How long should it take?
13 Jul 2023 19:53:04 UTC
Topic 229796
(moderation:
Howdy all,
Just wondering how long it should take to run a "Gamma_ray pulsar search #5 1.08 (FGRPSSE)" task on an AMD processor running at 4+Ghz?
Mine are running over a day and a half which seems ridiculously long to me. WU says it should take about 11 hours at first start, but then it runs well over a day before it's done.
Thanks,
Allen
Allen wrote:.. how long it
)
Provided you're talking about a relatively modern CPU architecture (eg Ryzen 7) and for a single task not overly having to compete for CPU resources, probably somewhere in the range around 10-16 hours. I haven't run these tasks in quite a while but I don't think there's been any drastic change over the years. Have you done the experiment of running just a single task (no other competition for resources) to get an idea of the best possible crunch rate?
If you're using all available threads, it will slow things down quite a lot since 2 compute heavy tasks will be fighting over the same resources from a single CPU core.
Which seems to indicate that you are over committing the available resources. Nobody can give you a precise recipe for the optimum mix of tasks. You have to do the experiments yourself.
I started to look at your hosts list - 4 Intels and 4 AMDs. I wondered if you might be talking about the Ryzen 7 but then found that the FX-8350 was the one with the long crunch times. That's a very old architecture that was quite slow even when it was first released - around 2012. Because its per core performance was quite bad, and two integer units shared an FPU, it's going to give longer crunch times. If you want to see more details, take a look at this Wikipedia article, particularly under the "Performance" sub-heading. AMD was hit with a class action lawsuit for misrepresenting the core count - also mentioned in that article.
Cheers,
Gary.
Thanks Gary! No, actually
)
Thanks Gary! No, actually it's the FX-8300 I was talking about. It has 2 RX560 on it and is still only gettting about 138k per day. It just seems strange since my laptop an i3-8130u does the wu's much faster, but yes, I am only running one wu's on the cpu and one on the gpu, which are of course, the same chip. Pretty impressed with the i-3, since I'm getting around 46k a day. Guess I will cut back on the FX-8300 threads, even though it has 16 gigs of memory.
Thanks again.
Allen
Allen wrote:... it's the
)
The FX-8300 has exactly the same architecture and would just perform a bit worse than the FX-8350 since it has a lower default clock speed and a lower power draw rating. With the two RX560 GPUs, it should be doing a *lot* better than 138K. If your CPU tasks take up to 1.5 days each and if you were running even 6 of them on the 8 threads, they would be contributing less than 3K to that 138K total if they were running continuously.
Whilst I've never used any FX-8xxx CPUs, I was donated a machine with an FX-6300 CPU some time ago. The motherboard had a pair of PCIe x16 slots so I populated it with a pair of RX 460s - slightly lower performance than RX 560s. I don't run any CPU tasks at all, just FGRPB1G at x2 on each GPU. I have a script (called psa) installed on each machine to monitor parameters of interest to me. I can launch this over an ssh connection from my daily driver to grab the current values and display them on my screen. Here is what that script shows (host IDs redacted for privacy reasons) at the moment:-
[gary@fx_6300-04 ~]$ psa
Hostname: fx_6300-04 Host ID: xxxxxx GPU: Baffin Checked: Fri Jul 14 03:12:23 PM AEST 2023
Disk: dos/Legacy Size: 149.05 GiB Root (Tot/Free): 20G/13G Home (Tot/Free): 9.8G/9.1G
Up: 12d 18h 37m 24s BOINC: 7.16.11 OCL: 20.40 Kernel: 5.15.16-pclos1 On_Frac: 0.985408
4328 ? 00:25:13 boinc
20473 ? 00:01:13 hsgamma_FGRPB1G Frac_done: 0.721525
20480 ? 00:01:01 hsgamma_FGRPB1G Frac_done: 0.594273
20496 ? 00:00:29 hsgamma_FGRPB1G Frac_done: 0.264961
20504 ? 00:00:13 hsgamma_FGRPB1G Frac_done: 0.090315
Current credits:- User Tot: 101.58B User RAC: 56.2M Host Tot: 329.1M Host RAC: 566.2K
[gary@fx_6300-04 ~]$
I also have some even older machines with a single RX 560 GPU. One of those has an E6300 Pentium dual core CPU from around 2009. Once again, no CPU tasks - just FGRPB1G at x2. For comparison, here is what the psa script invoked on that machine shows:-
[gary@e6300-02 ~]$ psa
Hostname: e6300-02 Host ID: xxxxxx GPU: Baffin Checked: Fri 14 Jul 2023 15:14:29 AEST
Disk: dos/Legacy Size: 18.65 GiB Root (Tot/Free): 13G/5.9G Home (Tot/Free): 4.5G/3.8G
Up: 4d 6h 39m 23s BOINC: 7.16.11 OCL: 20.40 Kernel: 5.15.53-pclos1 On_Frac: 0.994960
2421 ? 00:06:05 boinc
15799 ? 00:00:57 hsgamma_FGRPB1G Frac_done: 0.726287
15816 ? 00:00:27 hsgamma_FGRPB1G Frac_done: 0.254601
Current credits:- User Tot: 101.58B User RAC: 56.2M Host Tot: 571.2M Host RAC: 301.0K
[gary@e6300-02 ~]$
As you can see, the single Rx 560 is giving a bit more than half the output (current RAC) of the dual RX 460s.
