Run Multiple WU's on Your GPU

So in a sense you have done exactly what I said above. Contrary to what I said I managed to get to tasks to run however I actually think it is slower because each task is taking around 3 minutes so that 6 minutes in total. However if I was to run one at a time I could complete 2 tasks in approximately just over 4 minutes so that is a speed increase running one at a time from where I am sitting. What are your thoughts? Currently running 0.5 CPU and 0.5 GPU

If you are running 0.5 GPU, then you are running 2 tasks on that GPU. I had created a spreadsheet showing the GPU performance for 1, 2, 3, and 4 tasks running simultaneously. The sweet spot turned out to be 3 simultaneous tasks for the radeon R9 280x. Unfortunately, I deleted that a while ago. But I have another one for n body, which shows the same principle. PM me, and I can send you a PDF version if you like. It would be interesting to see what you get with 1.0 GPU, 0.333 GPU, and 0.25 GPU as well. I also have a 3900x. Basically, each task takes 90 seconds to run, but there are 3 running in parallel, so I knock out one every 30 seconds, on average. It just boggles the mind how much better the GPUs are compared to CPUs.

I have sent you to PM. This project certainly favours ATI cards

MY PC STATS YOUTUBE LINKS // ENJOY
RYZEN 9 5950X ALLCORE 4.5Ghz 128GB DDR4 // SOLO MILKYWAY@HOME WUs BOINC 16C 32T WATT/TEMP STATS https://youtu.be/IRV6louJXZU
RYZEN 9 5950X ASUS LC 6800XT // MILKYWAY@HOME x3 WUs BOINC 16C 32T WATT/TEMP STATS https://youtu.be/5q-CADthO5U

---

im in the country on 2grids they switch alot , so ofcourse my pc crashes, ive lowered my overclock ALL CORE ryzen 9 5950x to
CLOCK RATIO *44.25
CPU VOLT *1.244v
ENABLE S.A.M (SMART ACCESS MEMORY with amd 6000 series gpu) in bios make sure bios version is up to date
turn CSM SUPPORT *DISABLED
turn ABOVE 4G DECODING *ENABLED
turn RE-SIZE BAR SUPPORT *AUTO

in AMD GPU driver under GRAPHICS sub/ ADVANCED (latest driver version)
turn *OPENGL TRIPPLE BUFFERING *ENABLED
turn on S.A.M in AMD driver

---

Yes, your correct.

Based on the other discussions I try to write a script that will work around the limit by pausing the networking

Did this script ever get put into play?

Thanks to Keith M. for the concurrency recommendation on the RTX 2080. Running doubles is working well.

On a related note, I picked up an RTX A2000 for fun this summer to see what the little thing could accomplish. It has 6GB VRAM. Anyone want to throw out a recommendation on running single WUs versus doubles or triples on this thing?

Maybe I'll get motivated & experiment with all configurations, myself, but maybe not. :)

Thanks to Keith M. for the concurrency recommendation on the RTX 2080. Running doubles is working well.

On a related note, I picked up an RTX A2000 for fun this summer to see what the little thing could accomplish. It has 6GB VRAM. Anyone want to throw out a recommendation on running single WUs versus doubles or triples on this thing?

Maybe I'll get motivated & experiment with all configurations, myself, but maybe not. :)

It was posted before:

<app>
<name>milkyway</name>
<max_concurrent>0</max_concurrent>
<gpu_versions>
<gpu_usage>0.25</gpu_usage>
<cpu_usage>0.10</cpu_usage>
</gpu_versions>
</app>

<app>
<name>milkyway_nbody</name>
<max_concurrent>0</max_concurrent>
<gpu_versions>
<gpu_usage>0.25</gpu_usage>
<cpu_usage>0.10</cpu_usage>
</gpu_versions>
</app>

<app>
<name>milkyway_separation</name>
<max_concurrent>0</max_concurrent>
<gpu_versions>
<gpu_usage>0.25</gpu_usage>
<cpu_usage>0.10</cpu_usage>
</gpu_versions>
</app>

Just be aware that using 0.25 as the app is written will run 4 tasks at one time on the gpu, changing it to 0.50 will run 2 tasks at a time and using 0.33 will run 3 tasks at a time

N-Body doesn't have GPU application and "milkyway_separation" has been removed long time ago, so just the "milkyway" part is necessary, the other two should be preferably not included in the app_config file until Milkyway actually has them, sends to you and you know how they run on your hardware with standard settings.

