New user with AMD 7660D Linux questions

[dervisionar]

Hello Everyone! I'm a new user with some quick questions regarding my folding efforts.

To get started, here's a snapshot of what my folding rig is (an abridged version of FAHControl's System Info):

CPU: AMD A10-5800K APU w/ Radeon HD graphics
CPUs: 4
Memory: 6.7 GiB
Threads: POSIX_THREADS
OS Version: 3.16 (Linux Mint 17.3 Rosa w/ newer kernel)
OS Arch: AMD64
GPUs 1
GPU : ATI:5 Trinity [Radeon HD 7660D]
CUDA: (obviously not...)
Build Version 7.4.4
Video Driver: AMD Radeon Software Crimson Edition 15.12 Proprietary Linux x86 Display Driver

So I added my GPU to my folding slots, no problem. But my question is in regards to PPD, which to my understanding is the mark of how fast the hardware can produce "work".

To get to my questions: I'm generating ~3060 PPD. 1384 of that is coming from the CPU (project 6393). 1677 of that is coming from the GPU (project 10467). I'm running at "Full Folding Power", but do have my CPU work capped @ 1 Thread. The GPU, to my knowledge, is uncapped. The GPU seems very low PPD when I feel it should be more.

Does anyone else have any experience with the A10-5800K APU's GPU under Linux? Is my GPU's PPD up to par? Is there some special configuration I should be doing?

EDIT: I'm also adding some info that the "new user start here" post suggested.

Network Connection: DOCSIS 3.0 Cable Internet, Ethernet Wired directly to router
Overclocked?: GPU (APU portion) - Yes. up +100mhz from stock (900mhz). CPU - No. Technically the CPU is not overclocked, even though the GPU portion is.
Stable?: Yes. Originally was a Windows cheap gaming rig for a while before I swapped it to my linux server. Played Sleeping Dogs and Skyrim with it at those O.C. settings with no problem. No artifacts. No weird crashes. No problems.

Here's also a log from the GPU unit. You can see where I was playing with OpenCL settings...

