Time to test acemd3 for windows. It worked locally. Now I've sent a few WUs named ...TEST31... . There are a few successes but also several failures.

Common errors appear to be

* ERR_RESULT_START couldn't start app: CreateProcess() failed - Access is denied.
* 195 (0xc3) Unknown error number # Engine failed: Error compiling program: nvrtc: error: invalid value for --gpu-architecture (-arch)

The workunits SHOULD be sent to hosts with CUDA 9.2, Kepler and beyond, driver >= 397.44 as per [1]. Each WU uses NVRTC to recompile its kernel at runtime: this (+ driver/card/arch mismatch) may explain the second error.

As for the first error, still no clue.

[1] https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html

Also, I needed to make the assumption that the SystemRoot is c:\windows

Error 195 must be the 20x0s ! We need cuda 10 for that.

Thanks Toni, just in time for a more interesting period in the GPU market (Turing SUPER refresh)!

MrS

---

Scanning for our furry friends since Jan 2002

I just noticed I'm crunching with a new acemd3 2.04 application. Still only beta tasks but all have crunched successfully with the Linux OS and CUDA10.

Yes. acemd3 linux is working for what I can tell.

Windows version is next.

Thanks Toni, just in time for a more interesting period in the GPU market (Turing SUPER refresh)!

MrS

Is this for the newer nVidia 20XX cards? This has been the main reason why I've been holding off water cooling the rest of my GPU's

Yes these new beta wrapper apps correctly work with Turing cards.

Yes. acemd3 linux is working for what I can tell.

I saw this post and thought it was time to come back. My few Win7 computers are getting work but not my Linux rigs. I checked every box in my Preferences.
Too soon for a steady work flow???

---

Yes, Toni only threw out another limited run of beta tasks again. If you didn't grab them right away, you missed them.

I gather we are still a long way before proper generation of non-beta, new acemd3 tasks.

I have a hunch we still have another period of beta work to come for further testing of the Windows application. I don't expect the app to be mainlined until both Linux and Windows apps are validated.

Yes. acemd3 linux is working for what I can tell.

Windows version is next.

I cannot seem to get work for my NVidia Linux system. I just converted it from windows 10 to ubuntu 18.04 as windows could not handle my mix of nvidia boards on risers.

tb85-nvidia

67	GPUGRID	7/13/2019 1:00:27 PM	Sending scheduler request: To fetch work.	
68	GPUGRID	7/13/2019 1:00:27 PM	Requesting new tasks for NVIDIA GPU	
69	GPUGRID	7/13/2019 1:00:29 PM	Scheduler request completed: got 0 new tasks	
70	GPUGRID	7/13/2019 1:00:29 PM	No tasks sent	
71	GPUGRID	7/13/2019 1:00:29 PM	No tasks are available for Short runs (2-3 hours on fastest card)	
72	GPUGRID	7/13/2019 1:00:29 PM	No tasks are available for Long runs (8-12 hours on fastest card)	
73	GPUGRID	7/13/2019 1:00:29 PM	No tasks are available for New version of ACEMD	
74	GPUGRID	7/13/2019 1:00:29 PM	No tasks are available for Anaconda Python 3 Environment	

I am guessing the Linux app is not ready?

I am guessing the Linux app is not ready?

My limited understanding is that it is ready, but they are waiting for the Windows version in order to release them both at the same time.

It is too hot for me anyway. They can wait until September.

It is too hot for me anyway. They can wait until September

I hear you.

Went to open frame mining rig to help with cooling. Windows choked with 5th gpu. I switched to ubuntu with total of 6 gpus. NVidia driver did not spin the fans enough to cool. Spend 2 days figuring out how to enable fan control. Going to make a note here to myself and anyone else:

sudo apt install nvidia-driver-390
// the above puts in the proprietary driver
sudo nvidia-xconfig -a --cool-bits=4
// above created my 6 gpu entries and enabled fan control for all 6
// needs to run every time a board is added or removed.
nvidia-settings &
// above brings up the 6 devices where the fan speed can be set
// hopefully there is a way remember the setting after a reboot

NVidia driver did not spin the fans enough to cool. Spend 2 days figuring out how to enable fan control.

Thanks. I normally don't bother with controlling fans on my Ubuntu machines, but that may be because I didn't know of any way to do it.

sudo apt install nvidia-driver-390
// the above puts in the proprietary driver

I suggest: (suspend GPU tasks first)

sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt-get update
sudo apt-get install nvidia-driver-430

If the last one fails then:

sudo apt-get install libnvidia-compute-430

then the previous
This way you'll have CUDA 10.2 capable drivers.
(If you like the GUI better, then you can use only the first line, then go show apps -> software & updates -> other drives -> select the 430 driver and apply changes, wait for the driver download)

It is too hot for me anyway. They can wait until September

I hear you.

Went to open frame mining rig to help with cooling. Windows choked with 5th gpu.

I'm with both of you there. Hitting over 100F everyday. Shut everything down.

