I recently added a Sapphire R9 290 video card (AMD Hawaii) paired with an AMD eight core processor to participate in Seti. The setup does process work but with extremely long completion times. I have tried different boinc clients and different video drivers to no avail. I am also using Lunatics optimized Apps. Output doesn’t compare to the nvidia equipped rigs.

For instance the cpu can take up to 35,000 sec runtime and the gpu can take up to 3 to 5,000 seconds. Needless to say the credits are very low. Occasionally the gpu will turn out a work unit in 344 seconds for a much more equitable credit.

I have no Astropulse units in this.

This card is pretty new and info is hard to find. Does anyone here have any experience with this card?

---

I recently added a Sapphire R9 290 video card (AMD Hawaii) paired with an AMD eight core processor to participate in Seti. The setup does process work but with extremely long completion times. I have tried different boinc clients and different video drivers to no avail. I am also using Lunatics optimized Apps. Output doesn’t compare to the nvidia equipped rigs.

For instance the cpu can take up to 35,000 sec runtime and the gpu can take up to 3 to 5,000 seconds. Needless to say the credits are very low. Occasionally the gpu will turn out a work unit in 344 seconds for a much more equitable credit.

I have no Astropulse units in this.

This card is pretty new and info is hard to find. Does anyone here have any experience with this card?

Not that card, specifically.

The first question that comes to mind is "how many CPU work units are you processing?" The second is, "Are you running more than one work unit at a time on the GPU?"

If you use System Information Viewer or CPU-Z, what speed does it show the RAM is running AND importantly, what is the CPU temperature?

Beyond that, anything else you can tell us is going to help us figure this out.

My main and best guess is that you are running too many CPU tasks and starving the GPU for the CPU-attention it needs. But that's just a guess.

EDIT: If it were me I would stop all CPU tasks until I got the GPU doing its best.

I echo tbret's first questions - a quad-module Piledriver CPU should not be taking 8 hours to finish one MB task, closer to a quarter of that time.

For the GPU tasks, it does also appear that the GPU tasks are not getting enough attention from the CPU. While MB CUDA still performs better than MB OpenCL on AMD GPUs, an R9 290 should not take 40 minutes on Win7. It's an unfortunate fact that the current generation of OSs seem to require a CPU core available to maximise GPU throughput.

---

Soli Deo Gloria

I've got a little more time now and will explain my questions:

A lot of people are having trouble cooling the 8350s. The CPU will throttle itself to keep it within operating temperatures, which is why I, personally, prefer to use System Information Viewer from rh-software.com. It shows the CPU's frequency as well as the temperatures (core and socket). If you see your CPU going to 60C, and you see the multiplier back-off and the CPU down-clock, then the temps come down and the CPU ramps back up, then the temps come back up, then the CPU throttles -> You know you have a cooling problem. I've seen this happen on one of mine. Although I could have sworn I had, I had-not properly seated the heatsink on the CPU.

Secondly, most AMD motherboards come with "Cool & Quiet" enabled, so unless you've gone into the BIOS and told it not to use that, you are probably throttling-down every few seconds. Any time the CPU has not-much-to-do the power saving features are slowing the CPU. That won't matter much, but it will matter some. That is not enough to explain your times, however.

Thirdly, as I keep having to explain to people, the 8350 is an 8-core processor, BUT it only has four FPUs. So, if you are doing number crunching that requires the FPU (and it does), then you have one core competing with another core for the FPU's attention and that slows everything down; swapping one bit of work for another while the first bit waits. It's great for "multitasking" but not so great for number-crunching. I just looked and a v7 task on one of mine is taking 4,356.89 seconds, but I only have two cores doing CPU work. The other six are running a dozen MB GPU tasks on four cards.

My R9 270X is running the latest WHQL Catalyst (14.4?) and is taking 1,204.93 seconds to run MB work, three at a time, on an older Phenom II 1090T (in fact, on the same computer as I originally had the heat problem).

If you've goofed-around with a lot of drivers and SDK packages and other things, do this:

Get DriverSweeper, get CCleaner, and UNINSTALL MSI Afterburner. If you've ever had an NVIDIA card in that computer it is VITAL that you COMPLETELY AND ABSOLUTELY uninstall, erase, and otherwise get-rid-of any and every trace of NVIDIA drivers.

