[ANN]Bminer: a fast Equihash/Ethash/Cuckaroo29z miner for AMD/NVIDIA GPUs 16.4.9

[hash8]

> 1070ti on windows 7 for c31 algo. insufficient memory.
+1
"Insufficient memory 7847.66 MB available on device"
418.81 drivers

[cryptoyes]

Besides the malloc crash above, I also get lots of these:

Code:

[INFO] [2019-02-18T01:09:07+02:00] CUDA error DRIVER: '709' in func 'CopyEdges' line 174

or

Code:

[INFO] [2019-02-18T01:48:34+02:00] Received new job 90412
[FATA] [2019-02-18T01:48:38+02:00] Irrecoverable errors: [D7] GPU hang detected

or without any actual error:

Code:

[INFO] [2019-02-18T01:26:49+02:00] Total 9.30 G/s Accepted shares 55 Rejected shares 0 
[INFO] [2019-02-18T01:26:51+02:00] Received new job 0                           
[WARN] [2019-02-18T01:26:57+02:00] Miner died! It will be restarted soon...     

14.0.0 was the most stable. ALL VERSIONS AFTER 14.0.0 crash very often like above. Please fix these!

Trying to reproduce. What's the configuration of the machine, as well as software versions, etc.?

The more information the better. Thanks

I already provided all of those to you in the previous post, you even replied to it! https://bitcointalk.org/index.php?topic=2519271.msg49794934#msg49794934

Once again: Ubuntu 16.04, Nvidia 415.27, CUDA 10, GTX1070, GTX1070Ti, GTX1080Ti. Motherboard is ASRock H110 Pro BTC+ (13 PCIe slots). Intel G3930, Intel G4600, Intel i3-7100.

Crashes happen on all machines (it's a farm). No malloc crashes and no CUDA "CopyEdges" or "CollectEdges" crashes on 14.x. Constant crashes on 15.0 for all of these.

[cryptoyes]

Once again: Ubuntu 16.04, Nvidia 415.27, CUDA 10, GTX1070, GTX1070Ti, GTX1080Ti. Motherboard is ASRock H110 Pro BTC+ (13 PCIe slots). Intel G3930, Intel G4600, Intel i3-7100.

Crashes happen on all machines (it's a farm). No malloc crashes and no CUDA "CopyEdges" or "CollectEdges" crashes on 14.x. Constant crashes on 15.0 for all of these.

[admco]

Hi, hopefully this helps on the Windows 10 side of bug fixes for Grin31.

It seems that if there are spaces in the folder path, bminer quits with:

Code:

The system cannot find the path specified.

I had the miner working with two GPUs, but after Windows updated from build 1803 to build 1809 it stopped working with the following error:

Code:

[FATA] [2019-02-18T17:59:35-07:00] CudaError: Unknown error:C:\WINDOWS\System32\bcrypt.dll
[WARN] [2019-02-18T17:59:35-07:00] Miner died! It will be restarted soon...

I have another rig on build 1803 and it seems to be working.

I've also encountered OpenCL errors which are fixed by doing a clean install of the NVIDIA drivers.

[sharmanov]

So.... even though it sais AMD in your title your miner doesn't support AMD cards?

[cryptoyes]

Once again: Ubuntu 16.04, Nvidia 415.27, CUDA 10, GTX1070, GTX1070Ti, GTX1080Ti. Motherboard is ASRock H110 Pro BTC+ (13 PCIe slots). Intel G3930, Intel G4600, Intel i3-7100.

Crashes happen on all machines (it's a farm). No malloc crashes and no CUDA "CopyEdges" or "CollectEdges" crashes on 14.x. Constant crashes on 15.0 for all of these.

To give you an idea of how often it crashes: 648 crashes in the past 24h for 17 different rigs. On average, *every* rig crashes about once every 40 minutes ... super wasteful.

[NimaV]

I can't mine grin31 with GTX 1080ti on Windows 10 build 1809. Bminer reports 9170MB free memory on the device and gives me insufficient memory error. I've checked it on several monitoring softwares and there is exactly 11264MB free dedicated GPU memory on the device. windows is running on another GPU and it does not use any amount of the GTX 1080ti memroy. why Bminer reports more than 2GB less free memory on my 1080ti? is there any way to fix this?

Also Bminer 15 doesn't work with my rig because of bcrypt.dll error.

[admco]

Confirmed, it seems to be a Window 10 1809 issue. Builds 1803 and 1709 seem to work fine. The must have made changes to the bcrypt.dll library. I've tried using the 1803 version of the library on 1809 with no luck.

On a positive note, the Windows version appears to be pretty stable.
I have it running on 6 rigs with no issue. 6x 1080Ti rigs work at full speed.
Titan XP cards are getting 1.4 G/s with a 100 MHz core overclock.

