On this new 'puter I have an AMD RX 580 card that I finally got working for BOINC stuff.

Let me ramble a bit to get to the real question...

In the last week I managed to finagle a app_config.xml file for E@H that allows one of the apps to run two concurrent threads on the GPU. Seems to be working OK although run times are all over the place. I don't have enough experience with this yet to know just how 'OK' this is.

<app>
<name>einsteinbinary_BRP7</name>
<user_friendly_name>Binary Radio Pulsar Search (MeerKAT)</user_friendly_name>

<gpu_versions>
<cpu_usage>1.0</cpu_usage>
<gpu_usage>0.5</gpu_usage>
</gpu_versions>
</app>

Because this box is running Linux Mint (v20.3, they broke BOINC install in v21.1) and the lack of B550 sensor detection in the kernel I'm ONLY able to do GPU temp & fan monitoring in the desktop (PMW fan setup for all else done in BIOS).

I noticed today that my Milkyway@home Separation 1.46 tasks are taking about 2:42 to complete. I'll assume this is OK for an RX 580 card out of the box. What I also noticed is when running this MW app my GPU temp is nearly 10C LOWER than that E@H app running (two threaded).

Finally... the question: Is this any indication that the Separation 1.46 tasks could be dual threaded on this single card? Anybody with any actual experience with this?

Thanx, Skip

Yes, the MW Separation app is rather poorly optimized. It won't use very much of the gpu at all at any time. It begs running at least 2X on the card to get the gpu utilization up closer to 90%. 2X for Nvidia cards is about max becase of how the Nvidia drivers have to virtualize their OpenCL component to run the Separation app.

AMD or ATI cards OTOH, typically can run at least 4X or maybe 5X depending on how much memory the card has. The app doesn't really use all that much memory. I don't have any running currently to tell you exactly how much the app uses. I remember its about 1.6GB I think but don't quote me.

---

Yes, the MW Separation app is rather poorly optimized. It won't use very much of the gpu at all at any time. It begs running at least 2X on the card to get the gpu utilization up closer to 90%. 2X for Nvidia cards is about max becase of how the Nvidia drivers have to virtualize their OpenCL component to run the Separation app.

AMD or ATI cards OTOH, typically can run at least 4X or maybe 5X depending on how much memory the card has. The app doesn't really use all that much memory. I don't have any running currently to tell you exactly how much the app uses. I remember its about 1.6GB I think but don't quote me.

Great! I'll give it a shot... now if I can find the app short name

Skip,

Keith is right. You should be able to run atleast 5 units on your RX 580. I think I am running 8 or 10 on mine and I only have 8 cpu cores and you have 16. Just depends on how long you want to limit your run time to. I just kept on increasing the number of units I was running until it was not that advantageous to run anymore.

Good Luck,

Allen

Great! I'll give it a shot... now if I can find the app short name

This is the one I use here at MilkyWay
<app_config>

<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>

</app_config>

It's setup to run 2 tasks at a time, you can see the 0.5 in the gpu_usage line, to run 3 tasks at a time you would change that to 0.33, to run 4 tasks would be 0.25 etc.

Great! I'll give it a shot... now if I can find the app short name

You can always get an applications short name from the client_state.xml file.

---

Skip,

Keith is right. You should be able to run atleast 5 units on your RX 580. I think I am running 8 or 10 on mine and I only have 8 cpu cores and you have 16. Just depends on how long you want to limit your run time to. I just kept on increasing the number of units I was running until it was not that advantageous to run anymore.

Good Luck,

Allen

Just changed app_config from .5 gpu to .333
Hope this goes well.

Thanx again, Skip

Skip,

Keith is right. You should be able to run atleast 5 units on your RX 580. I think I am running 8 or 10 on mine and I only have 8 cpu cores and you have 16. Just depends on how long you want to limit your run time to. I just kept on increasing the number of units I was running until it was not that advantageous to run anymore.

Good Luck,

Allen

Just changed app_config from .5 gpu to .333
Hope this goes well.

Thanx again, Skip

Hope it's working great for you. I'm using.......v....on mine.

<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.12</gpu_usage>
<cpu_usage>0.12</cpu_usage>
</gpu_versions>
</app>
</app_config>


Allen

If anybody cares... my 2x vs 3x on MW Sep 1.46 RX 580 results:

Ya'll got better RX580s than me. Don't get me wrong, I'm happy with this. It is a $95 used Sapphire non-Nitro. Here's what some cutting, pasting and number manipulation is showing me for 2x vs 3x on this app. Sub 2 second advantage to 3x... I doubt I can get anything out of 4x (as it turned out on an E@H app also).

Hmmm, I don't seem to know how to paste 7 pages of .pdf here so just the bottom totals lines & notes from my .ods file.

The lower number of 2x tasks included was primarily because 2x was established. 2x numbers included echo prior days of running with that setup but going back further would've included running with a default core clock @ 1366Mhz for this card.

Current 3x running with 1393Mhz core clock but throttled to about 1380Mhz @ 153~154w package limit imposed by BIOS to keep it under the 155w limit. I suspect the 13Mhz throttling makes a very minimal difference, A twin box to this one will get a cheapo off brand RX580 with a low (1286MHz) default core clock tonight so we'll see.

		
						57	# of pending	
						78	# of .333s	
						36	# of .500s	
						114	total	

			Average all	209.32			no division applied
			Average 2x	83.99				
			Average 3x	82.34		1.655	3x secs avg advantage*
	
Notes:
Core temp almost rock steady @ 70C with new fan profile.
Slight throttling of the 2% OC Core Freq on 155w package limit on this Sapphire card.
Dropping to 1300MHz (Level 6) drops package to ~123w and 69C.
Pending count as of 4/5/2023 local.
*The 3x advantage seems to be growing very slightly over time.

Skip

---

- da shu @ HeliOS,
"A child's exposure to technology should never be predicated on an ability to afford it."