Code: Select all

*********************** Log Started 2016-01-24T15:14:42Z ***********************
15:19:38:WU01:FS01:Connecting to 171.67.108.45:80
15:19:38:WARNING:WU01:FS01:Failed to get assignment from '171.67.108.45:80': 10001: Server responded: HTTP_INTERNAL_SERVER_ERROR
15:19:38:WU01:FS01:Connecting to 171.67.108.204:80
15:19:39:WU01:FS01:Assigned to work server 140.163.4.233
15:19:39:WU01:FS01:Requesting new work unit for slot 01: READY gpu:0:Trinity [Radeon HD 7660D] from 140.163.4.233
15:19:39:WU01:FS01:Connecting to 140.163.4.233:8080
15:19:39:WU01:FS01:Downloading 4.28MiB
15:19:41:WU01:FS01:Download complete
15:19:41:WU01:FS01:Received Unit: id:01 state:DOWNLOAD error:NO_ERROR project:10467 run:0 clone:76 gen:248 core:0x17 unit:0x000001ad538b3db9538bbbddb582e716
15:19:41:WU01:FS01:Downloading core from http://web.stanford.edu/~pande/Linux/AMD64/ATI/R600/Core_17.fah
15:19:41:WU01:FS01:Connecting to web.stanford.edu:80
15:19:41:WU01:FS01:FahCore 17: Downloading 3.01MiB
15:19:46:WU01:FS01:FahCore 17: Download complete
15:19:46:WU01:FS01:Valid core signature
15:19:46:WU01:FS01:Unpacked 8.16MiB to cores/web.stanford.edu/~pande/Linux/AMD64/ATI/R600/Core_17.fah/FahCore_17
15:19:46:WU01:FS01:Starting
15:19:46:WU01:FS01:Running FahCore: /usr/bin/FAHCoreWrapper /var/lib/fahclient/cores/web.stanford.edu/~pande/Linux/AMD64/ATI/R600/Core_17.fah/FahCore_17 -dir 01 -suffix 01 -version 704 -lifeline 7630 -checkpoint 15 -gpu 0 -gpu-vendor ati
15:19:46:WU01:FS01:Started FahCore on PID 7908
15:19:46:WU01:FS01:Core PID:7912
15:19:46:WU01:FS01:FahCore 0x17 started
15:19:47:WU01:FS01:0x17:*********************** Log Started 2016-01-24T15:19:46Z ***********************
15:19:47:WU01:FS01:0x17:Project: 10467 (Run 0, Clone 76, Gen 248)
15:19:47:WU01:FS01:0x17:Unit: 0x000001ad538b3db9538bbbddb582e716
15:19:47:WU01:FS01:0x17:CPU: 0x00000000000000000000000000000000
15:19:47:WU01:FS01:0x17:Machine: 1
15:19:47:WU01:FS01:0x17:Reading tar file state.xml
15:19:47:WU01:FS01:0x17:Reading tar file system.xml
15:19:47:WU01:FS01:0x17:Reading tar file integrator.xml
15:19:47:WU01:FS01:0x17:Reading tar file core.xml
15:19:47:WU01:FS01:0x17:Digital signatures verified
15:21:21:WU01:FS01:0x17:Completed 0 out of 5000000 steps (0%)
15:50:07:FS01:Paused
15:50:07:FS01:Shutting core down
15:50:07:WU01:FS01:0x17:Caught signal SIGINT(2) on PID 7912
15:50:07:WU01:FS01:0x17:Exiting, please wait. . .
15:50:59:WU01:FS01:0x17:Lost lifeline PID 7908, exiting
15:50:59:WU01:FS01:0x17:ERROR:103: Lost client lifeline
15:50:59:WU01:FS01:0x17:Folding@home Core Shutdown: CLIENT_DIED
15:51:00:WU01:FS01:FahCore returned: INTERRUPTED (102 = 0x66)
15:52:56:FS01:Unpaused
15:52:56:WU01:FS01:Starting
15:52:56:WU01:FS01:Running FahCore: /usr/bin/FAHCoreWrapper /var/lib/fahclient/cores/web.stanford.edu/~pande/Linux/AMD64/ATI/R600/Core_17.fah/FahCore_17 -dir 01 -suffix 01 -version 704 -lifeline 7630 -checkpoint 15 -gpu 1 -gpu-vendor ati
15:52:56:WU01:FS01:Started FahCore on PID 8908
15:52:56:WU01:FS01:Core PID:8912
15:52:56:WU01:FS01:FahCore 0x17 started
15:52:56:WU01:FS01:0x17:*********************** Log Started 2016-01-24T15:52:56Z ***********************
15:52:56:WU01:FS01:0x17:Project: 10467 (Run 0, Clone 76, Gen 248)
15:52:56:WU01:FS01:0x17:Unit: 0x000001ad538b3db9538bbbddb582e716
15:52:56:WU01:FS01:0x17:CPU: 0x00000000000000000000000000000000
15:52:56:WU01:FS01:0x17:Machine: 1
15:52:56:WU01:FS01:0x17:Digital signatures verified
15:54:17:WU01:FS01:0x17:Completed 0 out of 5000000 steps (0%)
15:58:13:FS01:Paused
15:58:13:FS01:Shutting core down
15:58:13:WU01:FS01:0x17:Caught signal SIGINT(2) on PID 8912
15:58:13:WU01:FS01:0x17:Exiting, please wait. . .
15:58:36:WU01:FS01:0x17:Lost lifeline PID 8908, exiting
15:58:36:WU01:FS01:0x17:ERROR:103: Lost client lifeline
15:58:36:WU01:FS01:0x17:Folding@home Core Shutdown: CLIENT_DIED
15:58:38:WU01:FS01:FahCore returned: INTERRUPTED (102 = 0x66)
16:18:57:FS01:Unpaused
16:18:57:WU01:FS01:Starting
16:18:57:WU01:FS01:Running FahCore: /usr/bin/FAHCoreWrapper /var/lib/fahclient/cores/web.stanford.edu/~pande/Linux/AMD64/ATI/R600/Core_17.fah/FahCore_17 -dir 01 -suffix 01 -version 704 -lifeline 7630 -checkpoint 15 -gpu 2 -gpu-vendor ati
16:18:57:WU01:FS01:Started FahCore on PID 9668
16:18:57:WU01:FS01:Core PID:9672
16:18:57:WU01:FS01:FahCore 0x17 started
16:18:58:WU01:FS01:0x17:*********************** Log Started 2016-01-24T16:18:57Z ***********************
16:18:58:WU01:FS01:0x17:Project: 10467 (Run 0, Clone 76, Gen 248)
16:18:58:WU01:FS01:0x17:Unit: 0x000001ad538b3db9538bbbddb582e716
16:18:58:WU01:FS01:0x17:CPU: 0x00000000000000000000000000000000
16:18:58:WU01:FS01:0x17:Machine: 1
16:18:58:WU01:FS01:0x17:Digital signatures verified
16:20:19:WU01:FS01:0x17:Completed 0 out of 5000000 steps (0%)

