Autosuggest issues and fine tuning about xmr-stak HOT 23 CLOSED

from xmr-stak.

@gboddin can you confirm that the testing was on topic-nonce-final branch?

Also, as psychocrypt said - the logs are vital, we can't deduce a lot from "I'm getting errors".

from xmr-stak.

It was without the topic-nonce-final.

I'm merging

• topic-fix
• topic-nonce-final

on the dev branch, and will let it do it's magic before reporting

from xmr-stak.

Ok first report on 950m :
Letting autoconfig gives :

"gpu_threads_conf" :
[
  // gpu: GeForce GTX 950M architecture: 50
  //      memory: 4000/4044 MiB
  { "index" : 0,
    "threads" : 68, "blocks" : 15,
    "bfactor" : 0, "bsleep" :  0,
    "affine_to_cpu" : false,
  },

],

however, this ends up with :

-------------------------------------------------------------------
xmr-stak 2.0.0-predev mining software.
Based on CPU mining code by wolf9466 (heavily optimized by fireice_uk).
NVIDIA mining code was written by KlausT and psychocrypt.
AMD mining code was written by wolf9466.
Brought to you by fireice_uk and psychocrypt under GPLv3.

Configurable dev donation level is set to 2.0 %

You can use following keys to display reports:
'h' - hashrate
'r' - results
'c' - connection
-------------------------------------------------------------------
[2017-10-18 18:53:38] : Starting GPU thread, no affinity.
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
[2017-10-18 18:53:39] : WARNING: No AMD OpenCL platform found. Possible driver issues or wrong vendor driver.
[2017-10-18 18:53:39] : WARNING: backend AMD disabled.
[2017-10-18 18:53:39] : hwloc: memory pinned
[2017-10-18 18:53:39] : Starting single thread, affinity: 0.
[2017-10-18 18:53:39] : hwloc: memory pinned
[2017-10-18 18:53:39] : Starting single thread, affinity: 1.
[2017-10-18 18:53:39] : hwloc: memory pinned
[2017-10-18 18:53:39] : Starting single thread, affinity: 2.
[2017-10-18 18:53:39] : Connecting to pool xmr-eu1.nanopool.org:14444 ...
[2017-10-18 18:53:42] : Connected. Logging in...
[2017-10-18 18:53:42] : Difficulty changed. Now: 120001.
[2017-10-18 18:53:42] : New block detected.
HASHRATE REPORT
| ID | 10s |  60s |  15m | ID | 10s |  60s |  15m |
|  0 | (na) | (na) | (na) |  1 | (na) | (na) | (na) |
|  2 | (na) | (na) | (na) |  3 | (na) | (na) | (na) |
-----------------------------------------------------
Totals:   (na) (na) (na) H/s
Highest:  0.0 H/s
[CUDA] Error gpu 0: </xmr-stak/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu>:230
terminate called after throwing an instance of 'std::runtime_error'
  what():  [CUDA] Error: the launch timed out and was terminated

Lowering threads to 34 fixes the issue, btw, this is a laptop so : sudo optirun ./bin/xmr-stak was used to make sure the card was enabled and nothing else was blocking.

with 34 threads :

-------------------------------------------------------------------
xmr-stak 2.0.0-predev mining software.
Based on CPU mining code by wolf9466 (heavily optimized by fireice_uk).
NVIDIA mining code was written by KlausT and psychocrypt.
AMD mining code was written by wolf9466.
Brought to you by fireice_uk and psychocrypt under GPLv3.