Uninstall NVIDIA drivers. Uninstall Catalyst. (you've already uninstalled Afterburner) Run DriverSweeper. Reboot. Run CCleaner. Run CCleaner's Registry cleaner. Reboot. Now, believe it or not, re-run DriverSweeper. I'd lay a wager that something will be found. Clean it and reboot. (you can do the same thing by hand-editing the registry, but why?)

Go onto your hard drive and delete every subdirectory that has anything to do with NVIDIA drivers or AMD video drivers.

NOW install the Catalyst software and don't ever install another Catalyst build until you've thoroughly uninstalled that one (or hand-edited the registry). I prefer to do a "Custom" install and tell it to install in a subdirectory I name, myself. There's no way for a stray registry value to pick-up an old path that way.


NOW you can reinstall Afterburner if you want-to. (for anyone who says "It's not necessary to uninstall Afterburner," I'll reply, "Yeah, you're right. Unless it is necessary (I've seen this exactly twice in years and years, but the last time was recently enough that I remember the experience vividly) then you can spend many hours trying to figure-out why OpenCL won't seem to work.

If you do all of that, and you don't do ANY CPU work, you should see your Hawaii card scream past anything my pitiful little Pitcairn cards will do.

I HATE AMD's driver packages. Hate, hate, hate.

But FIRST do the other stuff. Check the heat. Check the speed. Stop trying to run all 8 cores of the CPU, or 7, or 6. Start with none. There's no use in deciding it's a driver problem until you know the whacky BIOS doesn't have the RAM running at 1/3rd speed (one of my ASRocks, and only one, loves to do that), or the CPU throttled to 400MHz because of heat, or something clearly not driver-related.

If you DO have a driver problem, then go into maximum overkill mode the first time and save yourself having to repeat the uninstall/re-install routine over and over and over again.

Are you using the "-use_sleep" option with Lunatics'? I do and prefer it.

EDIT: Yes, it is entirely possible to have both NVIDIA drivers and Catalyst installed on the same computer.

Haven't seen anything from Pokey in reply yet, but I concur that one shouldn't be trying to run 8 CPU tasks on an FX-8350. Also that NV and AMD drivers can co-exist on a Windows system.

Given that Pokey is running an R9 290, I don't think use_sleep is necessary or helpful.

---

Soli Deo Gloria

Given that Pokey is running an R9 290, I don't think use_sleep is necessary or helpful.

You are correct, sir.

I was on a roll.

I am running 8 cpu tasks and 1 gpu (0.04 CPUs+1 ATI GPU)
GPU Mem clock: 1300-1396
MHz
CPU-Z : Dram Freq.: 798.3
CPU temp according to CoreTemp: 67 C

I saw the 0.04 cpu indication and assumed I was ok with cores. I will cut back to 7 and see what happens.

It is late here, I will get back to you tomorrow.

Thanks

---

Pokey i have the 8350 chip but not the GPU card . I have Nvida GTX650 i do not run all 8 cores as it does heat up quickly . (i do have to get Liquid cooling)
and it does slow down i run 2 units on the GPU and only 6 of the cores and if i do AP i go down to 5 and at 5 cores and 2 MB units on the GPU it is saying it's running at 88%- 92% on the CPU's this is to much to kick in the extra core without it slowing a little or over heating to much

time aprox cpu 2.5-3.5 hours

AP aproxx 12.5 -13.5 hours

MB aprox 2.5 -3.1 hours doing 2 units

V7 Cuda aprox 27-40 mins doing 2 units

---

From the sounds of it, 8 concurrent CPU tasks + CPU throttling is the cause of the low performance.

If you're comfortable editing the app_info.xml, I suggest changing the 0.04 CPU to 1.0. Establish whether or not this helps with the GPU performance first before trying something like running 2 GPU tasks concurrently.

I'm running an FX-8320 at 4.2 GHz with air cooling (Hyper 212 Evo - relatively low-cost yet still reasonably effective) and without CPU throttling, but only with 4 CPU tasks concurrently. From memory, publicly-available Linux builds were processing normal-length MB tasks in about 2.25 hours. After I compiled the CPU application with Piledriver optimisations processing time comes to less than 2 hours. So that should give a rough idea of what to expect when running at peak performance.

---

Soli Deo Gloria

Thanks guys.
I will try these suggestions.

BTW, when I shut down to check some things, I found a process called amdacpusersvc.exe that was using 12-17% of the cpu even though the computer was idle. I stopped it. That isn't "the" problem but it couldn't be helping. Anyone know what it is?

