I'm getting v0.03 on my linux. 1 errored, the others finished.
That is CPU. I wish they would separate out the GPU ones better. In fact, that was my original suggestion years ago when I first joined the project. Not much has changed.
The first one of the "0.11 Gravitational Wave Engineering run on LIGO O1 Open Data (GW-opencl-nvidia-V1)" is running nicely on my GTX 750 Ti on a Ryzen 2600 motherboard (Win 10 1809, 388.13 drivers). At least after 20 minutes it has not crashed, which is progress; the total time is estimated at 3.7 hours.
But what struck my attention is that even though the GPU usage is 98%, the power is only 22 watts (and the temperature 51 C) as measured by GPU-Z. That is quite nice for summer use; I will take it. But I wonder how much more work is getting done by the GPU version as compared to the CPU version? It will be interesting to compare the efficiency with other cards also.
EDIT: If the GPU work units are identical to the CPU work units, then you could just compare run times. The 0.03 LIGO CPU work units are running about 8.7 hours on my Ryzen 2600. So the GPU ones are 8.7/3.7 = 2.35 times faster. The real question is then energy efficiency. Also, I am assuming for my estimates that the "Progress %" is accurate for both the GPU and CPU versions. The actual run times may be different.
(Also, my Ryzen 2600 is running on only 10 out of 12 cores; if all cores were in use, then the 0.03 CPU times would probably be a little over 9 hours.)
Also, I am assuming for my estimates that the "Progress %" is accurate for both the GPU and CPU versions. The actual run times may be different.
The first 0.11 GPU work unit that I noted below actually came in at 4 hours 45 minutes, and the next one (at 50% complete) is the same, so they seem to be consistent. (That at the 452.20 Hz frequency.)
On the other hand, my first 0.03 CPU work unit (at the 255.50 Hz frequency) is finishing (edit) in 8 hours.
So a GPU work unit on the GTX 750 Ti is only (edit) 68% faster than a CPU work on a Ryzen 2600, assuming they are doing the same thing. But the GTX 750 Ti draws over 22 watts (since GPU-Z does not measure the total card power), which is considerably more the Ryzen 2600 uses per core. Therefore, I don't see an improvement in efficiency by using a GPU as compared to using a CPU. And there are few GPUs that are much more efficient than the GTX 750 Ti. (My GTX 1060 and 1070 are a little more efficient, but not enough to make up this difference.) I think for power efficiency, the CPU is better.
overnight my Radeon VII machine received 68 0.11 Gravitational Wave Engineering run on LIGO O1 Open Data (GW-opencl-ati-V1) tasks.
I suspended Gamma-Ray Pulsar work, and one of the .11 tasks shows is running on my Radeon VII now.
There are odd things about the Radeon VII status as shown by GPU-Z:
Showing averages as computed by GPU-Z for a few minutes running the 0.11 GW task:
GPU-Load 64%
GPU power draw 39 W (very, very low for this card)
GPU clock 463 MHZ (the graph shows it to be varying, but in a very low range)
Memory clock 798 MHz (also way low)
GPU temperature 36.8C
Hot spot GPU temperature 39.4) (extremely low)
Memory used (dedicated) 1361 MB
Memory used (Dynamic) 166 MB
Meanwhile BoincTasks reports that this task is so far averaging use of 89% of a CPU, whereas a Gamma-ray pulsar task run at 2X on this machine reports about 28%.
At this point the task is posting percentage complete progress of a little over 1% per minute of elapsed time.
While I have been running gamma-ray work at 2X, my settings for GW mean my current GW task is running at 1X.
I'm attempting to raise the "X value" to 3X, but have not gotten the necessary new downloaded task yet.
To restate the obvious, this current Engineering run GW for AMD cards on Windows is a very different beast from the current gamma-ray Pulsar task.
At this point the task is posting percentage complete progress of a little over 1% per minute of elapsed time.
I'd keep an eye on that, on checkpointing, and on the stderr.txt file being written into the slot directory - just to eliminate the possibility that boinc is reporting pseudo-progress for a looper.
I'm still getting v03 on my
)
I'm getting v0.03 on my linux. 1 errored, the others finished.
Zalster wrote:I'm getting
)
That is CPU. I wish they would separate out the GPU ones better. In fact, that was my original suggestion years ago when I first joined the project. Not much has changed.