Configurable dev donation level is set to 2.0 %

You can use following keys to display reports:
'h' - hashrate
'r' - results
'c' - connection
-------------------------------------------------------------------
[2017-10-18 18:58:35] : Starting GPU thread, no affinity.
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
X server found. dri2 connection failed! 
[2017-10-18 18:58:36] : WARNING: No AMD OpenCL platform found. Possible driver issues or wrong vendor driver.
[2017-10-18 18:58:36] : WARNING: backend AMD disabled.
[2017-10-18 18:58:36] : hwloc: memory pinned
[2017-10-18 18:58:36] : Starting single thread, affinity: 0.
[2017-10-18 18:58:36] : hwloc: memory pinned
[2017-10-18 18:58:36] : Starting single thread, affinity: 1.
[2017-10-18 18:58:36] : hwloc: memory pinned
[2017-10-18 18:58:36] : Starting single thread, affinity: 2.
[2017-10-18 18:58:36] : Connecting to pool xmr-eu1.nanopool.org:14444 ...
[2017-10-18 18:58:36] : Connected. Logging in...
[2017-10-18 18:58:36] : Difficulty changed. Now: 120001.
[2017-10-18 18:58:36] : New block detected.
HASHRATE REPORT
| ID | 10s |  60s |  15m | ID | 10s |  60s |  15m |
|  0 | (na) | (na) | (na) |  1 | (na) | (na) | (na) |
|  2 | (na) | (na) | (na) |  3 | (na) | (na) | (na) |
-----------------------------------------------------
Totals:   (na) (na) (na) H/s
Highest:  0.0 H/s
HASHRATE REPORT
| ID | 10s |  60s |  15m | ID | 10s |  60s |  15m |
|  0 | 140.0 | (na) | (na) |  1 | 45.5 | (na) | (na) |
|  2 | 46.6 | (na) | (na) |  3 | 46.5 | (na) | (na) |
-----------------------------------------------------
Totals:   278.5 (na) (na) H/s
Highest:  270.2 H/s
HASHRATE REPORT
| ID | 10s |  60s |  15m | ID | 10s |  60s |  15m |
|  0 | 139.8 | (na) | (na) |  1 | 48.4 | (na) | (na) |
|  2 | 49.4 | (na) | (na) |  3 | 49.3 | (na) | (na) |
-----------------------------------------------------
Totals:   287.0 (na) (na) H/s
Highest:  285.5 H/s

from xmr-stak.

from xmr-stak.

Increasing bfactor to 8 seems to make the miner not crash anymore, but the H/s report for the 950M is on N/A and doesn't seem to move. ( I get 50/s in the 60s column but nothing in the 15s anymore ).

Anyhow, this is a hybrid laptop graphic setup (the nvidia chip is used only for mining, I'm using intel for display), so I expected rough edges and at this point, only halving the threads to 34 works atm ( i'm on 384.90 ).

Afaik optirun probes the drivers and launches an X server. Only then we can run nvidia-settings to OC the memory.

I tried to kill everything once xmr-stak runs but the hashrate went down to 80.

I will look at nvidia-persistenced, maybe that can help, also I wouldn't concern too much about this, it's low hashrate and can kill laptop, if you still want some debug output let me know !

On the GTX 1060 side, I have left the 52 threads with the merge of topic-fix and topic-nonce-final branches, and no reject to report anymore. ( It was quickly noticeable before, I completely forgot to write down the messages that started with "NVIDIA:" though -_-' ).

I'll close this in 24h if nothing else to report. Thanks for the support !

( reproducible build : https://gist.github.com/gboddin/ede88ed2beb43cd44e8205b8f4de6ce3 )

from xmr-stak.

from xmr-stak.

My bad, ok so I have 2 systems :

Server computer running Debian sid ( Kernel 4.11 with default amdgpu driver and nvidia 384.90, no X except to bootstrap nvidia memory clock, killed afterwards )

1. AMD RX580 4GB ( bios timings modded, mem clock at 2100Mhz ) -> generated-amd-rx580-4g.txt : ~780H/s

2. nVidia 1060 3GB ( mem clock at + 900mhz ) -> generated-nvidia-1060.txt : ~510H/s

3. CPU i3 7100T (3.4GHZ no OC) -> generated-cpu-i3-7100t.txt : ~30H/s x2

4. Intel HD graphics ( not used for mining, wish I could , beignet is crashing even clinfo, probably Debian sid )

Laptop running Debian sid ( Kernel 4.9 with nvidia 384.90 )

1. Secondary nVidia 950M ( mem clock at +175mhz ) -> generated-nvidia-950m.txt : ~160H/s

2. CPU i7 i7-6700HQ ( 2.6Ghz no OC ) -> generated-cpu-i7-6700HQ.txt : ~60H/s x3

3. Intel HD graphics ( not used for mining but as default display adapter, wish I could , beignet is crashing even clinfo, probably Debian sid )

