Welcome to MilkyWay@home

Posts by Andrew McFarland

1) Message boards : Number crunching : A10-7850K APU not recognized correctly (Message 60835)
Posted 27 Jan 2014 by Andrew McFarland
Post:
7 Professional SP1
2) Message boards : Number crunching : Is doing CPU WUs worth it, if you do GPU WUs? (Message 60822)
Posted 26 Jan 2014 by Andrew McFarland
Post:
Short answer: probably not.
Long answer: It depends on how much it takes away from the CPUs feeding the GPUs. My rule of thumb is 1 (dedicated) core per GPU. I found that 4x Radeon 6950's being fed by a quad-core i7, slowed their throughput more than the microscopic credit earned by running CPU-only tasks in tandem.
3) Message boards : Number crunching : A10-7850K APU not recognized correctly (Message 60821)
Posted 26 Jan 2014 by Andrew McFarland
Post:
Just got a Kaveri, using driver version 13.300.0.0 from 1/8/2014, and get this:

Milkyway@Home | Message from server: ATI GPU R600 (R38xx) does not support OpenCL

FWIW, Einstein@home does find it.

4) Message boards : Number crunching : 4/25 8AM-Work Generation issue (Message 54150)
Posted 25 Apr 2012 by Andrew McFarland
Post:
4 hours and counting...
5) Message boards : Number crunching : Download more WUs (Message 54037)
Posted 15 Apr 2012 by Andrew McFarland
Post:
Yeah. I finally just wrote a script to manually update the project periodically.
Change the Timeout parameter (seconds) to suit your needs.

CHDIR C:\Program Files (x86)\BOINC
:BEGIN
boinccmd --project http://milkyway.cs.rpi.edu/milkyway update
TIMEOUT /T 900
GOTO BEGIN
6) Message boards : Number crunching : No ATI work (Message 53657)
Posted 13 Mar 2012 by Andrew McFarland
Post:
I.e. under "Team Challenges" http://boincstats.com/bam/challenges.php

or

"Project Challenges" http://boincstats.com/stats/challenges.php

on the left side navigation frame.
7) Message boards : Number crunching : No ATI work (Message 53649)
Posted 13 Mar 2012 by Andrew McFarland
Post:
Wonder if the current situation is related to the challenge, or coincidence? It's been blissfully rock solid for many weeks now.
8) Message boards : Number crunching : Error: An ATI GPU supporting double precision math is required (Message 53571)
Posted 7 Mar 2012 by Andrew McFarland
Post:
In case you're curious, here's the rundown on the various ATI GPU capabilities
https://secure.wikimedia.org/wikipedia/en/wiki/Comparison_of_AMD_graphics_processing_units#Northern_Islands_.28HD_6xxx.29_series
9) Message boards : Number crunching : Not getting WU's (Message 52953)
Posted 9 Feb 2012 by Andrew McFarland
Post:
... I'm not talking about increasing WU limits, because in my experience, we're clearly not in danger of reaching them (I believe its at 40 per GPU?) Much of the time my client is living hand to mouth, often getting only one WU, finishing it, then getting one more, ad infinitum. Other times, it just runs dry.


Mystery solved. It seems that this behaviour only occurs when i set my additional work buffer to 1 day or less. I set it to 5 days and voila! 20 WU's instantly. The WU cap of 20? = .014 days for a 6950, so i would expect any setting above that to keep me full. Somethings funky here.

Anyhoo... keep up the good work!
10) Message boards : Number crunching : Not getting WU's (Message 52912)
Posted 9 Feb 2012 by Andrew McFarland
Post:
I've been biting my tongue on the work availability issue, because the server is so blissfully stable.[ thanks guys !]

2/8/2012 9:41:25 PM | Milkyway@Home | Sending scheduler request: To fetch work.
2/8/2012 9:41:25 PM | Milkyway@Home | Requesting new tasks for ATI GPU
2/8/2012 9:41:29 PM | Milkyway@Home | Scheduler request completed: got 0 new tasks
2/8/2012 9:41:29 PM | Milkyway@Home | No work available

This isn't really a shock to me because I see that our global throughput has yet to peak. (0.7 Petaflops and counting :-) It does make me curious, though, as to whether is it possible to assess just what percent of requests for work are going unfilled by analyzing server logs, etc.

NOTE: I'm not talking about increasing WU limits, because in my experience, we're clearly not in danger of reaching them (I believe its at 40 per GPU?) Much of the time my client is living hand to mouth, often getting only one WU, finishing it, then getting one more, ad infinitum. Other times, it just runs dry.

