Ordered an R9 290

[Bradtech519]

Should be here next week. Newegg had a killer deal on the MSI R9 290 with $100 headset and 3 games. Along with mail in rebate and promo code. Witcher 3 is coming out and now I can get back into milkyway@home.

 

[petrusbroder]

Nice - and welcom back! 🙂

 

[Neurodog]

There is an issue with 290 and 290x with Boinc. I can't it to work on Milkyway because it is being detected incorrectly

 

[Bradtech519]

There is an issue with 290 and 290x with Boinc. I can't it to work on Milkyway because it is being detected incorrectly

Ouch that is still going on months later? I would have thought the 290/290x issues would have got ironed out by now. Not like the 290/290x are still new cards on the market.

 

[VirtualLarry]

Yeah, my HIS Radeon 7790 1GB GDDR5 card has the same issue with MW@H, cannot use it for that. At least, it works with my NV GT630 1GB DDR3 Kepler 384SP card. Although, the WUs run 30-45 minutes, whereas, on my HD4850 cards, they are like 5 minutes.

 

[Bradtech519]

Got the card in and installed it. Updated my BOINC to 7.4.22 experimental and milkyway@home still barfs at it claiming no opencl support. :'(

 

[Bradtech519]

Looks like Einstein@home is welcoming the R9 290 with open arms. It grabbed four tasks. Perseus Arm survey & Arceibo. Each Arecibo is estimating 1 hour 34 minutes, and the Perseus is around 2hr 33min per BOINC. I haven't added an aftermarket cooler yet. I used AMD overdrive to enable fan manual control. Spun it up to 100% and let the tasks run at 100% load. It is averaging around 60-65c. Think I will let it run for another hour before heading to bed. These are pretty much repackaged 7970s correct? So far the "Twin Frozr" is doing well. I have a 240 GB SSD for OS/Games & 1 TB HDD coming tomorrow. Plan on installing BOINC on the HDD. Poor old WD Black 750 GB in there now is going on 6-7 years.

 

[biodoc]

Check this thread out:

http://boinc.berkeley.edu/dev/forum_thread.php?id=9312

 

[Pokey]

I would offer two observations in addition to Biodoc’s.
The R9 290 needs the latest Catalyst driver which is 14.4 or later. And you need to employ a display driver remover like DDU in order to get all the older stuff cleared out. I have heard the AMD drivers can get conflicted real easy. (my words not theirs)
Also, I have no experience running MW@H but I have run Seti and Folding. For some reason the first catalyst drivers for this card were not where they needed to be. (Imagine that!)

 

[Bradtech519]

Thanks guys for the information. I went home for lunch & got my Crucial 240 GB SSD & 1 TB WD HDD. I started a new install of Windows 8.1 Pro on the SSD. So I should be good to go on the latest of everything this afternoon when I get home to load drivers & get my FX 8350/R9 290 rig back to doing projects.

 

[GLeeM]

and the Perseus is around 2hr 33min per BOINC

Looks like those Perseus WUs took more like about four hours?
That 290 should do better. Did you leave a CPU core idle for the GPU to use?

Were you running two WUs at the same time together?

When running two Perseus WUs my GTX 780 takes about 2 hours 15 minutes on these current WUs. And the 7950 is not far behind.

 

[Bradtech519]

Looks like those Perseus WUs took more like about four hours?
That 290 should do better. Did you leave a CPU core idle for the GPU to use?

Were you running two WUs at the same time together?

When running two Perseus WUs my GTX 780 takes about 2 hours 15 minutes on these current WUs. And the 7950 is not far behind.

I had other workunits going.I could pause LHC tonight and give Einstein another go. Just got done with a Fresh 8.1 install & latest AMD Catalyst drivers. I tried out the 7.4.22 BOINC build to see if Milkyway would work and no go.

 

[GLeeM]

I could pause LHC tonight and give Einstein another go.

On my eight core/thread i7 I free up one core for the GPU by setting "Tools -> Computing preferences...", processor usage tab, "On multiprocessor systems, use at most 87.5% of the processors"

By doing this you can leave LHC going.

 

[Bradtech519]

On my eight core/thread i7 I free up one core for the GPU by setting "Tools -> Computing preferences...", processor usage tab, "On multiprocessor systems, use at most 87.5% of the processors"

By doing this you can leave LHC going.

Ahh I had it at 100% I believe. All I have running tonight is Einstein@home. It has four tasks going right now. Three perseus & one arecibo. Says Running, high Priority (0.5 CPU + 0.25AMD/ATI). I believe I turned off accepting CPU tasks from Einstein & told it to feed me only ATI/Nvidia work units. It appears I am getting mostly perseus arm survey tasks and very very few Arecibo. It's possible through the night I may be doing four GPU tasks all being Perseus at the same time.

 

[GLeeM]

0.25AMD/ATI

OK, four WUs at the same time ... then four hours is good!

The "0.5 CPU" means that with four WUs going BOINC manager should reserve 0.5 times 4 = 2 CPUs. It doesn't always work correctly. With some projects not enough CPU is held back from running CPU WUs and it helps to set the "use at most" to not use all cores.