[dervisionar]

Update: I think I may have answered my own question with this thread:

viewtopic.php?f=81&t=28030

TLDR: Joe-H said if you're using your computer (inputting and monitor on), FAH scales back resources used.

Here's my questions:

Would having the computer put the display to sleep up my APU's GPU output?

Also, does using X11 forwarded apps make a difference in this? Case in point: I set up FAH through command line, am running FAHControl through an X11 forwarded session remotely, and am posting to this forum using an X11 forwarded browser all from my FAH rig which also does torrenting and file sharing. Does that slow down my GPU's PPD?

[toTOW]

Don't bother running the iGPU, it's a very low end form of the VLIW4 architecture (used in HD6xxx series) which is not very good at running FAH, so it won't probably make the deadlines. Check the ETA of the GPU slot after it completes a couple of % ...

Also, there have been reports of GPU core running on CPU instead of GPU with recent Linux drivers : make sure that the GPU part of your is loaded and that the GPU core is not using all CPU cores instead ...

You'll probably get better results by using your 4 CPU cores to their maximum potential.

Also, make sure you have set up a passkey to be eligible to Quick Return Bonus.

[7im]

One report, not reports. Try folding. If you don't like the results, change it back.

[dervisionar]

Don't bother running the iGPU, it's a very low end form of the VLIW4 architecture (used in HD6xxx series) which is not very good at running FAH, so it won't probably make the deadlines. Check the ETA of the GPU slot after it completes a couple of % ...

Yeah... I was kind of worried that VLIW4 wasn't up to snuff. But I did check deadlines for the iGPU workload, and at it's current rate of PPD it should complete it by the imposed deadline by about a day. Assuming it doesn't choke. I'll check it again more when I get home.

Also, there have been reports of GPU core running on CPU instead of GPU with recent Linux drivers : make sure that the GPU part of your is loaded and that the GPU core is not using all CPU cores instead ...

... You know... I noticed something like that in TOP when I was just checking to see. I don't have it in front of me, but FAHcore_17 was running at 287% of my CPU load, but my overall CPU load was around 5-9% including all of the other processes I have running. So... that's... normal? I'm by no means a linux expert. I just run it on that machine for some of the specific functionality it offers.

You'll probably get better results by using your 4 CPU cores to their maximum potential.