In short, how close (or far) are we to having work supply keep up with demand?
11) Message boards : MilkyWay@home Science : Radeon HD 4870x2 berechnungsfehler (Message 52683)
Posted 27 Jan 2012 by Andrew McFarland
Post:
Abnahme Kern-und Speichertakt. Stellen Sie außerdem sicher, sie sind cool. Get neuesten Treiber ...
12) Message boards : Number crunching : GPU reference - Time to complete a WU (Message 52606)
Posted 22 Jan 2012 by Andrew McFarland
Post:

An even sweeter spot is an HD6950 2GB which unlocks to the full HD6970 shaders.


Touche'. I looked for an unlockable reference design, but all the feedback (from Newegg in the States) left me unable to be sure I'd found one that was. Have you bought one recently that you've personally successfully unlocked?
13) Message boards : Number crunching : GPU reference - Time to complete a WU (Message 52597)
Posted 21 Jan 2012 by Andrew McFarland
Post:
If you're going for an ATI/AMD card, the work has already been done for you. Just find the most GFLops for your dollar. You'll find all the details here:
https://secure.wikimedia.org/wikipedia/en/wiki/Comparison_of_AMD_graphics_processing_units

Note that even for a particular model, speeds will depend somewhat on what clocks speed and power limiting you use. For what its worth, I think the current sweet spot is the Gigabyte 6950 1GB: $240USD for a theoretical 2200 GFlops (single precision - double is around 500) I run them at 900MHz clock and +20% power and they generally do MW tasks in about 63 seconds.

Happy crunching!
14) Message boards : Cafe MilkyWay : Project throughput milestone (Message 52578)
Posted 19 Jan 2012 by Andrew McFarland
Post:
Since the new server, the MW throughput has been steadily rising. Yay! Having just passed the 1/2 Petaflop point a day or two ago, I decided it counted as a milestone. I'm saving some champagne for when we make it all the way! :-)

Here's an interesting chart of the Moore's law march of computing power. Crunch on!
http://en.wikipedia.org/wiki/FLOPS#Cost_of_computing
15) Message boards : Application Code Discussion : Problems with AMD Catalyst 11.11 (Message 52553)
Posted 17 Jan 2012 by Andrew McFarland
Post:
AMD Radeon HD 6800 Series (Sapphire/PCPartner)

1/14/2012 8:00:42 PM | Milkyway@Home | Message from server: An ATI GPU supporting double precision math is required

Other projects are utilizing my 6870 just fine just so ya know.


Your 68xx card does not have double precision.
The other projects don't need double precission.
From the 6000 series you would need a 69xx to run MW on it.
See the GPU requirement thread in the number crunching section for more info.


What he said. Here's a page laying out ALL the ATI cards deatails, including which has DP, in nice easy to compare charts.
http://en.wikipedia.org/wiki/Comparison_of_AMD_graphics_processing_units#Northern_Islands_.28HD_6xxx.29_series
16) Message boards : News : Raised GPU work limits (Message 52429)
Posted 12 Jan 2012 by Andrew McFarland
Post:
Just a little bit of optimism for the day... While there may be some work generation issues, I note that the project GFlops is approaching 400K, which is the highest I remember ever seeing it, so we're going in the right direction. Hurray for the new server! :-)
17) Message boards : Number crunching : No GPU tasks? (Message 52266)
Posted 6 Jan 2012 by Andrew McFarland
Post:


With the new server I updated the scheduler to correctly check for the OpenCL double extension rather than just checking the hardware; however only the most recent (and possibly unreleased) clients send back the supported OpenCL extension, so almost no Nvidia/OpenCL tasks are being sent out. I'll fix it soon.


Well, fwiw, I seem to get GPU task VERY sporadically, and when it does, it's usually only one (70 second) task. Came in to find it with no GPU tasks at all this morning...
18) Message boards : Number crunching : Screen saver (Message 52161)
Posted 1 Jan 2012 by Andrew McFarland
Post:
And why the work units are only 6 seconds? Thant`s stessfull...i dont`have enough work for my PC!


Patience, my friend. A new server is coming online in the next week or two, so this should change. Happy new year.
19) Message boards : News : Moving to a new server (Message 52095)
Posted 27 Dec 2011 by Andrew McFarland
Post:
All together now... hurrah! :-)
20) Message boards : Number crunching : New job ? (Message 52055)
Posted 26 Dec 2011 by Andrew McFarland
Post:
Is it really out of work or is something wrong again? :(

Well, the numbers right now don't look good - as many waiting for validation as in progress can't be a good sign.

Tasks ready to send 1
Tasks in progress 101,758
Workunits waiting for validation 94,181
Workunits waiting for assimilation 326
Workunits waiting for file deletion 0
Tasks waiting for file deletion 0
Transitioner backlog (hours) 0


Next 20

©2024 Astroinformatics Group