Posts by HAL9000


log in
21) Message boards : Number crunching : Just added 3rd GPU and CPU is 'Waiting for Memory' (Message 1656140)
Posted 5 days ago by Profile HAL9000
So previously you were running OK with 4 instances per GPU now?


That's correct, with 2 x GPUs (either the 2 x Titans or 1 x Titan & 1 x 750Ti) all was good, I had 4 CUDA tasks running per GPU (8 GPU CUDA tasks total) and 1 task per CPU core (8 x CPU tasks total).

Now, with 3 x GPUs (1 x Titan & 2 x 750Ti), 4 or 5 of the CPU tasks (it flips a bit) sit there saying 'Waiting for Memory'. I do, however, still have 4 CUDA tasks running per GPU still (12 GPU CUDA tasks total).

Makes sense?

As I said, I'm sure I'm not being efficient with the workload I'm assigning the GPUs/CPUs.

In your BOINC Computing preferences. What do you have for
    When computer is in use, use at most
    When computer is not in use, use at most
    Page/swap file: use at most

I think the default values are like 40 or 50%. Which if I'm doing my maths correctly should be fine for 12 GPU + 8 CPU tasks. However BOINC seems to think otherwise at the moment. So I would try bumping up those values if you have not already done so.
I pretty much maxed out the settings in the BOINC prefs for everything. SO I wouldn't have weird resource issues. At least not ones caused by BOINC deciding it needed to do something about.

22) Message boards : Number crunching : Just added 3rd GPU and CPU is 'Waiting for Memory' (Message 1656123)
Posted 5 days ago by Profile HAL9000
So previously you were running OK with 4 instances per GPU now?
23) Message boards : Number crunching : GPU driver version lists (Message 1655699)
Posted 7 days ago by Profile HAL9000
Besides the ATI driver list I have been keeping. I recently put together one for Nvidia drivers. The NV list needs lots of information filled in at the moment, but if it is useful then I'll populate it as I gain information. Something like the current NV list might be needed for Intel GPU drivers at this point as well.
ATI Driver Version Cheat Sheet
Nvidia Driver Version Cheat Sheet
24) Message boards : Number crunching : @Pre-FERMI nVidia GPU users: Important warning (Message 1655679)
Posted 7 days ago by Profile HAL9000
Or optionally upgrade from Pre Fermi to something that will compute faster burning less trees :).

What, The 8500 GT I am using is to slow for the amount of power is uses?!
I wish I could replace it, but it's a work compy. I keep getting teased that I'll get to replace the machine, but then we keep having budget issues. :/

Also is anyone keeping a list of NV drivers for people to reference like the ATI Driver Version Cheat Sheet I make?
25) Message boards : Number crunching : GPU Problem (Message 1655678)
Posted 7 days ago by Profile HAL9000
One way to observe any instability on the +12V rail, is to use a digital multimeter set to low AC volts. Ideally the AC content on the +12V(DC) to GND would be zero, but under load can bounce significantly if the PSU is either stressed/aged or insufficient. Less than about 200mVAC under full load would be (IIRC) in spec, though lower is better.

Some PSU models intended for high power multiple GPUs come with chokes on the PCI express power leads, while yet more expensive ones seem to have moved to extremely fast regulation circuitry.

Either way (AC there or not), as Keith describes a small GPU core voltage bump could help, provided nothing's actually faulty just on the edge of spec.

The R7 240 they have is rated for 30w. Which I would imagine is only coming from the PCIe slot.
However I did just noticed the R7 240 is designed to run 730-780MHz & it was previously stated the GPU was running at 925MHz. So it may just be a case of pushing a little card harder than it can go.
26) Message boards : Number crunching : Intel GPU (Message 1655267)
Posted 8 days ago by Profile HAL9000
Since we have the big boys in here talking about CPU usage settings :D

I have always wondered...

Is there any difference in:
GPU 0.5
CPU 0.4

or

GPU 0.5
CPU 0.3

Does the GPU task use more or less CPU as indicated, or is it just a limit as to when to shut down a CPU to feed the GPU?

The the full descriptions of what the settings for the app_info.xml & app_config.xml are on the BOINC wiki.

