PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
21) Message boards : Prime Sierpinski Problem : PSP Double Check is Complete! (Message 100543)
Posted 1912 days ago by wolfemancs
I managed to get 3 days worth of work in on the small stuff by setting a big buffer and then suspending the long tasks, but I'm up to a week's worth of work on old large tasks (~10 of them), so I guess I'll take a break on the DC to get through those.

See ya guys in a week!
22) Message boards : AP26 - AP27 Search : Should AP27 be GPU-only? (Message 96326)
Posted 2041 days ago by wolfemancs

1) AP27 is going to happen. Relatively soonish. It's got nothing to do with badges. There's no known AP27s and we'd like to find one. When we found the first AP26, it was impractical to look for an AP27 at that time. Six years later technology has improved enough such that it's a reasonable idea. Finding additional AP26s isn't a terrible thing either.

2) Every time I add a project, it makes the "no work" problem a little bit worse. How bad is it? Unfortunately, I don't know. That's a problem. All I have right now are anecdotal reports of not getting work, and my theories about why that's happening. High on my "to-do" list is finding out more about how bad the problem actually is.

3) This thread's purpose is to let all of you voice your opinions about some of the potential strategies we've been talking about internally, as well as to suggest ideas we may not have thought of ouselves.


Since this implies you might be looking for out of the box solutions you might not have thought of, I'll throw this weird one out there.

Change prime reporting to be shared equally by what are now considered the finder and the double checker.

There are a large number of prime finders who work on a 0.0 day queue so that the workunit they are crunching will be as new as possible to increase the chance to be the finder. If reporting was changed to being shared between both results needed to confirm the prime, people could set a higher queue, and while the number of requests wouldn't necessarily go down (they still finish WU just as often, so try to refill their queue just as often), a "no work units available" server response wouldn't be a problem as they'd still have work to crunch, and could wait for the server to refill, and just pull a bigger load next time they connect.

Personally, I don't think I'd feel any less of an accomplishment being a co-finder, and would even feel more of an accomplishment (rather than a feeling of just missing it) when I was the double checker (since I wouldn't know I was the double checker anymore).

CW
23) Message boards : Number crunching : Reign Record Challenge (Message 88446)
Posted 2317 days ago by wolfemancs
I must not understand how the red text on the front page works. . .

I was going to run some ESP Sieve tasks overnight, but when I looked at the front page, the message said:

*** 10 tasks, 9 affecting scoring positions, of Reign Record Challenge (CUL-LLR) cleanup work are currently available! ***


So instead of doing the sieve tasks, I grabbed 3 Cullen tasks to help with the cleanup. But when I look at the details of the 3 tasks I pulled, one of the waiting validations was sent Sep 4th, one Sep 9th (both before the challenge started, and on the third one, I'm the first person to have this test. So obviously none of the three are cleanup for the challenge.

Is there a way to request challenge cleanup tasks, and I just did it wrong? Should the challenge cleanup be heading to the front of the queue somehow? Or is the red text on the front page manually updating, and therefore not real time accurate?

Thanks for the help,
CW
24) Message boards : Problems and Help : ATI PPS Sieve WU sitting at "Read to Start" (Message 60300)
Posted 3335 days ago by wolfemancs
Afterburner is registering about 99% GPU usage. When I ran that one WU I was using the computer, so maybe thats why the time was so long. Weird this time though. I have GPU set to PPS(sieve), and CPU set to PPS (LLR) and (Sophie Germain), and it first downloads 3 CPU PPS (Sieve), and then two more for the video card.

Seems weird.

Currently at 9%, looks like the GPU will take about 45 minutes to complete the task (and about 10 hours for each of the CPU ones. . .