[cryptoyes]

Just to add to data...Getting constant crashes on one of my rigs on c31 w/15.0 release, "CUDA error DRIVER: '700' in func 'CopyEdges' line 173".  Ubuntu 18.04, 1070tis.  Current drivers.

Same here, see my posts above. Very buggy release ... but a major effort from the developer, I have to appreciate that.

[od1n]

HiveOS with v15 and 6x 1070 [Core +100 | MEM +0 | PL 85W]
C29 is mining with 22,8 gps
C31 is mining with 3,8 gps

C31 is running since 10 minutes, I will check if miner dies ...

1 hour update: Miner is still running but accepted share rate is <90%!
With C29 Bminer finally reaced >95% accepted shares.

After 2 hours I have nearly 15% rejected shares  
But sparkpool predicts nevertheless 10% more revenue then with C29 before.

Still running without freeze since 13 hours. But Hive shows 84% valid shares only, sparkpool too.

[cryptoyes]

The new 15.0.0 crashes often. Ubuntu 16.04, Nvidia 414.27, CUDA 10.0, GTX 1070/1070ti/1080ti

Code:

[INFO] [2019-02-18T00:51:38+02:00] [GPU 9] Speed: 0.76 G/s Fidelity 0.689 Temp: 34C Fan: 74% Power: 107W 0.01 G/J
[INFO] [2019-02-18T00:51:38+02:00] [GPU 10] Speed: 0.72 G/s Fidelity 1.795 Temp: 40C Fan: 75% Power: 99W 0.01 G/J
[INFO] [2019-02-18T00:51:39+02:00] [GPU 11] Speed: 0.74 G/s Fidelity 0.333 Temp: 43C Fan: 75% Power: 109W 0.01 G/J
[INFO] [2019-02-18T00:51:40+02:00] Total 8.88 G/s Accepted shares 3 Rejected shares 0
[INFO] [2019-02-18T00:51:48+02:00] Received new job 90422
*** Error in `/node/bminer-15.0.0': malloc(): memory corruption: 0x00007fa883d43de0 ***
======= Backtrace: =========
/lib/x86_64-linux-gnu/libc.so.6(+0x777e5)[0x7faac48387e5]
/lib/x86_64-linux-gnu/libc.so.6(+0x8213e)[0x7faac484313e]
/lib/x86_64-linux-gnu/libc.so.6(__libc_malloc+0x54)[0x7faac4845184]
/usr/lib/x86_64-linux-gnu/libcuda.so.1(+0x29b773)[0x7faac0921773]
/usr/lib/x86_64-linux-gnu/libcuda.so.1(+0x21892d)[0x7faac089e92d]
/usr/lib/x86_64-linux-gnu/libcuda.so.1(+0x1134c8)[0x7faac07994c8]
/usr/lib/x86_64-linux-gnu/libcuda.so.1(+0x29c8d0)[0x7faac09228d0]
/usr/lib/x86_64-linux-gnu/libcuda.so.1(+0x2bc099)[0x7faac0942099]
/usr/lib/x86_64-linux-gnu/libcuda.so.1(+0x1e966d)[0x7faac086f66d]
/usr/lib/x86_64-linux-gnu/libcuda.so.1(+0xfa8ce)[0x7faac07808ce]
/usr/lib/x86_64-linux-gnu/libcuda.so.1(+0xfd9f3)[0x7faac07839f3]
/usr/lib/x86_64-linux-gnu/libcuda.so.1(cuMemcpyDtoHAsync_v2+0x72)[0x7faac08d0d12]
/tmp/miner[0x7c07be]
/tmp/miner[0x7cb7be]
/tmp/miner[0x7b2fd8]
/tmp/miner[0x7cd846]
/usr/lib/x86_64-linux-gnu/libstdc++.so.6(+0xb8c80)[0x7faac578bc80]
/lib/x86_64-linux-gnu/libpthread.so.0(+0x76ba)[0x7faac4da86ba]
/lib/x86_64-linux-gnu/libc.so.6(clone+0x6d)[0x7faac48c83dd]
======= Memory map: ========
00400000-00401000 r-xp 00000000 00:2d 3063903                            /node/bminer-15.0.0
00401000-00dcc000 r-xp 00000000 00:00 0
00e00000-00e01000 rwxp 00000000 00:2d 3063903                            /node/bminer-15.0.0
00e01000-018c3000 rwxp 00000000 00:00 0
01a00000-01a01000 r-xp 00000000 00:2d 3063903                            /node/bminer-15.0.0
01a01000-01af5000 r-xp 00000000 00:00 0
01af5000-01af7000 -wxp 00000000 00:00 0
01af7000-01ecf000 r-xp 00000000 00:00 0
02000000-02d3b000 rwxp 00000000 00:2d 3063903                            /node/bminer-15.0.0
02f3a000-02fc6000 rwxp 00d3a000 00:2d 3063903                            /node/bminer-15.0.0
03f78000-077f4000 rwxp 00000000 00:00 0                                  [heap]
200000000-200200000 rwxs 00000000 00:06 727                              /dev/nvidiactl
200200000-200400000 ---p 00000000 00:00 0
200400000-200404000 rwxs 00000000 00:06 727                              /dev/nvidiactl
200404000-200600000 ---p 00000000 00:00 0
<snip>

Never saw this in the test. but it's interesting that it only happens in 15.0.0. Will dig into it.

