I3 4130 (2 cores, HT off, I've checked, it's faster without it), Radeon 7790. AVG GPU load 48-51%, depends on WU.

On my oldest machine 3/4 of the wu's fail with -1073741819 (0xffffffffc0000005) Unknown error number
http://www.gpugrid.net/result.php?resultid=13789647
http://www.gpugrid.net/result.php?resultid=13786661
http://www.gpugrid.net/result.php?resultid=13783126

all fail with - Unhandled Exception Record -
Reason: Access Violation (0xc0000005) at address 0x000000013F483B37 read attempt to address 0x0C4D5EB0

Its a AMD Phenom X4 with an AMD HD6950 GPU. Running win7 x64
Same setup works with HD7770 or on an A10 7700. Might be an issue with older hardware. I will change my setup.

BTW, still getting 915 wu's, not 916.

BTW, still getting 915 wu's, not 916.

Same here.

Try again. The Applications page says that 9.17 was deployed at 13:40 UTC today.

I've finished the first wu with sucess using 9.17 in a dual 7950 system.

So, good progress :))

My Win 8.1 box completed successfully a work unit. It used 95-98% of CPU and 50-63% of 1 GPU. Second GPU was not touched.

I tried using the <exclude_gpu> in cc_config.xml and ran 2 e@h apps on other gpu. CPU usage by mdrun-917 went from 98% to 65% and GPU for GPUgrid wu went from 65% to 43%. When I suspended the E@H project the GPUgrid numbers returned.

I know Vista was not listed on the list of OSes but my Vista box did get work units. Remaining time or completed time is always 100% and there is some activity on CPU, erratic and only a few spikes on GPU occasionally. I will disable it from getting apps unless you want to continue testing Vista 64bit.

Cheers!

---

Vista 64bit 2x 7870 6 core CPU.

---

My Win 8.1 box completed successfully a work unit. It used 95-98% of CPU and 50-63% of 1 GPU. Second GPU was not touched.

Jobs will only ever run on the first GPU. <exclude_gpu> is not respected.

I know Vista was not listed on the list of OSes but my Vista box did get work units. Remaining time or completed time is always 100% and there is some activity on CPU, erratic and only a few spikes on GPU occasionally running about 30% for an extended time, several minutes. I will disable it from getting apps unless you want to continue testing Vista 64bit.

We can't support Vista. If it works good, but if not, better just disable it.

Matt

My Win 8.1 box completed successfully a work unit. It used 95-98% of CPU and 50-63% of 1 GPU. Second GPU was not touched.

Jobs will only ever run on the first GPU. <exclude_gpu> is not respected.

I know Vista was not listed on the list of OSes but my Vista box did get work units. Remaining time or completed time is always 100% and there is some activity on CPU, erratic and only a few spikes on GPU occasionally running about 30% for an extended time, several minutes. I will disable it from getting apps unless you want to continue testing Vista 64bit.

We can't support Vista. If it works good, but if not, better just disable it.

Matt

That's cool. Just trying to be helpful. Not sure what info helps.

Cheers!

---

I have one rig with AMD Hainan running 'POEM++ OpenCL version v2.17 (opencl_ati_101)' and 'OpenCL for AMD GPUs v9.15 (mt)' at the same time.

http://www.gpugrid.net/show_host_detail.php?hostid=196913

---

I have one rig with AMD Hainan running 'POEM++ OpenCL version v2.17 (opencl_ati_101)' and 'OpenCL for AMD GPUs v9.15 (mt)' at the same time.

http://www.gpugrid.net/show_host_detail.php?hostid=196913

Yes, that's because as far as BOINC is concerned our AMD GPU app is a CPU-only app. If you're not happy with the two tasks sharing the GPU, your best option is to opt out of the AMD GPU app.

Matt

Thats ok. I like them both.

---

I've heard people on here say the app isn't well optimized yet, and I wonder if it's simply not scaling with GPU power at the moment.

On my system, it seems to run consistently at around 35 minutes each
My specs are as follows
CPU: Intel i7-4710HQ http://ark.intel.com/products/78930/Intel-Core-i7-4710HQ-Processor-6M-Cache-up-to-3_50-GHz
GPU: AMD Radeon R9 M265X http://www.notebookcheck.net/AMD-Radeon-R9-M265X.114535.0.html
RAM: 16 GB @ 1600 MHz
SSD: Toshiba Q-series (256 GB)

I've got it set for 5 threads, so it can potentially use a little HT if it helps. If it doesn't, for most apps its not much of a performance hit.
I also have the CPU set not to exceed 2.45 GHz, for thermal reasons.

Afterburner says the GPU utilization is about 74.25%
As of right now, I believe I'm CPU limited. If I allow the CPU to use Turbo Boost, it will go up to 2.9 GHz and the GPU utilization will increase to 78%

On my Nvidia system, with the 650M, utilization is usually around 90%, so, 74% isn't too far off.
Although the Nvidia app uses almost no CPU resources, it typically takes 23 hours to return even the smallests WUs on the laptop

I rather like the behavior of the AMD app using the CPU/GPU in unison.

What is the average run time for the AMD beta tasks?
and what kind of utilization and bottlenecks are you guys encountering?

The situation is different between
laptop an desktop pc
well developed app and beta app

Most posts refer to desktop pc's. It's usually no problem to run them @ 95% gpu load 7/24 and that's what people like to see. The beta app uses on my HD7770 not more than 55% regardless how many of the 4 cpu's are involved. This is why I try to run 2 wu's on the gpu. I would prefer to run two gpugrid app at a time but this is inhibited by system settings. So I try to run 2 different apps on the gpu which works fine now since 2 days.
The goals are different for Matt and the volunteers. Matt tryes to have a stable app and will try to optimize it later. I want my gpu's to run near 100%. You want to keep your notebook alive by limiting the temp.

Alexander

I've heard people on here say the app isn't well optimized yet, and I wonder if it's simply not scaling with GPU power at the moment.

This application is quite different to ACEMD. Whereas ACEMD really only needs a good GPU, this new one needs a good GPU, good CPU *and* good PCIe connectivity.

Inevitably, one of these will be a bottleneck - I'd never expect to see full utilisation of CPU and GPU- a realistic aim is to have both above 75%.

The current WUs are deliberately very short, so we get quick returns and good stats on failure rates.

Production work is probably a month or two away, but we'll ramp up WU sizes between now and then.

Matt

Can't wait till full production. Until then my 5960x and 290x are happily curching away.

I have a machine with two 7970s. The app appears to be using only one. Is this expected? It won't use them both?

Edit:

Also, it uses only 8 threads on a 12 thread machine. I have no restrictions set in any .xml files.

Arg 0 [projects/www.gpugrid.net/mdrun-917.exe]
Arg 1 [--nthreads]
Arg 2 [8]
BOINC running with [8] threads

CPU load confirms only 8 of 12.

---

Reno, NV
Team: SETI.USA

A little down in this thread:

My Win 8.1 box completed successfully a work unit. It used 95-98% of CPU and 50-63% of 1 GPU. Second GPU was not touched.

Jobs will only ever run on the first GPU. <exclude_gpu> is not respected.

So yes only on GPU is used.

---

Greetings from TJ

Working. Gpu usage is little bit over 50%. It using all of my 4 theards.

http://www.gpugrid.net/result.php?resultid=13800417

Matt are you working on the Access Violation problem?

http://www.gpugrid.net/result.php?resultid=13790793
http://www.gpugrid.net/result.php?resultid=13795778
http://www.gpugrid.net/result.php?resultid=13801334
http://www.gpugrid.net/result.php?resultid=13801633