CW

Event Log
12/10/2012 9:56:07 PM | | No config file found - using defaults
12/10/2012 9:56:07 PM | | Starting BOINC client version 7.0.28 for windows_x86_64
12/10/2012 9:56:07 PM | | log flags: file_xfer, sched_ops, task
12/10/2012 9:56:07 PM | | Libraries: libcurl/7.25.0 OpenSSL/1.0.1 zlib/1.2.6
12/10/2012 9:56:07 PM | | Data directory: D:\ProgramData\BOINC
12/10/2012 9:56:07 PM | | Running under account Craig
12/10/2012 9:56:07 PM | | Processor: 4 GenuineIntel Intel(R) Core(TM) i5-3570K CPU @ 3.40GHz [Family 6 Model 58 Stepping 9]
12/10/2012 9:56:07 PM | | Processor: 256.00 KB cache
12/10/2012 9:56:07 PM | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 sse4_1 sse4_2 syscall nx lm vmx tm2 popcnt aes pbe
12/10/2012 9:56:07 PM | | OS: Microsoft Windows 7: Professional x64 Edition, Service Pack 1, (06.01.7601.00)
12/10/2012 9:56:07 PM | | Memory: 7.90 GB physical, 15.79 GB virtual
12/10/2012 9:56:07 PM | | Disk: 931.51 GB total, 889.99 GB free
12/10/2012 9:56:07 PM | | Local time is UTC -8 hours
12/10/2012 9:56:07 PM | | ATI GPU 0: AMD Radeon HD (unknown) (CAL version 1.4.1741, 2048MB, 2008MB available, 6720 GFLOPS peak)
12/10/2012 9:56:07 PM | PrimeGrid | URL http://www.primegrid.com/; Computer ID 259996; resource share 100
12/10/2012 9:56:07 PM | PrimeGrid | General prefs: from PrimeGrid (last modified 09-Dec-2012 00:12:32)
12/10/2012 9:56:07 PM | PrimeGrid | Computer location: home
12/10/2012 9:56:07 PM | | General prefs: using separate prefs for home
12/10/2012 9:56:07 PM | | Reading preferences override file
12/10/2012 9:56:07 PM | | Preferences:
12/10/2012 9:56:07 PM | | max memory usage when active: 4043.19MB
12/10/2012 9:56:07 PM | | max memory usage when idle: 7277.74MB
12/10/2012 9:56:07 PM | | max disk usage: 100.00GB
12/10/2012 9:56:07 PM | | max CPUs used: 3
12/10/2012 9:56:07 PM | | (to change preferences, visit the web site of an attached project, or select Preferences in the Manager)
12/10/2012 9:56:07 PM | | Not using a proxy
25) Message boards : Problems and Help : ATI PPS Sieve WU sitting at "Read to Start" (Message 60281)
Posted 3336 days ago by wolfemancs
I was actually looking at it w/ GPU-Z while it was running, and it said 0% GPU load, but the GPU temps were going up to 55°C, which is what it runs at when the Heaven benchmark is cranking hard, so I'm pretty sure that was just a mis-reading somehow.

I'll check it with afterburner.

I'll get the events log sometime tonight. Long day at work ahead of me.

Thanks.
26) Message boards : Problems and Help : ATI PPS Sieve WU sitting at "Read to Start" (Message 60251)
Posted 3336 days ago by wolfemancs
So I updated the driver. . . and then I turned off the "Suspend GPU" option in the menu of the BOINC manager that I hadn't looked at. (Activity)

Oops.

Thanks for the help though.
CW
27) Message boards : Problems and Help : ATI PPS Sieve WU sitting at "Read to Start" (Message 60237)
Posted 3337 days ago by wolfemancs
I have a AMD HD7870, and it downloads the ATI version of the WU, but it just sits at Ready to Start, and doesn't calculate.

In the BOINC preferences I have the first 3 check boxes all checked (Computing allowed while: Computer on batteries, Computer in Use, Use GPU while computer is in use)
Also in BOINC prefrences I have it set to use 75% of the CPUs (i5 3570K, so 4 cores)

GPU-Z 0.6.6 has the OpenCL box checked, and hovering over it says:
Supported Version: Full
Supported Profile: OpenCL 1.2 AMD-APP
(1016.4)

Catalyst Drivers 12.10. I didn't install anything special, but when you install the catalyst drivers through the package, "AMD APP SDK Runtime (Version 10.0.1016.4)" is selected, and when I run the driver install package it says, 'AMD APP SDK Runtime' is already installed and up to date.