Actually I'm also getting malloc crashes on 14.3.1 when mining C31 on GTX1080Ti (same system, Ubuntu 16.03, Nvidia 415.27, CUDA 10). Never got this error when mining grin29. It seems linked to your c31 code more than the 15.0.0 version.

The CUDA "CopyEdges" and "CollectEdges" errors are new to 15.0.0.

With 15.0.0 I'm also now getting these:

Code:

[INFO] [2019-02-19T07:01:58+02:00] [GPU 11] Speed: 0.68 G/s Fidelity 1.053 Temp: 58C Fan: 93% Power: 83W 0.01 G/J 
[INFO] [2019-02-19T07:01:58+02:00] Total 8.35 G/s Accepted shares 220 Rejected shares 24 
bminer-15.0.0: tpp.c:87: __pthread_tpp_change_priority: Assertion `previous_prio == -1 || (previous_prio >= fifo_min_prio && previous_prio <= fifo_max_prio)' failed.

and also

Code:

[INFO] [2019-02-19T07:00:01+02:00] [GPU 9] Speed: 0.68 G/s Fidelity 0.967 Temp: 36C Fan: 82% Power: 108W 0.01 G/J 
[INFO] [2019-02-19T07:00:02+02:00] [GPU 10] Speed: 0.69 G/s Fidelity 0.877 Temp: 42C Fan: 83% Power: 120W 0.01 G/J 
terminate called after throwing an instance of 'std::bad_alloc'
  what():  std::bad_alloc
SIGABRT: abort
PC=0x7f28a2b0a428 m=0 sigcode=18446744073709551610
signal arrived during cgo execution

[WARN] [2019-02-19T07:00:04+02:00] Miner died! It will be restarted soon...     

Very buggy version

[leoragraves666]

C29 no crashes, but C31... damn I just woke up, on 187 rigs, thousands of crashes

[realbminer]

I can't mine grin31 with GTX 1080ti on Windows 10 build 1809. Bminer reports 9170MB free memory on the device and gives me insufficient memory error. I've checked it on several monitoring softwares and there is exactly 11264MB free dedicated GPU memory on the device. windows is running on another GPU and it does not use any amount of the GTX 1080ti memroy. why Bminer reports more than 2GB less free memory on my 1080ti? is there any way to fix this?

Also Bminer 15 doesn't work with my rig because of bcrypt.dll error.

We are working on fix the bcrypt.dll error. Sorry for the trouble.

[BTCspace]

better performance on 2080Ti card ?

to use turing architecture?

[jabujax]

How to setup API???

forexample: -api=192.168.1.106:8080

in web write this: http://192.168.1.106:8080/api/status

[jabujax]

Guys how are you able to get 1.2G/s per GTX 1080Ti at stock speed? I am unable to go above 0.6G/s per card with stock
Cards are at 50°C, so definitely not an overheating issue. Is here anybody with 6 GTX 1080Ti's on one rig running total at 7G/s?

OS: Win 7 x64
Drivers: 416.34
CPU: G3930

I get 1,33 G/s per 1080ti in a 6 card rig all 1080ti, but I'm using linux...

It could depends on GPU model. I get from one card 1.3 G/s and from the second only 1.1 G/s. Both on WinX system

I am able to get 1.2 G/s per every card, but only for 2 or 3 cards running at one time. When i add GPU 4,5,6 speed of all cards go down to 0.6 G/s.

make separate bat files for every gpu. Same problem occured for me i solved like this.

[jabujax]

> 1070ti on windows 7 for c31 algo. insufficient memory.
+1
"Insufficient memory 7847.66 MB available on device"
418.81 drivers

mee too same problem. İs there anyone who can run 1070ti on win7?

[jabujax]

C29 no crashes, but C31... damn I just woke up, on 187 rigs, thousands of crashes

Yes my rig is  crashing too. But in long hours, 9 hours etc.

My memory clock -502 , maybe because of this it is crushing long period. 1080ti hash 1.1 gps

[UnclWish]

Author, remove "AMD" from name of topic. Your miner didn't support AMD cards.

[UselessGuru]

Author, remove "AMD" from name of topic. Your miner didn't support AMD cards.

It does support BEAM on AMD:

https://www.bminer.me/examples/#specifying-devices