Sounds like you need a bash file to override the nvidia to turn the fans up to 100% all the time.

Keith was kind enough to send me his but you need to make several adjustments to Ubuntu to use them.

---

sudo apt install nvidia-driver-390
// the above puts in the proprietary driver

I suggest: (suspend GPU tasks first)

sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt-get update
sudo apt-get install nvidia-driver-430

If the last one fails then:

sudo apt-get install libnvidia-compute-430

then the previous
This way you'll have CUDA 10.2 capable drivers.
(If you like the GUI better, then you can use only the first line, then go show apps -> software & updates -> other drives -> select the 430 driver and apply changes, wait for the driver download)

I got errors from the NVidia download. My attempt
sudo sh ./NVIDIA-Linux-x86_64-430.34.run
failed within seconds. I then read the instructions that recommended using a repository and NOT using their download. Best I could google was that 390 driver but I also read that it fully support the "10" series of boards. I don't have any newer boards. I will try your repository when I get to a stopping point (seti offline)

I have since discovered the "seti special app" for Linux that can does 6-8 work units in the time it would normally take a gtx1070 to do a single one. I only looked into this app since the Linux app is not working on gpugrid. I will probably crunch on seti with all 6 of my "10" Maybe I can get into the top 10. I posted some performance graphs here https://setiathome.berkeley.edu/forum_thread.php?id=81271 If I can get into the top 3 I may not come back to gpugrid for a while.

It is too hot for me anyway. They can wait until September

I hear you.

Went to open frame mining rig to help with cooling. Windows choked with 5th gpu.

I'm with both of you there. Hitting over 100F everyday. Shut everything down.

Sounds like you need a bash file to override the nvidia to turn the fans up to 100% all the time.

Keith was kind enough to send me his but you need to make several adjustments to Ubuntu to use them.

Just run a bash script file each time you boot the host to set your overclocking and fan control once you have applied the coolbits tweak in xorg.conf.

This is the one I use for my daily driver. It all is accomplished with nvidia-settings and nvidia-smi if you are power limiting.

#!/bin/bash

/usr/bin/nvidia-settings -a "[gpu:0]/GPUPowerMizerMode=1"
/usr/bin/nvidia-settings -a "[gpu:1]/GPUPowerMizerMode=1"
/usr/bin/nvidia-settings -a "[gpu:2]/GPUPowerMizerMode=1"

nvidia-smi -i 0 -pl 215
nvidia-smi -i 1 -pl 215

/usr/bin/nvidia-settings -a "[gpu:0]/GPUFanControlState=1"
/usr/bin/nvidia-settings -a "[fan:0]/GPUTargetFanSpeed=100"
/usr/bin/nvidia-settings -a "[fan:1]/GPUTargetFanSpeed=100"
/usr/bin/nvidia-settings -a "[gpu:1]/GPUFanControlState=1"
/usr/bin/nvidia-settings -a "[fan:2]/GPUTargetFanSpeed=100"
/usr/bin/nvidia-settings -a "[fan:3]/GPUTargetFanSpeed=100"
/usr/bin/nvidia-settings -a "[gpu:2]/GPUFanControlState=1"
/usr/bin/nvidia-settings -a "[fan:4]/GPUTargetFanSpeed=100"

/usr/bin/nvidia-settings -a "[gpu:0]/GPULogoBrightness=20"
/usr/bin/nvidia-settings -a "[gpu:1]/GPULogoBrightness=20"
/usr/bin/nvidia-settings -a "[gpu:2]/GPULogoBrightness=20"

/usr/bin/nvidia-settings -a "[gpu:0]/GPUMemoryTransferRateOffset[4]=600" -a "[gpu:0]/GPUGraphicsClockOffset[4]=60"
/usr/bin/nvidia-settings -a "[gpu:1]/GPUMemoryTransferRateOffset[4]=600" -a "[gpu:1]/GPUGraphicsClockOffset[4]=60"
/usr/bin/nvidia-settings -a "[gpu:2]/GPUMemoryTransferRateOffset[3]=2000" -a "[gpu:2]/GPUGraphicsClockOffset[3]=30"

It only got tricky with the new Turing cards which have TWO fan interfaces since they have two fans on each card. They also have FOUR power levels compared to Pascal's 3 power levels. I had to figure out that you need to increment the fan count to properly identify the fans for control. Also you need to change the [X] number to identify which power level you are applying the overclock to.

This example is for two RTX 2080's and one GTX 1080. Should mention that the GPULogoBrightness command DOES NOT work on the Turing cards. That attribute is not exposed on the Turing cards anymore. Works fine for Maxwell and Pascal though. So for the Turing cards you either have to live with the logo being full on bright or use various levels of opaque tape to cover up the logo.

Guys, you're having a nice discussion here but please don't take this thread completely off-topic - important news could appear here.

MrS

---

Scanning for our furry friends since Jan 2002

Apes,

Maybe move this last few discussions to a new thread, something like "Turing adjustments for heat?"

---