 

[Bradtech519]

Thanks I've emulated your settings & started LHC@HOME back up around 12 PM CST. I had about 12 hours of nothing but einstein running. Seems like my average credits are still stuck at 0.09. Guess they have a delayed validation process on rewarding credit.

OK, four WUs at the same time ... then four hours is good!

The "0.5 CPU" means that with four WUs going BOINC manager should reserve 0.5 times 4 = 2 CPUs. It doesn't always work correctly. With some projects not enough CPU is held back from running CPU WUs and it helps to set the "use at most" to not use all cores.

 

[GLeeM]

It looks like you have quite a few Invalid WUs. I suggest to change your settings to run only one or at most two WUs at a time to see if that lessens the Invalids.

I use MSI Afterburner or EVGA Precision X or even GPU-Z to watch "GPU usage". This lets me know if the GPU is getting enough CPU. I also use these to determine how many WUs to run at the same time.

 

[Bradtech519]

Yeah, 46 invalids that is pretty rough :thumbsdown:. Pretty much an entire day.. Only had six valids. I'm going back to doing one workunit at a time. Will check back to see how it goes. Will move it up to 2 if my GPU isn't getting worked enough and I'm getting valids.

It looks like you have quite a few Invalid WUs. I suggest to change your settings to run only one or at most two WUs at a time to see if that lessens the Invalids.

I use MSI Afterburner or EVGA Precision X or even GPU-Z to watch "GPU usage". This lets me know if the GPU is getting enough CPU. I also use these to determine how many WUs to run at the same time.

 

[Bradtech519]

Seems as if my activity running one work unit is sporadically around 50-70% utilization of the GPU. The tasks stopped failing reducing it down to 1 WU.

 

[GLeeM]

running one work unit is sporadically around 50-70% utilization of the GPU

Looks like you could run two WUs at a time safely. Running three probably wouldn't increase PPD.
With some projects, when you run multi-WUs you may need to reserve more than one CPU. It may not be worth losing another CPU core!?!

Below shows 1 WU versus 2 WUs on my GTX 560 Ti 448 Core, HD7950 and GTX 780

# of WUs _ GTX560Ti _ 7950 __ GTX780

Einstein x1 __ 6500 ___ 5000 __ 4200 seconds

Einstein x2 __ 10800 __ 7380 __ 6900 seconds

PS: The above is from a while ago, so not with current WUs.

 

[Bradtech519]

Two is the sweet spot it appears & I'm not getting an invalids/errors. I have 6 cores dedicated to LHC & two free for Einstein. Getting ready 90-93% utilization on my CPU. I could prob put another core to work.

 

[GLeeM]

I could prob put another core to work.

Now you are figuring it out!

When you add the other core to CPU work watch your GPU usage to see if it goes down.

Each GPU card is different in how it works with different projects. You just have to try and see what works best.

 

[zzuupp]

One half successful. The first two modified fits crashed immediately. However, the milkyway apps are running, and now validating.
I just turned off modifiedfits in my preferences.

r290 pc

downloads:
http://milkyway.cs.rpi.edu/milkyway/download/

By using arkayn's advice to someone getting an apple/ATI combo working,
and changing filenames and version numbers, here is:

APP_INFO for 32bit windows & ATI:

<app_info>
<app>
	<name>milkyway_separation__modified_fit</name>
	</app>
<file_info>
	<name>milkyway_separation__modified_fit_1.36_windows_intelx86__opencl_ati_101.exe</name>
	<executable/>
</file_info>
<app_version>
<app_name>milkyway_separation__modified_fit</app_name>
<version_num>136</version_num>
<avg_ncpus>0.040000</avg_ncpus>
<max_ncpus>0.567833</max_ncpus>
<plan_class>opencl_ati_101</plan_class>
<file_ref>
<file_name>milkyway_separation__modified_fit_1.36_windows_intelx86__opencl_ati_101.exe</file_name>
<main_program/>
</file_ref>
<coproc>
<type>ATI</type>
<count>1.0000</count>
</coproc>
</app_version>

<app>
<name>milkyway</name>
</app>
<file_info>
<name>milkyway_separation_1.02_windows_intelx86__opencl_amd_ati.exe</name>
<executable/>
</file_info>
<app_version>
<app_name>milkyway</app_name>
<version_num>102</version_num>
<avg_ncpus>0.040000</avg_ncpus>
<max_ncpus>0.567833</max_ncpus>
<plan_class>opencl_ati_101</plan_class>
<file_ref>
<file_name>milkyway_separation_1.02_windows_intelx86__opencl_amd_ati.exe</file_name>
<main_program/>
</file_ref>
<coproc>
<type>ATI</type>
<count>1.0000</count>
</coproc>
</app_version>
</app_info>

Good luck & happy crunching!

 

[GLeeM]

the milkyway apps are running, and now validating.

Are you keeping a CPU core free for the GPU to use? If not, you will be surprised at how much better it will do!

 

[zzuupp]

Are you keeping a CPU core free for the GPU to use? If not, you will be surprised at how much better it will do!

Not at the moment. I've got a full load on the CPU for the primegrid race. I'll free up a core tomorrow night.