But the simple answer it is just to reserve CPU cores.
So you can set the value for the CPU from
GPU 0.5
CPU 0.01
to
GPU 0.5
CPU 0.49
and nothing will change for a single GPU system.
If you had 2 GPUs and set
GPU 0.5
CPU 0.25
Then 1 CPU core would get reserved for the 4 GPU instances.
27) Message boards : Number crunching : Panic Mode On (96) Server Problems? (Message 1655012)
Posted 8 days ago by Profile HAL9000
When I checked the SSP a few minutes ago, the ready to send for MB was 2!

And now it's over 30 minutes since the SSP last updated.

What did you expect? It's Friday, and it's a law that the system must prepare to go down for the weekend :-)

Yes, but it's still morning in Berkeley. Not even 10am yet.

And the SSP still hasn't updated.

I don't even pretend to do any work at work before 10am on a Friday. Once their coffee starts to kick in they might have some time to give things a look.
28) Message boards : Number crunching : Windows 10 may put SETI@home in Danger (Message 1654752)
Posted 9 days ago by Profile HAL9000
I don't think I have used the screen saver since about October of '99.
29) Message boards : Number crunching : Intel Integrated GPU not working (Message 1654642)
Posted 9 days ago by Profile HAL9000
What chipset does your motherboard use?

Cheers.

Ah yes the P67 chipset does not allow the iGPU to be used. While all of the other 6 series & all 7 series chipsets do.
30) Message boards : Number crunching : Intel GPU (Message 1654596)
Posted 9 days ago by Profile HAL9000
A setting of <max_ncpus>0.02</max_ncpus> in an app_info.xml has nothing to do with the applications CPU usage while processing. It is used by BOINC for scheduling purposes.
A setting of <cpu_usage>0.02</cpu_usage> in an app_config.xml is used to reserve CPU cores. Such as setting <cpu_usage>0.02</cpu_usage> while running 2 GPU instances to reserve 1 CPU core.

That won't have any effect at all. BOINC allows the CPU to be over-committed until the sum of all <cpu_usage> and <max_ncpus> fractions, across all running co-processor tasks, reaches 1.000000: actually, I think <avg_ncpus> is dominant in app_info.xml, but that makes little difference to the principle.

With a value of <cpu_usage>0.02</cpu_usage> in app_config, BOINC will allow 49 GPU tasks to run in parallel, before reserving an extra core - but your hardware would probably have crashed long before that. To reserve the core at two tasks, the <cpu_usage> value would need to be between 0.5 and 0.99

I was doing a copy/paste with <cpu_usage>0.02</cpu_usage> & forgot to change that to <cpu_usage>0.5</cpu_usage> for my example. Examples are much better when actually accurate! Thanks for catching that.
31) Message boards : Number crunching : Intel GPU (Message 1654539)
Posted 9 days ago by Profile HAL9000
I stopped one after 7 + hours and it was 46% complete

I wonder if that has to do with your driver version 10.18.14.4156. In the world of GPU crunching latest rarely is the greatest. As such the last driver version 15.36.14.4080 were listed as DO NOT USE for SETI@home. As it would produce errors. When we were discussing iGPU drivers last. Version 10.18.10.4061 with 10.18.10.3621 as fall back were known to be good. Versions 4156 may be good as well, but falls into unknown/untested.

Importantly the driver version is probably not related to that increase in CPU times. At least in the sense there isn't a driver that will magically fix the issue, at the moment.
32) Message boards : Number crunching : Intel GPU (Message 1654535)
Posted 9 days ago by Profile HAL9000
I'm not overly talented at this but ... I'm sure you can do something like 0.02 CPU usage for your GPU, that way it won't tax your CPU and give a little to the GPU.

Other people will know the command ... I just don't know it without looking.

A setting of <max_ncpus>0.02</max_ncpus> in an app_info.xml has nothing to do with the applications CPU usage while processing. It is used by BOINC for scheduling purposes.
A setting of <cpu_usage>0.02</cpu_usage> in an app_config.xml is used to reserve CPU cores. Such as setting <cpu_usage>0.02</cpu_usage> while running 2 GPU instances to reserve 1 CPU core.

On an i5, or probably all Ivy Bridge/Haswell CPUs, when running SETI@home it doesn't matter if you are running 1 CPU instances, leaving the other 3 cores free, when running the iGPU as well. The CPU times will be increased to nearly double.
It is probably a limitation of the hardware. As the iGPU does not have any memory of its own. It uses shared CPU cache & system memory.

