Moo WU taking an Hour per on 5870 and 5990

I have Boinc 7.0.28,
AMD 12.8 drivers.
t1045 processor,
8gm memory,
GPU: 560ti, 5990,5870 (5870 running from the 6" riser I bought, Thanks F$)
I have the app_info.xml so I can run moo @ 1 wu per gpu AMD. The Cuda card runs 4 wu's of Poem. I use Prolasso and designated 4 cores to Poem and the other 2 cores for Moo.

I was running 3 wu's under the old Boinc 6.xx in about 35 min. I added my 560ti to this box to split my Cuda cards for Poem. To run Poem I had to update to Boinc 7xx and now that I wanted to run Moo again it takes 2x longer to complete. 58 min. on the 5870 and 68+ mins on each of the 5990's gpus. I've checking with Afterburner and it says it running at 900/600 with 99% utilization. But something is not right. I double checked with CPU-Z it also says I'm running at 900 with 99% use. When I ran it prior to the upgrade my temps where in the high 70's when running Moo. Now they are in the high 60's. Here is the strangest part, one wu's ran for 36 mins and pushed my temps to 78. Meaning it was running that one wu properly. I can figure it out. Anyone got any ideas? It's like th gpu is downclocked or something.

Hey Mike,

I sounds like Moo is CPU starved. Have you tried editing the POEM app_info file <avg_ncpus></avg_ncpus> and <max_ncpus></max_ncpus> ? Try 0.5 and see if that helps.

Quote:

---

Originally Posted by Fire$torm

Hey Mike,

I sounds like Moo is CPU starved. Have you tried editing the POEM app_info file <avg_ncpus></avg_ncpus> and <max_ncpus></max_ncpus> ? Try 0.5 and see if that helps.

---

I switched the above with your suggestions.

I also noticed Moo is set at

<avg_ncpus>0.15</avg_ncpus>
<max_ncpus>1.0</max_ncpus>

Should I up the Ave. to 0.5?

DUH! Sorry, I wasn't being clear. I meant for you to edit POEM's app_info file to give more CPU resources to Moo. Also try not running any CPU only work for a spell to see if Moo times improve.

Thanks, not running any cpu the 6 cores are feeding the 4 gpus. I'm using prolasso and assigning cores. 1 for each Poem wu (4) the last 2 cores to feed the 3 gps in the 5870 and 5970.

My motherboard is the Asus M4A79T Deluxe. Cuda in the first slot, 5870 with the riser in the 2nd slot and the 5970 in the 3rd slot. I suspend Poem and the I can hear the cards ramp up. Temps climb to 78 on the 5870 and 66 on the 5990. If I can't figure this out I'll have to go to another project. Donate is not an option. Idea here was max credits while the Cuda card ran Poem. Hmm, how do I fix this?

I think Moo is still the 2nd best ATI project unless things have changed.

From your Max credit thread (Here)

Quote:

---

ATI/AMD GPU
1. Donate (With the reduced credits Donate is still King.)
2. Moo for 5xxx & above, MW for 4xxx Cards
3. DiRT (AMD cards run cool on this project.)
4. Milkyway Requires Double Precision GPU (Radeon HD 38xx, 47xx, 48xx, 58xx, 59xx, 69xx)
5. Collatz
6. PrimeGrid

---

Edit: Any chance you can update that thread to add POEM? If you are short on free time then I can figure it out. okay?

Quote:

---

Originally Posted by Fire$torm

I think Moo is still the 2nd best ATI project unless things have changed.

From your Max credit thread (Here)


Edit: Any chance you can update that thread to add POEM? If you are short on free time then I can figure it out. okay?

---

Done.

Quote:

---

Originally Posted by Fire$torm

Hey Mike,

I sounds like Moo is CPU starved. Have you tried editing the POEM app_info file <avg_ncpus></avg_ncpus> and <max_ncpus></max_ncpus> ? Try 0.5 and see if that helps.

---

Did not work. Can't understand why it's choking out when I assign 2 cores to Moo? I assign 4 to Poem. I even reduced the priority of Poem to below normal and raised Moo to high. No go. I stop using Prolasso, same results Moo is choked out.

I've also noticed that Poem takes 10 minutes longer to run when Moo is running vs Milk. I'll continue to play with this.

I think I figured this out.
10/11/2012 10:15:57 AM | | NVIDIA GPU 0: GeForce GTX 560 Ti (driver version 306.23, CUDA version 5.0, compute capability 2.1, 1024MB, 8381357MB available, 1382 GFLOPS peak)
10/11/2012 10:15:57 AM | | ATI GPU 0: Cypress (CAL version 1.4.1741, 1024MB, 991MB available, 5760 GFLOPS peak)
10/11/2012 10:15:57 AM | | ATI GPU 1: Cypress (CAL version 1.4.1741, 1024MB, 991MB available, 4928 GFLOPS peak)
10/11/2012 10:15:57 AM | | ATI GPU 2: Cypress (CAL version 1.4.1741, 1024MB, 991MB available, 4928 GFLOPS peak)


It seems that my AMD cards are running as devices 0,1,2. I have the Cuda card in the first slot closest to the cpu. Boinc Afterburner has labled the Cuda card as GPU4. Funny part is that it does not impact my Milk wu completion times at all. Just my Moo.
Hmm, how to fix this? If I can fix this I'll fix my problem.

Quote:

---

Originally Posted by Mike029

Did not work....... Hmm, how to fix this? If I can fix this I'll fix my problem.

---

The numbering thing is just the difference in how programmers tend to designate things in code. So widget #0 (programmers) = widget #1 (regular folk).

The problem is your AMD Hex-core. You have to keep in mind that your OS is taking up CPU cycles managing all the background stuff including CPU overhead. So anything that forces GPU work to wait for access to the CPU is going to slow down GPU production.

One thing that might help would be to OC the CPU.

And just in case, I found a couple of how-to's

Edward's Home: AMD 1045t Overclock Guide (Link)

YouTube: AMD Phenom II X6 Overclocking Tutorial (NCIX Tech Tips #66) (Link)

So having the cuda card as device 0 and the ATI card as Device 0 is not a problem?

Quote:

---

Originally Posted by Mike029

So having the cuda card as device 0 and the ATI card as Device 0 is not a problem?

---

Correct. Device #0 just means first device (of it's type) found in system. Since ATI and nVidia are different types of devices from BOINC point of view, your good.

Quote:

---

Originally Posted by Fire$torm

Correct. Device #0 just means first device (of it's type) found in system. Since ATI and nVidia are different types of devices from BOINC point of view, your good.

---

Ahh, ok. Then it has to be Moo. Milk runs fine on the 3 gpus with Poem taking 4 of 6 cores. I only have another few weeks then I'll move this box and start hitting Donate on the AMD's ;)