msr      register values for "intel" preset have been set successfully (55 ms)
[2021-08-31 11:41:29.750]  randomx  init dataset algo rx/0 (4 threads) seed 872505fbda609445...
[2021-08-31 11:41:29.779]  randomx  allocated 2336 MB (2080+256) huge pages 0% 0/1168 +JIT (29 ms)
[2021-08-31 11:41:35.492]  net      new job from rx-eu.unmineable.com:3333 diff 100001 algo rx/0 height 2439047 (42 tx)
[2021-08-31 11:41:43.142]  randomx  dataset ready (13363 ms)
[2021-08-31 11:41:43.142]  nvidia   use profile  rx  (1 thread) scratchpad 2048 KB
|  # | GPU |  BUS ID | INTENSITY | THREADS | BLOCKS | BF |  BS | MEMORY | NAME
|  0 |   6 | 09:00.0 |      2208 |      32 |     69 |  6 |  25 |   4416 | NVIDIA GeForce RTX 3070
[2021-08-31 11:41:44.173]  nvidia   READY threads 1/1 (1028 ms)
[2021-08-31 11:41:48.183]  miner    speed 10s/60s/15m n/a n/a n/a H/s max n/a H/s
[2021-08-31 11:41:48.827]  nvidia   thread #0 failed with error <randomx_prepare>:43 "out of memory"

This is AstroBWT:

 * CUDA GPU     #6 09:00.0 NVIDIA GeForce RTX 3070 1905/7001 MHz smx:46 arch:86 mem:6686/8192 MB
[2021-08-31 11:41:29.694]  net      use pool rx-eu.unmineable.com:3333  159.65.30.104
[2021-08-31 11:41:29.694]  net      new job from rx-eu.unmineable.com:3333 diff 100001 algo rx/0 height 2439047 (42 tx)
[2021-08-31 11:41:29.695]  cpu      use argon2 implementation SSSE3
[2021-08-31 11:41:29.750]  msr      register values for "intel" preset have been set successfully (55 ms)
[2021-08-31 11:41:29.750]  randomx  init dataset algo rx/0 (4 threads) seed 872505fbda609445...
[2021-08-31 11:41:29.779]  randomx  allocated 2336 MB (2080+256) huge pages 0% 0/1168 +JIT (29 ms)
[2021-08-31 11:41:35.492]  net      new job from rx-eu.unmineable.com:3333 diff 100001 algo rx/0 height 2439047 (42 tx)
[2021-08-31 11:41:43.142]  randomx  dataset ready (13363 ms)
[2021-08-31 11:41:43.142]  nvidia   use profile  rx  (1 thread) scratchpad 2048 KB
|  # | GPU |  BUS ID | INTENSITY | THREADS | BLOCKS | BF |  BS | MEMORY | NAME
|  0 |   6 | 09:00.0 |      2208 |      32 |     69 |  6 |  25 |   4416 | NVIDIA GeForce RTX 3070
[2021-08-31 11:41:44.173]  nvidia   READY threads 1/1 (1028 ms)
[2021-08-31 11:41:48.183]  miner    speed 10s/60s/15m n/a n/a n/a H/s max n/a H/s
[2021-08-31 11:41:48.827]  nvidia   thread #0 failed with error <randomx_prepare>:43 "out of memory"