You see, that's another issue in my usage scenario: The machine I have doing the folding is actively used by me as an X11 app-forwarding, Plex transcoding, file sharing, torrent box that is also used as a jukebox, media playback machine, and part time Steam Streamer and Arcade Emulator. I was hoping that FAH could be somehow set up like old school Seti@Home and work as a screen saver of sorts. But I wanted to get into initial configuration and optimization before exploring that option. How that relates to CPU usage, I wanted to tone that back because the AMD "Piledriver" architecture is pretty poor in hard calculations. And with the current workload it already has, I don't want to bog it down too much.

Obviously I'm just going to have to deal with barely meeting deadlines until I do my system upgrade in 2017 when AMD Zen APUs drop... or in my fantasy world where nVidia starts making x86_64 APUs... or if Intel starts putting Iris PRO graphics on desktop i3s, which is probably also sheer fantasy.

Also, make sure you have set up a passkey to be eligible to Quick Return Bonus.

Oh yes. I did do that.

[bruce]

If you take a 24hr average of torrent/X11/media/steam/seti etc. (excluding FAH) how many free CPUs do you have?

The seti screensaver option was good if the computer was mostly idle while you or your customers are sleeping. FAH has a silimar option, but it's best used with an inactive display. Mostly it detects when KVM are inactive and starts GPU folding until you use the mouse or keyboad again. Set the idle option in Advanced Control program or move the slider lower.

What a lot of people would do is buy a GPU board -- whatever you think you can afford -- and ignore the iGPU. It will require about 1 CPU to feed it and will do an excellent job of folding unless your transcoding requires long-term commitments of GPU resources.

Lacking that, go with CPU-only folding, as has been suggested above.

[dervisionar]

If you take a 24hr average of torrent/X11/media/steam/seti etc. (excluding FAH) how many free CPUs do you have?

I'll check that as soon as I get home. That's probably a fair way to look at it.

The seti screensaver option was good if the computer was mostly idle while you or your customers are sleeping. FAH has a silimar option, but it's best used with an inactive display. Mostly it detects when KVM are inactive and starts GPU folding until you use the mouse or keyboad again. Set the idle option in Advanced Control program or move the slider lower.

Ahh... that seems pretty elegant. 1 questions about that though:

How does X11 Window forwarding play into the KVM detection? Does that make a difference at all? Because I set all of the folding up remotely through a forwarded X11 session and ran FAHClient that way to get all of the info I shared. So could it be possible the folding results are better now that I haven't been accessing the box remotely?

What a lot of people would do is buy a GPU board -- whatever you think you can afford -- and ignore the iGPU. It will require about 1 CPU to feed it and will do an excellent job of folding unless your transcoding requires long-term commitments of GPU resources.

Yeah, I have a spare R9 280x sitting around right now. But the only reason I haven't dropped it in revolves around why I bought the APU platform in the first place. It was to partially eliminate the need for a video card for my usage scenario while bringing down my overall power consumption of this iteration of my "home server". Adding the video card defeats the purpose of that for me.

My takeaway from this is that I'm on a virtual "holy grail hunt" in my desire to fold vs. my desire for power efficiency. Seems like I can't have it both ways. At least for this iteration of hardware. Maybe that becomes a different story when new gear comes out.

Lacking that, go with CPU-only folding, as has been suggested above.

Yeah... you're probably right, bruce. Thanks for taking the time to answer my questions, man.

[bruce]

How does X11 Window forwarding play into the KVM detection? Does that make a difference at all? Because I set all of the folding up remotely through a forwarded X11 session and ran FAHClient that way to get all of the info I shared. So could it be possible the folding results are better now that I haven't been accessing the box remotely?{/quote]

I don't have a clue. Let us know when you figure it out.

Yeah, I have a spare R9 280x sitting around right now. But the only reason I haven't dropped it in revolves around why I bought the APU platform in the first place. It was to partially eliminate the need for a video card for my usage scenario while bringing down my overall power consumption of this iteration of my "home server". Adding the video card defeats the purpose of that for me.