I'm shutting down the cpu usage on this computer.

There was an Nvidia card in prior. And I did use DriverSweeper, but not CCleaner.

Will report back later today on any changes.

One more thing, I did run SIV and if I am reading it right, the multiplier is jumping back and forth 7-20.5 as are the frequencies.
Temps appear to be in the 67-68 C range, but one temp over on far right reads 77 C.
And the Hyper 212 plus is so hot it will burn you to touch it.
I'm going to shut down and reset the heat sink.
The Intel cpu's with Nvidia cards are running much cooler.

---

I`m running my 8350 with an Noctua D14 cooler at 50°C.
2 cores are freed to feed my HD 7970 which is similar to a 280X.
CPU times ~8000 sec. 3 instances on my GPU takes ~1500 seconds.

---

With each crime and every kindness we birth our future.

Interesting to know about the 8350. Explains a lot. I was already moving mine from a dual cpu/gpu cruncher to a pure GPU cruncher after I keep seeing some problems with it. Now I know why it was doing it. Thanks for the info

Zalster

running hot for a reason. The cpu cooler fan has seized up...............

---

Ouch, not good. Was it a stock cooler? I replaced mine with a liquid radiator (2-140 mm fans), under full load CPU temps don't get above 43C. Just make sure the radiator will fit your case. Some have the mounting holes are irregular intervals ( i guess so they only fit their cases).

running hot for a reason. The cpu cooler fan has seized up...............

Yeah. That'll do it.

I second the closed-loop liquid cooling idea. I have two running that way and have no heat issues at all.

Interesting to know about the 8350. Explains a lot. I was already moving mine from a dual cpu/gpu cruncher to a pure GPU cruncher after I keep seeing some problems with it. Now I know why it was doing it. Thanks for the info

Zalster

Don't take this as any sort of "expert" opinion. Long ago I realized that if I wanted to do CPU-crunching I needed to switch-over to Intel processors. It was only the "WOW Challenge" coming-up that made me think "Maybe I should wring these things out for every bit of credit they can stand."

Normally I consider it a waste of heat and electricity.

All three of my 8350 machines are supporting four video cards, each. I do two APs at a time on each card and 3 MBs at a time on each card. The MBs take less CPU and the APs don't take a lot, except when there is a highly blanked work unit.

You can't just set it up to maximize your production under every circumstance because the circumstances change even though you have changed nothing at all (AP blanking).

My "feeling" (intuition: a conclusion reached with inadequate evidence) is that it is *probably* safe for me to run two CPU tasks and crunch on four video cards doing either eight APs or twelve MBs or some mix.

That's just my current "feeling" about this. If I learn anything real that tells me otherwise, I'll change my mind.

One thing I have noticed that I wouldn't have expected is that you don't have to get one of these CPUs to run in the 90%+ range to seriously slow things down. I've started getting nervous about anything over 70%.

But... yeah, as much as I love being able to run all those "threads" and it serves me well when I have multiple things going-on on my computers (music, web surfing, a spreadsheet, and I open a PDF file and...), I have to admit that a Piledriver or Bulldozer just isn't really the right thing for SETI crunching.

I'd sure rather pay for the AMDs up-front, though.

OK, I think we have lift off.
I was able to dig up a cpu cooler in my junk pile to use until my Hyper 212 EVO comes. I am only running two cores (25%) now.
The GPU units have sped up considerably. We will see.
The SIV reports mem and gpu speeds in the correct range, gpu utilization range is 64-100%, temp 70 C, fan at 38%.

Thanks tbret and all. Getting rid of all the drivers was fun...........(sarcasm)

If anybody is still reading this, what is the way to run more than one instance on the gpu? I think I know but.........

---

just edit your app-inf.xml file

look for <cuda>?<cuda> . ?= number of units

ie. for 1 unit you put 1
for 2 units you put 0.50
for 3 you put 0.33

change all the instances of <cuda>?<cuda>

Also remember to edit with word pad not note pad . Little bug it stuff the estimate times up goes down to 5 mins then counts back up . If this does happen just reopen it in word pad and then save it and it fixes the little bug which happen when you use notepad

---

just edit your app-inf.xml file

look for <cuda>?<cuda> . ?= number of units

ie. for 1 unit you put 1
for 2 units you put 0.50
for 3 you put 0.33

change all the instances of <cuda>?<cuda>

Thank you Glenn savill

---

Your welcome pokey

---