Here is RandomXGraft

 * CUDA GPU     #6 09:00.0 NVIDIA GeForce RTX 3070 1905/7001 MHz smx:46 arch:86 mem:6686/8192 MB
[2021-08-31 12:21:32.369]  net      use pool gulf.moneroocean.stream:10032  195.201.124.214
[2021-08-31 12:21:32.370]  net      new job from gulf.moneroocean.stream:10032 diff 45744 algo rx/graft height 945359 (1 tx)
[2021-08-31 12:21:32.370]  cpu      use argon2 implementation SSSE3
[2021-08-31 12:21:32.400]  msr      register values for "intel" preset have been set successfully (30 ms)
[2021-08-31 12:21:32.400]  randomx  init dataset algo rx/graft (4 threads) seed 7f856bda85faa24c...
[2021-08-31 12:21:32.582]  randomx  allocated 2336 MB (2080+256) huge pages 100% 1168/1168 +JIT (182 ms)
[2021-08-31 12:21:41.670]  randomx  dataset ready (9088 ms)
[2021-08-31 12:21:41.671]  nvidia   use profile  rx  (1 thread) scratchpad 2048 KB
|  # | GPU |  BUS ID | INTENSITY | THREADS | BLOCKS | BF |  BS | MEMORY | NAME
|  0 |   6 | 09:00.0 |      2208 |      32 |     69 |  6 |  25 |   4416 | NVIDIA GeForce RTX 3070
[2021-08-31 12:21:42.480]  nvidia   READY threads 1/1 (809 ms)
[2021-08-31 12:21:46.109]  nvidia   thread #0 failed with error <randomx_prepare>:43 "out of memory"
RainbowMiner commented 2 years ago

Here is RandomX mining with the prior CUDA plugin v6.12.0:

 * ABOUT        XMRig/6.15.0 MSVC/2019
 * LIBS         libuv/1.42.0 OpenSSL/1.1.1l hwloc/2.5.0
 * HUGE PAGES   permission granted
 * 1GB PAGES    unavailable
 * CPU          Intel(R) Pentium(R) CPU G4560 @ 3.50GHz (1) 64-bit AES
                L2:0.5 MB L3:3.0 MB 2C/4T NUMA:1
 * MEMORY       3.7/7.9 GB (47%)
                ChannelA-DIMM0: <empty>
                DIMM_A1: 4 GB DDR4 @ 2133 MHz KHX2133C14D4/4G
                ChannelB-DIMM0: <empty>
                DIMM_B1: 4 GB DDR4 @ 2133 MHz KHX2133C14D4/4G
 * MOTHERBOARD  MSI - Z170A GAMING PRO CARBON (MS-7A12)
 * DONATE       1%
 * ASSEMBLY     auto:intel
 * POOL #1      stratum+tcp://rx-eu.unmineable.com:3333 algo rx/0
 * COMMANDS     hashrate, pause, resume, results, connection
 * HTTP API     127.0.0.1:35006
 * OPENCL       disabled
 * CUDA         11.3/11.3/6.12.0
 * NVML         disabled
 * CUDA GPU     #6 09:00.0 NVIDIA GeForce RTX 3070 1905/7001 MHz smx:46 arch:86 mem:6686/8192 MB
[2021-08-31 13:03:33.101]  net      use pool rx-eu.unmineable.com:3333  159.65.30.104
[2021-08-31 13:03:33.102]  net      new job from rx-eu.unmineable.com:3333 diff 100001 algo rx/0 height 2439079 (82 tx)
[2021-08-31 13:03:33.102]  cpu      use argon2 implementation SSSE3
[2021-08-31 13:03:33.191]  msr      register values for "intel" preset have been set successfully (89 ms)
[2021-08-31 13:03:33.192]  randomx  init dataset algo rx/0 (4 threads) seed 872505fbda609445...
[2021-08-31 13:03:33.205]  randomx  allocated 2336 MB (2080+256) huge pages 100% 1168/1168 +JIT (13 ms)
[2021-08-31 13:03:47.145]  randomx  dataset ready (13939 ms)
[2021-08-31 13:03:47.145]  nvidia   use profile  rx  (1 thread) scratchpad 2048 KB
|  # | GPU |  BUS ID | INTENSITY | THREADS | BLOCKS | BF |  BS | MEMORY | NAME
|  0 |   6 | 09:00.0 |      2208 |      32 |     69 |  6 |  25 |   4416 | NVIDIA GeForce RTX 3070
[2021-08-31 13:03:47.879]  nvidia   READY threads 1/1 (733 ms)
[2021-08-31 13:03:51.328]  net      new job from rx-eu.unmineable.com:3333 diff 100001 algo rx/0 height 2439079 (82 tx)
[2021-08-31 13:03:52.150]  miner    speed 10s/60s/15m n/a n/a n/a H/s max n/a H/s
[2021-08-31 13:04:02.219]  net      new job from rx-eu.unmineable.com:3333 diff 100001 algo rx/0 height 2439080 (19 tx)
[2021-08-31 13:04:02.297]  nvidia   accepted (1/0) diff 100001 (71 ms)

