RX 560 AMD Driver 18.1.1, topology error on every frame

[GPU timpster]

Hello,

I'm having an issue with my AMD card, it keeps having a topology error, and this has only been recently happening, but it's been a while since I've installed the 18.1.1 GPU drivers. I'd say it's probably been happening for a week or two, but I didn't know about it until at least, a week ago.

Code: Select all

00:16:51:WARNING:FS01:Size of positions 392 does not match topology 389
00:17:11:WU01:FS01:0x21:Completed 62500 out of 6250000 steps (1%)
00:17:51:WARNING:FS02:Size of positions 394 does not match topology 391
00:18:25:WU02:FS02:0x21:Completed 62500 out of 6250000 steps (1%)
00:18:31:WARNING:FS01:Size of positions 392 does not match topology 389
00:19:16:WU01:FS01:0x21:Completed 125000 out of 6250000 steps (2%)
00:20:12:WARNING:FS01:Size of positions 392 does not match topology 389
00:20:27:WARNING:FS02:Size of positions 394 does not match topology 391

This starts IMMEDIATELY after the config, which is as follows:

Code: Select all

00:15:32:Saving configuration to config.xml
00:15:32:<config>
00:15:32:  <!-- Network -->
00:15:32:  <proxy v=':8080'/>
00:15:32:
00:15:32:  <!-- User Information -->
00:15:32:  <passkey v='********************************'/>
00:15:32:  <team v='111065'/>
00:15:32:  <user v='timpster'/>
00:15:32:
00:15:32:  <!-- Folding Slots -->
00:15:32:  <slot id='1' type='GPU'/>
00:15:32:  <slot id='2' type='GPU'/>
00:15:32:</config>
00:15:32:WARNING:WU00:Slot ID 0 no longer exists and there are no other matching slots, dumping
00:15:32:WU00:Sending unit results: id:00 state:SEND error:DUMPED project:8206 run:7 clone:93 gen:103 core:0xa7 unit:0x0000007f868b340258ed39eeb5efcbb5
00:15:32:WU00:Connecting to 134.139.52.2:8080
00:15:33:WU00:Server responded WORK_ACK (400)
00:15:33:WU00:Cleaning up
00:15:49:Saving configuration to config.xml
00:15:49:<config>
00:15:49:  <!-- Network -->
00:15:49:  <proxy v=':8080'/>
00:15:49:
00:15:49:  <!-- Slot Control -->
00:15:49:  <power v='full'/>
00:15:49:
00:15:49:  <!-- User Information -->
00:15:49:  <passkey v='********************************'/>
00:15:49:  <team v='111065'/>
00:15:49:  <user v='timpster'/>
00:15:49:
00:15:49:  <!-- Folding Slots -->
00:15:49:  <slot id='1' type='GPU'/>
00:15:49:  <slot id='2' type='GPU'/>
00:15:49:</config>

I've defaulted my GPU settings in MSI afterburner, and overclocking by as much as 140+ MHz on the core clock speed doesn't make the issue any worse, nor better. Also, sorry for dumping that CPU WU, it was set to 5 threads, but I could have set it to four and let it complete, whoops.

Either way, I'd like to get this solved, and I'm open to rebooting (already done several times as GTA V crashed my system after EXITING!) I'll try reinstalling the display drivers is need be, but I last installed them 1/31/2018, so that could be the issue. I may not have clearly mentioned it but I've already re-installed FAH (everything, I'm on windows 10) just tonight, and did so before that as FAH 4.4.4 was causing my GPU to dump every single work unit, but only the RX 560, not my GTX 960.

[foldy]

I also see this logging "WARNING:FS01:Size of positions 392 does not match topology 389" sometimes .

That is just a warning for the FAH Viewer and does not hurt the work unit in any way.

Don't worry! Nothing needs to be done on your side.

[toTOW]

It's not an error, it happens with core21 v0.0.18 ... you can try to force an update of the core, v0.0.20 solves this issue.

[SteveWillis]

How, please.