Small issues

• The AMD RX580 seemed under-evaluated, probably because during the test the amount of memory detected (2815) is not what's really available in real conditions (4000) ? Changing intensity to 1000 instead of 864 gives +~25H/s
• Yesterday, the nvidia 1060 seemed unstable and gave rejects. however after merge of the fix branches and recompile today it's rock solid at ~510h/s -> so forget it problem solved

Blockers

• The nvidia 950M in the laptop won't start mining with the default suggested threads ( 68, see the previous log ). Setting the bfactor to 8 doesn't help, rather it makes the miner output 50H/s sporadically ( every minutes or so ). Resetting the bfactor to 0 and setting the threads to 34 on the other hand gives a solid 160H/s ( which is what I ended up with when using xmr-stak-nvidia anyway )

To be clear, I'm not blocked, just trying to help you improve auto-detect :)

from xmr-stak.

After too much checking -_-'

Disabling the X server on the nVidia chip makes the PowerMizer run the card in the lowest perf possible (450Mhz), and despites nvidia-smi -ac allowing the max frequency to be changed, doesn't prevent powermizer to kick in when X exits.

The gom option of nvidia-smi should enable full-on computing but :

# nvidia-smi --gom=0
GOM mode cannot be changed on GPU 00000000:01:00.0.
Treating as warning and moving on.
All done.

That happens on both the 1060 and the 950M.

If that's the output of your distro, there's no other choice but to keep an headless X server enabled ...

Here's the final sh** I ended up with :

optirun nvidia-smi -pm 1
optirun nvidia-settings -c :8 -a "GPUPowerMizerMode=1" -a "GPUMemoryTransferRateOffset[2]=400"
optirun ./bin/xmr-stak

There are countless of X headless scripts (and complaints on nVidia's forums) around on the Internet to work around this issue.

Probably this doesn't occur on more advanced cards made for computing only ( and therefore supporting nvidia-smi --gom=0 )

So remaining issue still :
The 950M is getting twice the threads it should during xmr-stak autoconfig ( or something else beyond my knowledge is wrong ).

HASHRATE REPORT
| ID | 10s |  60s |  15m | ID | 10s |  60s |  15m |
|  0 | (na) | (na) | (na) |  1 | (na) | (na) | (na) |
|  2 | (na) | (na) | (na) |  3 | (na) | (na) | (na) |
-----------------------------------------------------
Totals:   (na) (na) (na) H/s
Highest:  0.0 H/s
[CUDA] Error gpu 0: </xmr-stak/xmrstak/backend/nvidia/nvcc_code/cuda_extra.cu>:230
terminate called after throwing an instance of 'std::runtime_error'
  what():  [CUDA] Error: the launch timed out and was terminated

Again fine tuning to half of it works perfect.

from xmr-stak.

from xmr-stak.

I was trying to compile a debug version to give more output but :

[ 48%] Building CXX object CMakeFiles/xmr-stak-backend.dir/xmrstak/backend/backendConnector.cpp.o
/xmr-stak/xmrstak/backend/amd/amd_gpu/gpu.cpp:900:23: error: invalid operands to binary expression ('size_t *' (aka 'unsigned long *') and 'size_t' (aka 'unsigned long'))
                        assert(BranchNonces%w_size == 0);
                               ~~~~~~~~~~~~^~~~~~~
/usr/include/assert.h:89:5: note: expanded from macro 'assert'
  ((expr)                                                               \
    ^~~~

Doesn't happen on release.

Maybe I need some dbg packages ?

from xmr-stak.

from xmr-stak.

@gboddin your assert bug is fixed in #60

from xmr-stak.

@gboddin The nvidia timeout can be a result of an illegal memory access which is fixed in #57

from xmr-stak.

To allocate the full memory of your amd card you need to set some environment variables before you start the miner. The limitaion is coming from the opencl standard. The variables must be set before each start.

export GPU_FORCE_64BIT_PTR 0
export GPU_MAX_HEAP_SIZE 98
export GPU_MAX_ALLOC_PERCENT 100
export GPU_SINGLE_ALLOC_PERCENT 100

Never the less the autosuggestion will always only use max. 2GiB memory.
This will be changes for VEGA gpus soon.

from xmr-stak.

Hi variables are set everywhere now :)

Still same results.

Ok I'm done testing with the various builds ( the debug builds works now , perfect ! ).