My advice to you is to not attempt to run Einstein CPU tasks at all on your problem machine and allow your dual RX 560s to show you how much better they can perform on their own. They will still be constrained if you're running lots of CPU tasks from other projects. Obviously, that's a choice for you to make.
It's not strange at all. A core from the 2012 FX-8300 could be well and truly 'out-performed' by a core from a 2009 Intel E6300 Pentium dual core. Just imagine how much better a more modern i3-8xxx architecture can perform. AMD only got back into the higher performing CPU game when they finally released Ryzen.
Cheers,
Gary.
Very true about AMD being in
)
Very true about AMD being in a rut for awhile. I have just made some alterations in my configuration on this machine and will wait a short while to see what sorts out. I did a couple of quick tests just to see what I could see. It seems that when I cut my cpus threads down and increased my gpus by one additional task, the cpus really grunted and some even started the backward counting trend I spoke of before. So, I increased my cpus and cut the gpus back that one wu and everything is moving forward, except that one cpu wu is waiting for memory. BTW, I was wrong about this machine, it only has 8 gigs of ram.
Also, there must be a glich in the software, somewhere, as when I stopped Boinc and restarted, those wu's that said over a day, returned to less than a day to finish. "Screwy"
I should say hear that the software that is running the wu's seems VERY partial to nVidia GPU's. I have a couple of very old systems with 1050i's that are kicking some really amazing numbers. One is sporting a AMD X4 845 and the other an Intel Core(TM)2 Quad CPU Q6600 @ 2.40GHz and I am not restricting any of their theads.
Thanks again for all your help!
Allen
Seems very strange to me,
)
Seems very strange to me, seeing the Q6600 with the 1050i leading the pack .....
The FX-8350 has a RX580 and the X4-845 has a 1050i, the fx-8300 has the 2 rx560's , the Ryzen not only has the internal amd gpu, but also a 1050i and the A8-7600 has a 1050i. I don't get it, unless it is the GPU tasks that they arre running.
GHz
ice Pack 1, (06.01.7601.00)
r
Pack 1, (06.01.7601.00)
ice Pack 1, (06.01.7601.00)
5.00)
s 4C+6G
ice Pack 1, (06.01.7601.00)
5.00)
Allen wrote: Seems very
)
Are you using the "internal amd gpu" to crunch with as well? If so you need to stop as it's just the cpu and that could be a part of why that pc isn't doing as well as it should be doing. It's just like the Intel pc's having an Intel gpu in them, it's just the cpu doing double duty.
Well, if the Ryzen will let
)
Well, if the Ryzen will let me shutdown the internal GPU, I might try that since in all my other machines, the nvidia GPUs seem to be doing a good job. I guess I could just restrict the AMD gpu in the settings.
Thanks,
Allen
Allen wrote: Well, if the
)
Yes that's what most people do, you could also exclude it but just saying don't sent me amd gpu tasks is easier
Yes that's what most people
)
Mikey wrote:
Yes that's what most people do, you could also exclude it but just saying don't sent me amd gpu tasks is easier
But wouldn't that shutdown the nVidia gpu tasks too?
Turn off the igpu in the BIOS
)
Turn off the igpu in the BIOS and you won't use it and BOINC can't see it.
Or exclude AMD gpus from BOINC with exclude statements in the cc_config.xml.