Radeon RDNA / Navi DC performance and compatibility thread (RX 5700/XT, for now)

[amrnuke]

I wouldn't worry about it. That's why they are running the beta test.

Fair point!

 

[amrnuke]

Running the "Demo" protein, RX5700 utilization around 30-40% most of the time according to Radeon Settings / Adrenalin. Is that right? Seems low.

 

[crashtech]

Seems very low to me as well.

 

[biodoc]

@amrnuke , Do you have a free thread/cpu to feed the GPU? Maybe your GPU is CPU limited.

 

[amrnuke]

@amrnuke , Do you have a free thread/cpu to feed the GPU? Maybe your GPU is CPU limited.

As of now CPU is on R@H and WCG 50/50, using a total 100% CPU time.
My PPD for the 5700 is 15000, I'm not sure where to find any more information on the GPU's processing power.

 

[crashtech]

@amrnuke , You must leave some CPU free to feed the GPU for Folding. Sometimes other resources seem to be affected also, like RAM bandwidth. Easy way to see is to stop CPU crunching temporarily while monitoring GPU utilization. Running Folding alone should give you your best case GPU utilization. Then allow CPU cores to begin crunching incrementally while continuing to monitor GPU usage. When GPU usage falls too much, back off on number of CPU cores used. Sometimes I have to use much fewer CPU cores than expected to maintain maximal GPU usage, depending on the projects and the system.

 

[VirtualLarry]

My PPD for the 5700 is 15000, I'm not sure where to find any more information on the GPU's processing power.

Yeah, that sounds like CPU starvation to me.

My PPD (with a free dedicated thread on my 6C/12T Ryzen) is 515K PPD for my RX 5700.

Edit: Oh yeah, forgot about the passkey, just assumed everyone folding had one. If you don't, GET ONE, and after 10 successfully-completed WUs, you will get massively better PPD.

 

[biodoc]

@amrnuke , besides what @crashtech suggested, make sure you are folding with a passkey. Without a passkey you will not be eligible for bonus points. Once you enter your passkey, you will need to complete 10 work units to get bonus points and that makes a dramatic difference in PPD.

 

[amrnuke]

@amrnuke , You must leave some CPU free to feed the GPU for Folding. Sometimes other resources seem to be affected also, like RAM bandwidth. Easy way to see is to stop CPU crunching temporarily while monitoring GPU utilization. Running Folding alone should give you your best case GPU utilization. Then allow CPU cores to begin crunching incrementally while continuing to monitor GPU usage. When GPU usage falls too much, back off on number of CPU cores used. Sometimes I have to use much fewer CPU cores than expected to maintain maximal GPU usage, depending on the projects and the system.

Yeah, that sounds like CPU starvation to me.

My PPD (with a free dedicated thread on my 6C/12T Ryzen) is 515K PPD for my RX 5700.

Thanks to both of you. Will work on this today and hopefully have some better results!

@amrnuke , besides what @crashtech suggested, make sure you are folding with a passkey. Without a passkey you will not be eligible for bonus points. Once you enter your passkey, you will need to complete 10 work units to get bonus points and that makes a dramatic difference in PPD.

Thank you, I'll make sure I plug in the passkey for sure.

 

[amrnuke]

With passkey and some CPU time it helped out a lot. I didn't see much difference between single thread and two threads, 540-550K PPD with both, so leaving at single thread so I can pump out some more R@H and WCG.

 

[amrnuke]

My RX5700 was doing 500-550K per day, and here in the last couple of days I noticed it's at about 1 million/day. No driver update. Unclear what changed. Card is also using 99% (with single CPU thread) versus 85% before. Interesting results.

Anyone else see a recent bump?

 

[biodoc]

My RX5700 was doing 500-550K per day, and here in the last couple of days I noticed it's at about 1 million/day. No driver update. Unclear what changed. Card is also using 99% (with single CPU thread) versus 85% before. Interesting results.

Anyone else see a recent bump?