The first one of the "0.11
)
The first one of the "0.11 Gravitational Wave Engineering run on LIGO O1 Open Data (GW-opencl-nvidia-V1)" is running nicely on my GTX 750 Ti on a Ryzen 2600 motherboard (Win 10 1809, 388.13 drivers). At least after 20 minutes it has not crashed, which is progress; the total time is estimated at 3.7 hours.
But what struck my attention is that even though the GPU usage is 98%, the power is only 22 watts (and the temperature 51 C) as measured by GPU-Z. That is quite nice for summer use; I will take it. But I wonder how much more work is getting done by the GPU version as compared to the CPU version? It will be interesting to compare the efficiency with other cards also.
EDIT: If the GPU work units are identical to the CPU work units, then you could just compare run times. The 0.03 LIGO CPU work units are running about 8.7 hours on my Ryzen 2600. So the GPU ones are 8.7/3.7 = 2.35 times faster. The real question is then energy efficiency. Also, I am assuming for my estimates that the "Progress %" is accurate for both the GPU and CPU versions. The actual run times may be different.
(Also, my Ryzen 2600 is running on only 10 out of 12 cores; if all cores were in use, then the 0.03 CPU times would probably be a little over 9 hours.)
The Version 0.11 went without
)
The Version 0.11 went without any issues, I think. And used about 16% of the GPU. FGRP usually uses about 95%. Thats on a GTX1070 with 8GB RAM.
Jim1348 wrote:Also, I am
)
The first 0.11 GPU work unit that I noted below actually came in at 4 hours 45 minutes, and the next one (at 50% complete) is the same, so they seem to be consistent. (That at the 452.20 Hz frequency.)
On the other hand, my first 0.03 CPU work unit (at the 255.50 Hz frequency) is finishing (edit) in 8 hours.
So a GPU work unit on the GTX 750 Ti is only (edit) 68% faster than a CPU work on a Ryzen 2600, assuming they are doing the same thing. But the GTX 750 Ti draws over 22 watts (since GPU-Z does not measure the total card power), which is considerably more the Ryzen 2600 uses per core. Therefore, I don't see an improvement in efficiency by using a GPU as compared to using a CPU. And there are few GPUs that are much more efficient than the GTX 750 Ti. (My GTX 1060 and 1070 are a little more efficient, but not enough to make up this difference.) I think for power efficiency, the CPU is better.
tolafoph schrieb:The Version
)
The 0.11 ATI version doesn't yet. Running yes, but progress extremely slow. GPU load close to zero, about 3 % intermittently only. Had to abort it.
tolafoph wrote:The Version
)
Ouch. That's ludicrously CPU limited, I'm surprised they went public with these at all with so little of the app ported over.
overnight my Radeon VII
)
overnight my Radeon VII machine received 68 0.11 Gravitational Wave Engineering run on LIGO O1 Open Data (GW-opencl-ati-V1) tasks.
I suspended Gamma-Ray Pulsar work, and one of the .11 tasks shows is running on my Radeon VII now.
There are odd things about the Radeon VII status as shown by GPU-Z:
Showing averages as computed by GPU-Z for a few minutes running the 0.11 GW task:
GPU-Load 64%
GPU power draw 39 W (very, very low for this card)
GPU clock 463 MHZ (the graph shows it to be varying, but in a very low range)
Memory clock 798 MHz (also way low)
GPU temperature 36.8C
Hot spot GPU temperature 39.4) (extremely low)
Memory used (dedicated) 1361 MB
Memory used (Dynamic) 166 MB
Meanwhile BoincTasks reports that this task is so far averaging use of 89% of a CPU, whereas a Gamma-ray pulsar task run at 2X on this machine reports about 28%.
At this point the task is posting percentage complete progress of a little over 1% per minute of elapsed time.
While I have been running gamma-ray work at 2X, my settings for GW mean my current GW task is running at 1X.
I'm attempting to raise the "X value" to 3X, but have not gotten the necessary new downloaded task yet.
To restate the obvious, this current Engineering run GW for AMD cards on Windows is a very different beast from the current gamma-ray Pulsar task.
archae86 wrote:At this point
)
I'd keep an eye on that, on checkpointing, and on the stderr.txt file being written into the slot directory - just to eliminate the possibility that boinc is reporting pseudo-progress for a looper.
I'm not going to run these
)
I'm not going to run these for now because they massively underutilize my GPUs. Maybe there is an app_config that would help?