Average credits declining - Can anyone explain?
28 Mar 2014 13:12:21 UTC
Topic 197499
(moderation:
About a month ago, I was averaging ~42K credits daily. This production was on an AMD fx-4300 quad core CPU with a NVidia GTX480 GPU, windows7 platform. However, I began to notice that my average credits were declining daily. I run einstein@home alone on a single host, 24/7, and had not changed this schedule. I decided to move Einstein to one of my intel machines to see if the multithreading and on-board graphics would restore the production. I began to run Einstein on an intel i7-4770K cpu with the NVidia GTX 480 and the intel 4600 graphics. This machine also runs a windows7. Everything seemed to be working well. Boinc recognized all my hardware and the scheduler assigned tasks to the NVidia and intel gpu's and all 8 threads of the CPU. But though I continued to run Einstein 24/7, the average credits continued to fall. Its down to ~27K presently. I know that there are many others with similar hardware configurations that have much better production. Does anybody know what I'm missing? I really do want to understand this.
Average credits declining - Can anyone explain?
)
The AMD fx-4300 quad core CPU with a NVidia GTX480 GPU hasn't contacted the project in 11 days, if it doesn't continue crunching, reporting, and getting fresh work then your RAC will go down.
Claggy
Regarding RAC as compared to
)
Regarding RAC as compared to your peers, one issue on both of your PCs is that you have been allowing Gamma-Ray Pulsar #3 work to run on your nvidia GPUs. The current released code requires quite a lot of CPU support, and the GPU is not very busy--so the flops/hour (which approximately drives the credit award rate) is much lower for this work than for Arecibo GRP or Perseus Arm Survey nvidia GPU work. You could check the task lists for peers you are comparing yourself to, and may find that they are either not allowing GPR3 work at all, or are restricting it to their CPUs.
There may well be other issues--this is just one I noticed in a quick glance.
Thank you so much for your
)
Thank you so much for your insightful response. This seems to be the clue I need to restore my production.
RE: Thank you so much for
)
You might also consider leaving one cpu core free for the gpu to use so it doesn't have periods of waiting to be fed. It will slightly lower your cpu production but very well could increase your gpu production and thus overall rac.
Thanks Mikey. I'll give this
)
Thanks Mikey. I'll give this a try as well. And thanks again archae86. I've made your suggested task restriction and am now monitoring GPU usage.
RE: Thanks Mikey. I'll give
)
Let us know what works so we can pass it on to others.
The result of restricting
)
The result of restricting tasks to exclude running the gamma-ray pulsar #3 application showed an immediate improvement both in GPU utilization and in average credits. As for freeing a CPU core for GPU usage, I really can't say. I assumed you meant that I should restrict CPU usage to use at most 7 of the 8 threads. Shortly after making this adjustment, I noticed that only 7 threads were running tasks. This was short-lived however, because in the next couple of hours, I noticed that all 8 threads were running in addition to the 2 threads for the NVidia and Intel GPUs as it always has. So whether this enhanced improvements made through the task restriction, I cannot measure. P. S. -- This notion of restricting tasks to best utilize the machine's hardware configuration has opened a new chapter in project participation for me. I simply must know more about evaluating applications in terms of CPU and GPU utilization, but I will ask about this in another thread. I don't know who I am talking to - I assume you guys are project technicians assigned to the help desk. If so, I want you to know that I appreciated your responses. On the other hand, If I am talking to fellow participants, I am truly in awe!
The people that have posted
)
The people that have posted in this thread are fellow participants (me included), there are a few (or a lot) of us that have been doing this for a long time and you tend to pick things up on the way. =)
I too stand in awe of the
)
I too stand in awe of the knowledge that often flies around here, in depth and so freely given! I also learn heaps of stuff, and welcome any corrections given.
Thanks guys .... you'll receive the normal brown paper bags containing unmarked bills taped under the third bench on the left in the public park near the lake. North side as usual. :-)
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Hello Robert, I run 7
)
Hello Robert,
I run 7 hosts myself but only one with 8 cores running and that is just a laptop so your GeForce has 10X the GPU's and a better processor which speeds things up for your too.
So your Boinc Manager must be running similar to this (running Gravitational Wave S6 on those 8 cores with 28 tasks in progress)along with Binary Radio Pulsar Search (Arecibo)GPU's v1.34 and Binary Radio Pulsar Search (Arecibo, GPU)v1.39 and BRP (Perseus Arm Survey)GPU's all at once.
You may want to go to your Einstein Preferences and change your preferences to look similar to this and to run CPU's at the same time the S6 tasks to "Yes" and you can test the GPU settings for the BRP PAS "GPU utilization factor of BRP apps" to run at .5 to run two GPU's or .33 for 3 or .25 to run 4 GPU tasks at the same time (I only run .25 on my OC'd 660Ti)
But right now you might either wait until all the other tasks are finished (but set your preferences to "No" for those)or abort them but you do still have 41 of those BRP v1.34 in progress.
And set your "Run CPU versions of applications for which GPU versions are available" to blank if you just want to run Gamma Rays as your CPU tasks.
Here is an example
Some do run faster with one CPU free and some don't make that much difference so on mine I run all 8 running CPU's and still run GPU BRP PAS X2
(I run just Einsteins and a T4T on all of mine most of the time so on my 3-core hosts I run the T4T CPU task and VB along with BRP PAS X2 to get the best GPU times and run X3 along with that T4T task on my quads except the one with the 660Ti OC'd in fact all of mine are OC'd here)
http://einstein.phys.uwm.edu/hosts_user.php
Edit: oh and you will find the RAC rises slower than it drops