The theory of cache thrashing seems to be supported by the fact the J1900 Celeron, BayTrail hardware, I have does not show this same symptom when using the iGPU. Instead of a single shared cache it uses a separate for every 2 cores.
33) Message boards : Number crunching : Intel GPU (Message 1654353)
Posted 10 days ago by Profile HAL9000
I think unless someone has a magic formula I am going to forgo the igpu.

AP takes 15 hours to run vs 4.5 on the cores and if I run anything on GPU, all cores take twice as long to do anything.

Does not seem like a winning proposition.

One things that was suggested, & I have not tried. Is running different projects on each hardware type. So like SETI@home on iGPU & Einstein on CPU.

Also 15 hours is a rather long time for the iGPU to do an AP. I was seeing times of about 3 hours for my iGPU 4600 & it is only 100MHz faster than the one in your i5-4570
34) Message boards : Number crunching : Intel GPU (Message 1654352)
Posted 10 days ago by Profile HAL9000
My tests with Intel vTune did not show me any apparent cause differences when running the iGPU to not.

Would it show memory bus overloading anyhow?

It has been some time since I looked at it, but I seem to recall it showed a form of memory i/o for the apps. I just recall that data was very similar between each capture. With nothing standing out, to me, as to what the cause might be.
35) Message boards : Number crunching : Win 10 will be a FREE Upgrade to Win 7 & 8 Users!!! (Message 1654268)
Posted 10 days ago by Profile HAL9000
Windows 10 will be a free upgrade for all installations of win7/8/8.1--genuine or not.

Given they are moving the OS to a platform for services this makes sense. "Here is our OS, now buy our services."


Still leaves the very curious question: if they are going allow free upgrades, even for previously pirated versions, and most of their money comes from Software Assurance anyway, why not just go the full 10 yards and allow Windows 10 to be installed on fresh, new machines too? The idea for Microsoft here is obviously to reduce support costs for older OSes by giving it away as an upgrade. Certainly those remaining few licenses can't be that lucrative!

Yeah it is like they just can't bring themselves to say "Here it is a $0 to everyone." I think resellers, like HP, Dell, & such, are still going to have pay licensing fees. So it may be related to that.
36) Message boards : Number crunching : GPU Problem (Message 1654265)
Posted 10 days ago by Profile HAL9000
Yea, ANY stop/start kicks it back on.

I don't understand why it locks up - it seems totally random. And even more confusing is why it starts back up (seemingly random as well) .... IDK if an update task or something kicks in that causes it to restart automatically.

I have been trying to watch it ... but you know, toast never pops when you watch it.

Very strange. Something seems to be telling the app to snooze, but it does not look like BOINC. Have you tried running without TThottle to see if it may be causing the issue?
37) Message boards : Number crunching : GPU Problem (Message 1654253)
Posted 10 days ago by Profile HAL9000
If you restart BOINC, or stop/start GPU processing, does the GPU activity go back up? Sounds like the driver may have crashed. Leaving the GPU app in lala land.
38) Message boards : Number crunching : Intel GPU (Message 1654248)
Posted 10 days ago by Profile HAL9000
Guess I have some work to do if I want to use GPU

I suspended GPU work and CPU work dropped from 15000 to 20000 seconds to 2500 to 5000, like my other machine that is not using GPU

The CPU times being nearly double when using iGPU+CPU for SETI@home is what I was finding with my Haswell back in July, A journey: iGPU slowing CPU processing. I did some further testing iGPU tuning. My tests with Intel vTune did not show me any apparent cause differences when running the iGPU to not. So I have not yet found an ideal configuration to run iGPU+CPU.
39) Message boards : Number crunching : Win 10 will be a FREE Upgrade to Win 7 & 8 Users!!! (Message 1654236)
Posted 10 days ago by Profile HAL9000
Windows 10 will be a free upgrade for all installations of win7/8/8.1--genuine or not.

Given they are moving the OS to a platform for services this makes sense. "Here is our OS, now buy our services."
40) Message boards : Number crunching : GPU Problem (Message 1654193)
Posted 10 days ago by Profile HAL9000
In BOINC what is the Status of the task. Running, Waiting to run?


Previous 20 · Next 20

Copyright © 2015 University of California