If you were working on a "normal" machine, having two GPUs isn't a bad idea. The dGPU can fold without interruptions and the iGPU is probably sufficient to drive your screen.

Nevertheless, the spare R9 280x can be put to use on FAH (... or not) in any machine for an equal power consumption.

[dervisionar]

I don't have a clue (regarding X11 window forwarding affecting folding performance). Let us know when you figure it out.

Well... I'm by no means a scientist and have performance numbers fully fleshed out. But it seems like my total average PPD is still right where it was. All of the work units have progressed per initial estimates with no fluxuations. So it appears not. But again, I'm not the best person to cite.

If you were working on a "normal" machine, having two GPUs isn't a bad idea. The dGPU can fold without interruptions and the iGPU is probably sufficient to drive your screen.

Nevertheless, the spare R9 280x can be put to use on FAH (... or not) in any machine for an equal power consumption.

I have to be honest, bruce, I'm batting around the idea of dropping it in. But the other "gotcha" about the FM2 platform is adding a discrete card that isn't the Radeon HD6670 disables the onboard because it assumes the one you're plugging in is better. I'm about 80% positive of this. I think they have may have addressed that in FM2+ and the Kaveri APUs but it's not the case for the Trinity chips. So I'm not sure I have the option to flex both, even though that would make a lot of sense.

But it all boils down to doing this in the most power efficient way possible for me. I could honestly care less about the raw number of work units I produce as long as I don't totally blast my electric bill doing it. Maybe there's a way to turn this into an income tax credit?

I think all of my questions have been answered here. Again, it appears that the Trinity platform just isn't optimal for folding on a power constrained budget. Obviously 2017 will be my year to do better once AMD Zen APUs featuring Polaris GPUs come out. Then I could at least humor folding decently while keeping total system power consumption under 200w. Hopefully. Assuming AMD can address their power consumption problems.

[Nathan_P]

If you are concerned about power consumption consider a Nvidia GTX 750Ti, they use very little power and are fairly reasonable folding cards.

[bruce]

The income tax credit idea has been explored previously on this forum. General conclusion (1) Check with your tax advisor and (2) Probably not unless you donate money or useful equipment directly to Stanford.

Somebody said they figured out how to enable both the iGPU and the dGPU at the same time. I've never been able to do it.

[dervisionar]

If you are concerned about power consumption consider a Nvidia GTX 750Ti, they use very little power and are fairly reasonable folding cards.

Yeah, I've thought of going that route a while ago with the 750Ti. But my issue with that is knowing that in about 5 months that the 14/16nm video cards are coming out, I can't justify doubling down on a 28nm card with less power efficiency. Even knowing that the 750Ti is actually a decent video card for what it is and is very power efficient as-is. Personally, I'm just more apt to wait it out until the better stuff comes out.

The income tax credit idea has been explored previously on this forum. General conclusion (1) Check with your tax advisor and (2) Probably not unless you donate money or useful equipment directly to Stanford.

Somebody said they figured out how to enable both the iGPU and the dGPU at the same time. I've never been able to do it.

I was half joking about the tax credit thing. But I can't say I'm surprised someone thought of the same thing. Thanks for the update on that.

I know my list of compatible dual cards was super abridged in my FM2 assessment. I could tool around with it. But the other concern with my APU setup is my power supply. It's only a 500w gold. I haven't ran my config through the PSU calculator, but I'm expecting that adding the R9 280X is going to put me over the tipping point. 8 HDDS and 1 SSD is a bit of a power draw on it's own.

Funny epitaph to this: for poops and giggles I installed FAH on my Windows10 Gaming rig w/ an i7-2600k @ 4.25ghz and nVidia 780Ti: ~222160 PPD. 212763 PPD on the 780Ti alone. And here I am trying to milk another 1000ppd out of a beat APU.

Tangential thought: Anyone know how well Intel IRIS PRO graphics stack up in Folding?

[7im]

Flat. Iris GPU is not supported.