A new beta project (core 22; 11789) has been released. Other's have seen a bump in PPD in the beta forum.

 

[VirtualLarry]

After reading that, I fired up F@H, sure enough, my Navi 10 is hitting 930K-945K PPD, for a total of over 1.6Mil PPD on this PC. (Secondary is a GTX 1660 ti, on a PCI-E 3.0 x4 connection). My Navi 10 is on a PCI-E 3.0 x8 connection, perhaps that's why it's not hitting 1Mil PPD by itself? I think that unlike most mining apps, F@H is somewhat dependent on PCI-E bus bandwidth.

 

[[H]Coleslaw]

Has anyone attempted to get their 5700 to run on Moo!? Last I checked older cards that only had CAL worked but I didn't know if the new cards were still capable of CAL or not...

 

[amrnuke]

After reading that, I fired up F@H, sure enough, my Navi 10 is hitting 930K-945K PPD, for a total of over 1.6Mil PPD on this PC. (Secondary is a GTX 1660 ti, on a PCI-E 3.0 x4 connection). My Navi 10 is on a PCI-E 3.0 x8 connection, perhaps that's why it's not hitting 1Mil PPD by itself? I think that unlike most mining apps, F@H is somewhat dependent on PCI-E bus bandwidth.

My RX5700 is now failing to consistently run F@H, cycling between Ready and only briefly Running (for perhaps a second or two).

 

[VirtualLarry]

My RX5700 is now failing to consistently run F@H, cycling between Ready and only briefly Running (for perhaps a second or two).

Did you remember to put "client-type" "beta" in the GPU slot options? Is it running FahCore22?

 

[amrnuke]

Did you remember to put "client-type" "beta" in the GPU slot options?

Yes and double-checked it's still in there.

Is it running FahCore22?

FahCore listed as 0x22.

 

[VirtualLarry]

Hmm. Tried re-booting? Maybe the GPU got into a "funny" hung state or something? My GTX 1660ti gets that way sometimes. It's my secondary card, for mining and F@H, my XFX RX 5700 8GB GDDR6 is the primary.

 

[biodoc]

Has anyone attempted to get their 5700 to run on Moo!? Last I checked older cards that only had CAL worked but I didn't know if the new cards were still capable of CAL or not...

There is an opencl app for ati/amd and i see rigs in the top computers list with RX 570/580, Radeon VII and vega cards but I haven't seen any Navi cards. I searched the moo forum for 'navi' and came up with nothing.

 

[amrnuke]

Hmm. Tried re-booting? Maybe the GPU got into a "funny" hung state or something? My GTX 1660ti gets that way sometimes. It's my secondary card, for mining and F@H, my XFX RX 5700 8GB GDDR6 is the primary.

Rebooted twice. Tried shutting down BOINC R@H/WCG as well, and running F@H solo, to no avail. Weird. I'm gonna check for driver updates.

 

[biodoc]

Maybe there's no core 22 tasks available.

 

[[H]Coleslaw]

There is an opencl app for ati/amd and i see rigs in the top computers list with RX 570/580, Radeon VII and vega cards but I haven't seen any Navi cards. I searched the moo forum for 'navi' and came up with nothing.

Right. That is why I was wondering if anyone had attempted to not run the OpenCL but rather the CAL work units. I honestly don't know if the newer cards are capable of running the obsolete CAL work anymore. I was just curious if that could be a work around.

 

[amrnuke]

Maybe there's no core 22 tasks available.

It oddly was downloading, then trying to start, and not working.
It is running now, but instead of the previous 1M ppd, it's back down to 500-550K ppd.
May consider updating to optional Adrenalin 20.1.1 to see if that maybe helps. But also worried it might break things.

Edit: For no apparent reason, no driver updates or anything, it just popped back up to 1M ppd.

 

[biodoc]

The latest windows drivers for Navi cards works now for Einstein@home. See this thread.

 

[VirtualLarry]

https://www.reddit.com/r/Amd/comments/eqfgz0