I'm going to be letting my Ryzen 2700x machine with a Sapphire Nitro+ Vega 64 do some crunching sometimes... can anyone provide me with a command line to increase the Vega 64's performance like you helped me with my GTX 1060 on my 24/7 rig. I tried googling and came up with nothing but confusing stuff.

---

All of this has happened before and will happen again -Battlestar Galactica

You need to get Mike to help out with the command line for ATI cards. I know nothing about tuning ATI as I've never used one. Maybe you can entice other ATI card users to offer some suggestions.

Or search the forums for keywords ATI tuning command line

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

I never even thought to search "ATI" as I was just searching for "AMD".... LOL!

I'll see what that brings up

---

All of this has happened before and will happen again -Battlestar Galactica

Hmmm.... didn't find anything :(

---

All of this has happened before and will happen again -Battlestar Galactica

I found an old post of mine where Mike said the same command line tuning works well enough for both ATI and Nvidia for 1060 class cards. I thought there are some extra command line tunings for ATI that locks cpu cores to instances. But my quick keyword search didn't turn up that command line.

You can try the standard one for 1060's.

-sbs 1024 -period_iterations_num 1 -tt 1500 -high_perf -high_prec_timer -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

I'll try that because this Vega 64 is showing 12 mins per wu and I know that it can compute faster than that...

---

All of this has happened before and will happen again -Battlestar Galactica

I still think it would be beneficial if you shoot Mike a PM and ask for help with an optimized command line tuning for Vega64.

Mike from Seti. Expert on SoG app tuning. UserID = 9826
User profile https://setiathome.berkeley.edu/show_user.php?userid=9826

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

This is good for a start.

---

With each crime and every kindness we birth our future.

Thanks Mike... I'm running this command line but not seeing any performance difference. Should this Vega 64 be doing work units faster than 12 mins each?


Also I have noticed that HWMonitor shows only 32-38% gpu uasage where as my GTX 1060 in my other rig runs 100%. Is there anyway to have 100% of the Vega 64 compute? I have it setup 1CPU + 1GPU right now with the app_config file so it has one core for the gpu coming off a Ryzen 2700x.

---

All of this has happened before and will happen again -Battlestar Galactica

I'm pretty sure also the Vega64 should be faster than 12 minutes. I think it runs them in 6-7 minutes. I'd have to find another host with a Vega64 card to verify. I think the issue is the command line. I'm pretty darn sure there is a way to do something with assigning affinity of a cpu core to a gpu thread. I think that is the ticket to better speed. I was hoping Mike would step in with the correct command line parameters.
[Edit] Maybe not possible in Linux. I knew there was something about cpu_lock and found the references I was thinking of. It is in the astropulse_7.08_README_x86_64-pc-linux-gnu__opencl_nvidia_100 document in the /Docs folder in the Seti project folder.

On Linux and MacOSX :
Due to OS permission setting rules you have the chance to achieve normal priority by setting <no_priority_change>1</no_priority_change> in <options>
section of your BOINCs "cc_config.xml" file. Check BOINC manuals/wiki for details where to find and how to set this up.

-cpu_lock : Enables CPUlock feature. Results in CPUs number limitation for particular app instance. Also attempt to bind different instances to different CPU cores will be made.
Can be used to increase performance under some specific conditions. Can decrease performance in other cases though. Experimentation required.
Now this option allows GPU app to use only single logical CPU.
Different instances will use different CPUs as long as there is enough of CPU in the system.
To use CPUlock in round-robin mode GPUlock feature will be enabled. Use -instances_per_device N option if few instances per GPU device are needed.
(not available on 64bit linux or MacOSX)

-cpu_lock_fixed_cpu N : Will enable CPUlock too but will bind all app instances to the same N-th CPU (N=0,1,.., number of CPUs-1).
(not available on 64bit linux or MacOSX)

-gpu_lock :Old way GPU lock enabled. Use -instances_per_device N switch to provide number of instances to run.
(not available on 64bit linux or MacOSX)

-instances_per_device N :Sets allowed number of simultaneously executed GPU app instances per GPU device (shared with MultiBeam app instances).
N - integer number of allowed instances.
(not available on 64bit linux or MacOSX)

I was thinking there was a similar way to do cpu affinity with the MB tasks. I use the schedtool app in my Linux hosts to assign affinity to the science applications. I keep the cpu tasks on the physical cores and then let the hyperthread cores support the gpu applications.

sudo apt install schedtool

This doesn't directly speed up the Vega64 card via parameter tuning though.