I have this particular computer set as "Home" and both "Default" and "Home" are set to allow ATI GPU and CPU both.


Anything else I'm missing?
Thanks, Craig
28) Message boards : General discussion : Prime Proth points (Message 46743)
Posted 3670 days ago by wolfemancs
I think what you're seeing there is a leftover WU from the last challenge.

Because they were expecting so many PPS Sieve WU to be turned in so fast, they started crunching PPS LLR WU w/ n > 1,000,000. This results in significantly longer work units, which results in significantly higher credits. . .

This looks like it took your computer 1500 seconds to crunch, I'm guessing a normal one takes you much less than that?

Just a guess . . .
29) Message boards : Number crunching : Proposal for the 2012 challenge series (Message 45362)
Posted 3691 days ago by wolfemancs
Assuming my average over 5800+ sieves is familiar, you get 29.4 factors per unit.


at 5800 sieves, I'm guessing you've been sieving for quite a while, including when there were far more candidates to find factors for.

The factors per WU have been decreasing over time. I've only been crunching them in the last couple weeks, and mostly since the increased WU size. I've run 3 of the 6P WU, and 38 of the 9P WU, and my average factors/WU has been bouncing between 7.6 and 8.5, and is currently at 8.3

I'm running an AMD GPU, so much slower than the top NVIDIA cards, but for my particular card, it calculated out to pretty much a toss up whether my Athlon 64 X2 6000+ (20 min/small PPS LLR x 2 cores) or my HD5670 (2.5 hrs/9P PPS Sieve) at 8 factors/ Sieve file.

So at about 8 factors/ sieve (which is what I'm seeing the CURRENT average to be), a dual core w/ a hot new NVIDIA card, would definitely prioritize Sieves, but an older quad core, w/ a middle to higher end AMD GPU might find that LLR tests score them more points, so it's not worth dedicating an entire core to handle the GPU. (Not to mention the dimensions that GPU-LLR could add to the equation)

As to the fact that reliable hosts will still get sent double checks because someone has to double check the unreliable hosts. . . it's probably not worth re-coding everything just for a challenge, but during a challenge, there would be plenty of "only here for the challenge" hosts that would be "un-trusted" to all double check each other, leaving the "trusted" hosts to all work on their own no-double-check-required WU.
30) Message boards : Number crunching : Proposal for the 2012 challenge series (Message 45360)
Posted 3691 days ago by wolfemancs
So you would punish the Prime finders by giving them a 1/2 point? Primes have to be double checked and in the scenario written out earlier in this thread would punish those that have to have a double checker.


Like I said. . . I think it would have to be discussed.

I'm not sure exactly how the optimized double check procedure works. . . If I understand it right, as the server begins to "know" a host, if the host has consistently produced valid double checked WU, the server begins to "trust" the host. Once a host becomes "trusted" they begin to be sent a majority of WU that don't require a double check, and are only occasionally sent a WU that requires a double check, just to make sure dust buildup or other circumstances haven't made the host unreliable.

So I was thinking of the 1/2 vs 1 pt for an LLR test to be more of a reward for a trusted host, rather than a punishment for prime finders.

In reality, it's just a way to measure true work done:
- When a factor is found sieving, it completely eliminates a candidate from the test bank.
- When a trusted computer completes an LLR test that doesn't require a double check it eliminates a candidate from the test bank
- When an LLR test requires a double check, each of the 2 successful tasks eliminates half of the candidate from the test bank.

All though I can see the argument that each test eliminated the candidate, from the bank, we were just making sure, which would give credence to the idea of giving each tester a point for each LLR test validated, regardless of whether it was a single check, or part of a double check. The other downside to this though is total challenge points awarded would not equal total candidates eliminated from the test bank.

I'd be happy either way. The fun of optimizing the hosts, and keeping the fingers crossed for lots of factors/sieve would make this challenge plenty of fun for me.

CW


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2022 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 2.24, 1.58, 1.21
Generated 29 Jan 2022 | 4:27:40 UTC