[bruce]

To force an update to a FAHCore, pause any WUs that are using the particular FAHCore you're trying to update. Navigate to the FAH data directory (See CWD near the top of your log). Open the "cores" subdirectory and navigate down a long chain of subdirectories until you find that FAHCore and delete it. (See that long chain in the log just before a new WU (re-)starts.) Delete it.

When you resume folding, a fresh copy will download. It will be whatever is considered the "latest" version which may or may not be the same version you already had.

It will only reload a FAHCore when it's actually needed by an active WU. Thus, if you pause ALL slots, you can delete the entire "cores" directory and allow them all to re-load as required for current and future WU assignments.

[SteveWillis]

Followed your instructions but no joy. Still get does not match topology lines. Oh well......
But thanks anyway

[bruce]

That is just a warning for the FAH Viewer and does not hurt the work unit in any way.
Don't worry! Nothing needs to be done on your side.

Agreed.

The message has something to do with how FAH separates the protein image displayed by FAHViewer from the solvent which isn't displayed. The science doesn't use that information.

[GPU timpster]

I think I've got a major issue. I did this, and I'm still having the same issue. Both my GPUs are showing horrible performance. It's similar to what happens when a driver crashes, but because I only output video to the AMD card (and only use it for things other than FAH), I don't think the Nvidia driver should be crashing.

I've got a 12+ minute TPF on GTX 960, Project 11432, and I'll detail what happened when I paused and deleted the core folders. When I resumed the WU, it pulled a FERMI core, which is probably not what I need, so I think the FAH software is misreading my hardware now.

For the AMD card, I added the slot (but it did not say "update core" as with the Nvidia card, and pulled an R600[/] which is the Radeon 2000 series!! I have FAH 7.4.16, so could this be a detection issue in the beta software? I'm not running FAH 7.4.4 because I got nothing but "bad work units" one after another on my AMD card.

[bruce]

Yes, you may have an identification issue, but it's not a major issue. Sometimes FAHClient does interchange the GPU descriptions. That may be confusing for you when you read the log, but it won't disrupt the science. Both NVidia and AMD GPUs support OpenCL which sits on top of their respective hardware drivers. The WUs being assigned currently use FAHCore_21 so from the scientific perspective, the assignments will run on either GPU using whatever processing resources are available on the GPU to which the WU is assigned.

[GPU timpster]

No, it's listing the cards correctly. I tested by pausing the Nvidia GPU and seeing the data in afterburner reflect a low gpu load. The issue I'm having now is horrid performance on the RX 560, it shows currently a TPF of 1h 8m, and an ETA of 4.73 days. I have the card clock / memory speeds on default settings, and have the fan on 80% with extra power.

This card used to do very well in FAH within about 45 seconds TPF difference between it and the GTX 960. I have just now installed the 18.2.3 (optional) AMD driver, AND before that, deleted the GPU slot, and deleted the ATI core folder in AppData.

I then rebooted, added the GPU slot (rebooted again, because the card didin't work correctly (neither the Nvidia) so I did that, and the Nvidia card functions as expected, and I'm getting the currently stated performance.

[bruce]

This is probably useful information. I've seen reports of specific WUs that are more/less efficient when comparing two NV GPUs (say a GTX7xx with a GTX10xx) and I asked an expert to look into it but this is the first direct comparison of an RX 560 to a GTX 960 where the relative performance is that dramatically different. I'll pass this one on, too.

It should be noted that TPF is not a reliable measurement of performance, though and I need a bit more information.

Various projects have wide variations in the number of atoms. They also have wide differences in the number of steps per WU. You can't compare two different projects unless you either are running the same project or you somehow compensate for the differences in those two numbers. More steps mean your hardware has to work harder per frame -- but at least it's linear. More atoms means your hardware needs to work harder per frame (but the increase is greater than linear).

Each project is assigned points based on the expected scientific difficulty so that if your hardware has to work longer to complete a WU, you end up with about the same number of points for the time you spent working on that WU.

Thus you'll need to give me data on which projects you are comparing ... rather than the time-per-frame. ... and what PPD are you earning on each project, both when the TPF was similar and when the TFP was radically different.

Somewhere in all of that is a measure of "efficiency" which is really pretty difficult to extract.