[Edit2] I think the basic cpu_lock parameter is still valid in Linux64, just the other ones mentioned are not allowed.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

I'm pretty sure also the Vega64 should be faster than 12 minutes. I think it runs them in 6-7 minutes. I'd have to find another host with a Vega64 card to verify. I think the issue is the command line. I'm pretty darn sure there is a way to do something with assigning affinity of a cpu core to a gpu thread. I think that is the ticket to better speed. I was hoping Mike would step in with the correct command line parameters.

Thanks Keith! I'm sure we will get it sorted. This machine wont be doing a lot of crunching anyway since its my main machine but I would like to get it dialed in for when I do let her run BOINC.

---

All of this has happened before and will happen again -Battlestar Galactica

I would try adding the -cpu_lock parameter to your command line for MB and see if it either throws an error or speeds the task up or does nothing.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

It looks as if a CPU core hasn't been Reserved for the GPU. The observed behavior is what would be expected if All CPU cores are busy with CPU tasks.
Try reserving one or More CPU cores for the GPU using the 'Use at most ____ % of the CPUs' setting.
Start with changing the 100% to 99%.

OK, I don't know why I didn't think of this before. You should look at RueiKe's hosts. https://setiathome.berkeley.edu/show_user.php?userid=10276073
He runs exclusive AMD/ATI gpu hardware including older Radeon Fury, Vega64 and Vega VII gpus. If you look at the times for his gpu tasks on his various hosts, he is running MB tasks in 3 -5 minutes.
If you look at his stderr.txt outputs for his tasks you can garner some of the settings he is using in his MB tuning command line parameters. Some of his threads are here.
https://setiathome.berkeley.edu/forum_thread.php?id=81872#1886964
https://setiathome.berkeley.edu/forum_thread.php?id=82949#1936392

I see he has -hp and -high_perf in his command line but I don't see any instance of cpu_lock. So know that card can go faster.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Here is Mike's response to a Vega 56 question in the "GPU FLOPS: Theory vs Reality" thread.

https://setiathome.berkeley.edu/forum_thread.php?id=81962&postid=1963448#1963448

A good start for vega would be.

-sbs 2048 -period_iterations_num 1 -spike_fft_thresh 4096 -high_perf -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64

Tom

---

A proud member of the OFA (Old Farts Association).

and after the -period_iteration_num 1 u can add -tt 600

---

and after the -period_iteration_num 1 u can add -tt 600

I often use -tt 1500 but for whatever reason Mike didn't on that response. Maybe there is a difference between the Nvidia and the Vega on that parameter? Or else, that is what he meant by a "good start" :)

Tom

---

A proud member of the OFA (Old Farts Association).

Here is Mike's response to a Vega 56 question in the "GPU FLOPS: Theory vs Reality" thread.

https://setiathome.berkeley.edu/forum_thread.php?id=81962&postid=1963448#1963448

A good start for vega would be.

-sbs 2048 -period_iterations_num 1 -spike_fft_thresh 4096 -high_perf -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64

Tom

Which txt file does the command line go in?

---

All of this has happened before and will happen again -Battlestar Galactica

Here is Mike's response to a Vega 56 question in the "GPU FLOPS: Theory vs Reality" thread.

https://setiathome.berkeley.edu/forum_thread.php?id=81962&postid=1963448#1963448

A good start for vega would be.

-sbs 2048 -period_iterations_num 1 -spike_fft_thresh 4096 -high_perf -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64

Tom

Which txt file does the command line go in?

Whatever the filename that has mb_cmdline or similar in its name in the Seti project directory.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

OK, I don't know why I didn't think of this before. You should look at RueiKe's hosts. https://setiathome.berkeley.edu/show_user.php?userid=10276073
He runs exclusive AMD/ATI gpu hardware including older Radeon Fury, Vega64 and Vega VII gpus. If you look at the times for his gpu tasks on his various hosts, he is running MB tasks in 3 -5 minutes.
If you look at his stderr.txt outputs for his tasks you can garner some of the settings he is using in his MB tuning command line parameters. Some of his threads are here.
https://setiathome.berkeley.edu/forum_thread.php?id=81872#1886964
https://setiathome.berkeley.edu/forum_thread.php?id=82949#1936392

I see he has -hp and -high_perf in his command line but I don't see any instance of cpu_lock. So know that card can go faster.

Here are my command line options:

 -v 1 -instances_per_device 1 -sbs 2048 -period_iterations_num 1 -tt 600 -spike_fft_thresh 4096 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64 -hp -high_perf -no_defaults_scaling -cpu_lock -nobs -tune 1 64 1 4 -no_use_sleep

In addition, I am using amdgpu-utils to underclock my Vega64s. I am running with a 140W power cap, max Sclk p-state of 6 and a constant Mclk p-state of 3. There are all running in compute mode. All have waterblocks and run at <50C.