So yeah the 950M definitely won't start mining with the default suggested settings. I have to reduce either the threads or blocks by half, everything else is top notch 👍 👍 👍

It's a mobile gpu so it could be a bit exotic, let me know if you need any debug, strace or whatever ;)

A small sumary for hybrid laptop user :

• Use your intel as main graphic
• Don't config Xorg for the nvidia ( except if you want overclocking with coolbits through nvidia-settings, but on a mobile, not recommended )
• Install the nvidia drivers probing them will enable compute

# enable the gpu, check the documentation depending on your situation ( switcheroo ,ect ) 
modprobe nvidia
# resets GPU 
nvidia-smi -r
# sets permanent mode to 0, otherwise clocks stay at the lowest :| < see me confused
nvidia-smi -pm 0 

I guess this bug can be closed since it's a corner case.

from xmr-stak.

from xmr-stak.

Yes.

I also noticed that reducing the thread counts on 1060s ( I have 2 now ) gave a few H/s more.

Maybe the suggest just overshoot nvidia by default ? How is this computed ?

Here's a small table with numbers which should be easier to grasp :

from xmr-stak.

I'll leave my expertise here, even if the ticket was closed.

I'm running 8 x nVidia P106-100, so cards that are made for mining, under Linux (ubuntu-16.04).

You still need to run Xorg. No way around it. I tried. Once you stop Xorg, the card puts itself into P8 state and you can't take it out with nvidia-smi.

Also, you can't enable GOM on these, either:

# nvidia-smi --gom=0
GOM mode cannot be changed on GPU 00000000:01:00.0.
Treating as warning and moving on.
GOM mode cannot be changed on GPU 00000000:02:00.0.
Treating as warning and moving on.
GOM mode cannot be changed on GPU 00000000:03:00.0.
Treating as warning and moving on.
GOM mode cannot be changed on GPU 00000000:04:00.0.
Treating as warning and moving on.
GOM mode cannot be changed on GPU 00000000:06:00.0.
Treating as warning and moving on.
GOM mode cannot be changed on GPU 00000000:07:00.0.
Treating as warning and moving on.
GOM mode cannot be changed on GPU 00000000:08:00.0.
Treating as warning and moving on.
GOM mode cannot be changed on GPU 00000000:09:00.0.
Treating as warning and moving on.
All done.

Currently getting around 490H/s per GPU with the default (auto) settings.

from xmr-stak.

@Znuff yes you always need Xorg with nVidia when gom is not available.

Here's a one shot command to config xorg :

nvidia-xconfig -a --allow-empty-initial-configuration --cool-bits 28

This will config all your cards to run headless ( without monitor having to be connected ) and with overclocking and fan control enabled.

You can then create a small script with

1. nvidia-settings commands for overclocking
2. at the end of the script run you miner

#!/bin/bash
/usr/bin/nvidia-settings -c :0 -a "[gpu:0]/GPUFanControlState=1" 
/usr/bin/nvidia-settings -c :0 -a "[gpu:1]/GPUFanControlState=1" 
/usr/bin/nvidia-settings -c :0 -a "GPUTargetFanSpeed=100"
/usr/bin/nvidia-settings -c :0 -a "[gpu:0]/GPUGraphicsClockOffsetAllPerformanceLevels=200"
/usr/bin/nvidia-settings -c :0 -a "[gpu:1]/GPUGraphicsClockOffsetAllPerformanceLevels=100" 
/usr/bin/nvidia-settings -c :0 -a "[gpu:0]/GPUMemoryTransferRateOffset[3]=1600" 
/usr/bin/nvidia-settings -c :0 -a "[gpu:1]/GPUMemoryTransferRateOffset[3]=1200"
cd /opt/xmrstak
./bin/xmrstak

Then run it with :

xinit your-script.sh

Memory OC is the key with XMR ;)

from xmr-stak.

I am very well aware on how to run it, even wrote my own system, that's not a problem.

It would just be nice for other people to run into this message to figure out that you can't run the mining cards without Xorg when they Google for an answer.

I know, it's stupid.

from xmr-stak.

how to Disable CPU mining ? & my PC is Running slow hang after i start the mining ? what to do ?
& how to reduce speed of GPU

from xmr-stak.

Read the help file, maybe?

Pass --nocpu for no CPU Usage. Edit the nvidia.txt file to change the threads/blocks count to lower usage.

from xmr-stak.