---

.

At the risk of being boring is there any change of Intel GPU's being supported on the options list please. Whilst they are not as fast as the big boys I am sure they could make a useful contribution.

Well, according to this, the DP FLOPS of the new Arc GPUs are quite impressive, those cards should be actually faster than most of the "big boys", in particular all Nvidia consumer GPUs. OTOH in Einstein forum people report no DP capability. So no idea wether those cards can run Milkyway or not.

---

.

Apparently the info on wikipedia is wrong, so Intel GPUs are still completely useless for Milkyway.
https://www.tomshardware.com/news/intel-arc-will-not-support-fp64-hardware

---

.

I have 2x W8100's waiting for a new motherboard (I'm currently using just one with an 5800X+A320 board). Does the syntax change if I wanted to run 2x tasks per GPU? So 4 concurrently across 2 GPU's.

I have 2x W8100's waiting for a new motherboard (I'm currently using just one with an 5800X+A320 board). Does the syntax change if I wanted to run 2x tasks per GPU? So 4 concurrently across 2 GPU's.

No. The app_config entry for 0.5 gpu usage applies to all detected gpus in BOINC. As long as BOINC sees both gpus, then each will run at 2X tasks per card.

You can get even more direct control over each gpu with more elaborate config file writing but that does not seem to be needed in your situation.

---

Thanks Keith. I have a Supermicro EPYC (Gen 1) board on the way. Looking forward to ~4.2 TFlops of FP64 in one machine :)

Thanks Keith. I have a Supermicro EPYC (Gen 1) board on the way. Looking forward to ~4.2 TFlops of FP64 in one machine :)

Nice! Always good to see more Epycs in use in BOINC.

For the earlier generations, you can get a lot of horsepower from cheap, pulled older Epycs from Ebay.

---

I have a Supermicro EPYC (Gen 1) board on the way. Looking forward to ~4.2 TFlops of FP64 in one machine :)

What GPU's are you going to run with your machine when it arrives?

Unless he is running a PCIE Gen 4 or higher gpu, it does not really matter which gpu he uses. The great thing about the Epyc platform is the tons of PCIE lanes and bandwidth it provides to PCIE cards and the 5 or more PCIE slots spaced consistently to allow the gpus to be put in any slot you desire.

The first generation Epyc or Threadripper boards are limited to PCI Gen 3 or lower. You need to move to the 3rd generation boards to pick up Gen 4 speeds. But even at Gen 3 speeds, since the slots you get are at least X8 and normally a full X16 lane width, the speeds don't really matter.

---

i use afterburner for overclocking and gpu z and hwinfo64 for temperature monitoring

Can you take the side off the pc? If so try that and see if it helps or not, if not you can always put a small floor fan blowing into the open side or as you said get more fans and leave the side on. If your current case doesn't have top fans in some cases it's pretty easy to swap to a new case, Dell and HP's being the notable exceptions.

i use afterburner for overclocking and gpu z and hwinfo64 for temperature monitoring

Can you take the side off the pc? If so try that and see if it helps or not, if not you can always put a small floor fan blowing into the open side or as you said get more fans and leave the side on. If your current case doesn't have top fans in some cases it's pretty easy to swap to a new case, Dell and HP's being the notable exceptions.

Small fans are like small dogs. They make a lot of noise but do not do much. Larger fans push more air with less noise but it is possibe to get too big a fan

Duz it fly ?
Did this fall of china`s balloon that `drifted` over USA .