.. works perfect.

Also, AstroBWT mining with CUDA plugin v6.12.0 with xmrig v6.15.0:

 * OPENCL       disabled
 * CUDA         11.3/11.3/6.12.0
 * NVML         disabled
 * CUDA GPU     #6 09:00.0 NVIDIA GeForce RTX 3070 1905/7001 MHz smx:46 arch:86 mem:6686/8192 MB
[2021-08-31 12:57:30.190]  net      use pool gulf.moneroocean.stream:10032  195.201.124.214
[2021-08-31 12:57:30.202]  net      new job from gulf.moneroocean.stream:10032 diff 4323 algo astrobwt height 6114555
[2021-08-31 12:57:30.203]  nvidia   use profile  astrobwt  (1 thread) scratchpad 20480 KB
|  # | GPU |  BUS ID | INTENSITY | THREADS | BLOCKS | BF |  BS | MEMORY | NAME
|  0 |   6 | 09:00.0 |       672 |      32 |     21 |  6 |  25 |   6248 | NVIDIA GeForce RTX 3070
[2021-08-31 12:57:31.190]  nvidia   READY threads 1/1 (987 ms)
[2021-08-31 12:57:35.373]  miner    speed 10s/60s/15m n/a n/a n/a H/s max n/a H/s
[2021-08-31 12:57:35.501]  nvidia   accepted (1/0) diff 4323 (60 ms)
tymoteuszrogalewski commented 2 years ago

Yea, I also have in this version similar errors. And see that under Ubuntu16 it stops compiling (gcc-5) Must use Ubuntu18 with gcc-7

[2021-09-01 11:13:56.030] nvidia thread #0 failed with error _cryptonight_extra_cpuinit:384 out of memory [2021-09-01 11:13:56.044] nvidia thread #4 self-test failed [2021-09-01 11:13:56.056] nvidia thread #2 failed with error _cryptonight_extra_cpuinit:384 out of memory [2021-09-01 11:13:56.100] nvidia thread #5 self-test failed [2021-09-01 11:13:56.107] nvidia thread #0 self-test failed [2021-09-01 11:13:56.147] nvidia thread #2 self-test failed [2021-09-01 11:13:56.153] nvidia thread #3 failed with error _cryptonight_extra_cpuinit:384 out of memory [2021-09-01 11:13:56.201] nvidia thread #3 self-test failed

Prev versions of xmrig and xmrig-cuda works ok.

commandline options used: --opencl --cuda --no-cpu -o stratum.ravenminer.com:3838 -a kawpow -u RH2xxxxxxxxxx -p x --print-time=5 --http-port=3333

xboxplayer9889 commented 2 years ago

nvidia thread #0 failed with error :43 "out of memory"

I've got the same with xmrig-cuda-6.15.0, and works well with 6.12.0:

I don't know it's a bug or just e.g. block sizes has changed, but I've modified thread/block count and works well:

This is now the fastest (biggest h/s) settings, I could say the optimal for my card. It is exactly the half of 6.12's with the same speed. ( 6.12's optimal was 48 threads and 16 blocks )

xboxplayer9889 commented 2 years ago

https://github.com/xmrig/xmrig-cuda/pull/124 solved for me, again available 48threads/16blocks for me..

RainbowMiner commented 2 years ago

Solved. Thank you!