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
1) Message boards : Number crunching : The Dog Days of Summer Challenge (Message 39647)
Posted 3000 days ago by Chuck
PG Admins and Staff,

Thanks VERY much for the hard work in getting things running again.

In light of the internal challenges encountered and impact it had on the goal of the challenge, may I
respectfully ask the Admins, Staff, and Management to consider releasing the system to 'normal' operating
status at this point and rescheduling another go at it?

I think that would be most fair to you folks as you have had to deal with the internal issues beyond your
scope of magnitude and issues simply out of your control.

I would think in the interests of the science & 'friendly challenge' senses, this would make the most sense.

Sorry you had to endure such pain,
I, as a world-wide scale computing engineer, do appreciate,

Auto-pilot and sleep modes on,
Get some rest gang, you deserve it,

Cheers,
Chuck
2) Message boards : Number crunching : The Dog Days of Summer Challenge (Message 39517)
Posted 3001 days ago by Chuck
My compliments to ALL of you for doing such a WONDERFUL JOB!!!!!

I like the effectiveness of your local wu cache management!!!!!

It gave me WUs 2 minutes AFTER the challenge began AND it gave me a bunch.....

In a word.... OUTSTANDING!


Great management of a huge task with limited staff and hardware resources!


I wish you all the best (in getting rest and in the success of the challenge).


Cheers,
Chuck
3) Message boards : Number crunching : PPS (Sieve) Cobblestone Recalibration (Message 38391)
Posted 3042 days ago by Chuck
Regarding GPU version of srsieve / tpsieve / sr1sieve / et ali.

Gang,

I develop GPU apps for a living. My career path is winding down from the huge demand it's been in the past few years and why absent so much since 2005. Doing modeling + graphics at approx 128 Gigabits/frame (multi-panel) with the frame rate of 75 fps makes for some interesting challenges (lighting, hidden line removal, aka... GL-like stuff) without dropping a frame over udp. You can find examples of this on the net... Take a F/A-18 and fly (emulation not simulation) it at mach 2+ and do a high speed barrel roll. That makes for some interesting stresses on the GPUs and CPUs not to mention the LANs (it's all DV until it hits the display system).


Math side of this is (from the DV perspective only) (1920×1080×24×4×9×75)/1048576 = 128144.53125 Mbit/sec

However, with that now 'stable product', and me having moved on to a new task, I have more time (hence my quiet return to PG).

I still have my tools, both ATI and nVidia GPUs, so will be, unless directed/requested otherwise to see what, if anything, I can help port over from Geoff's great code to the GPUs to 'BOINCify' . Geoff took over when I left for work and Joe finished the code. Since then, Geoff has been doing his normal 'fantastic work'.

Those of you who have access to my email (or can get it) please don't hesitate to contact me off-board directly with your thoughts, experiences, or direction on what is best to do. I ask for your guidance.


Thanks,
C.


edit: provide the math behind the work.
4) Message boards : Number crunching : PPS (Sieve) Cobblestone Recalibration (Message 38238)
Posted 3047 days ago by Chuck
First, thank you VERY much for the recalibration announcement being formalized.

Given the tools at PG's disposal, is it now practical to use sr1sieve and single K/file sieves in BOINC? Have the candidate list(s) been sieved to sufficient depth to permit this finer grain sifting or is this not practical at PG which would permit highly efficient memory utilizationand wildly fast CPU and GPU work?

I've been doing theoretical math and even the Riesel Sieve project since the early days. I've also a minor in Mathematics (with my double major EE/CE). I'm doing Quantum work now at work... That will make your head spin around a few times... (LOL).


Thank you for the rapid reply and I hope PG can remain the premier project it's always been.


Thanks,
C
5) Message boards : Proth Prime Search : PPS Sieve credit nerf (Message 38236)
Posted 3047 days ago by Chuck
I'm getting 4127 for the past few days. Runtimes are consistent with what they've been since I started in Jan 2011, but the estimated time provided with the WU is WAY OUT OF WHACK *AND* it's making the machine run a LOT hotter than normal (> 6C for the last 8 days of work)

My 'Cobblestones' hasn't changed... nothing in the machine has changed regarding BOINC. no glibc, etc or other changes which would impact task runs. FC-14 updates coming in are typical gui-only fixes.... no kernel, .ko, .so or other significant changes.

Can someone please look at the WUs being generated to ensure they are being estimated correctly? The PG server is reporting longer than normal runtimes for Windows users (mostly). My preferences now shows the recent average runtime (or similar wording) to have shown an increase.

Is there an announced project work change? More work per WU w/ less credit announcement ? Is this version of tpsieve up to the task? Are GPUs now the preferred method of working on Primegrid tasks? If GPUs are the preferred, could someone please see about expediting the OpenCL availably for all PG tasks?


My nVidia is running the OpenCL but not the Cuda driver because nVidia does not permit/support Cuda in a laptop GPU coprocessor for the primary video driver. Yes, I have two configs... one nVidia w/ 512MB dedicated RAM in one machine, and an ATI 4200 as the display/HDMI (when HDMI activated) processor w/ a 5000 HD series (with 1GB dedicated memory for the GPU) for the 3-D rendering (which I do none of) OpenCL in the other.

I don't understand but windows-7 professional 64-bit has gotten TERRIBLY slow at both PPS Sieve CPU and ati13ati apps using the existing drivers. Windows now takes 50 % longer to turn a task around (CPU) and 2x longer for ati13ati than in March-April. Is this PG unique? The PPS Sieve WUs are coming in under windows-7 at 400+. No other apps have slowed down. They have gotten faster if anything thanks to the fixed by MS.

The current status makes me think my machines are 'running sideways' Yet everything else... Even code I've written/compiled last fall is running the same (or faster). I know tpsieve, sr1sieve and the whole family of programs Geoff has done well. He's well respected in the mathematics community for making things run faster / more efficiently. He's got one unique, great mind when it comes to putting theoretical mathematics (what prime number theory is) into practice as a computer program.

It's coming to the point where giving PG a laptop with a GPU in it is becoming no longer viable. Is this PG's choice or an unknown problem now showing itself so it can be fixed? In contrast... I know an ATI 6900 card in a Windows machine (which can return sieves in ~20 minutes elapsed). it Does have all the current ATI/OpenCL drivers for certain (i have them here for win, but not quite working on linux YET). My Team's participation in PG has dwindled to only 4 from the once 10-15+ we've had a few months ago. Two top producers have moved to other projects. The last 'mainstay' left just the other day. He has a pharm of AMD and NV GPUs.

Help please?

C.
6) Message boards : Proth Prime Search : PPS Sieve credit nerf (Message 37800)
Posted 3057 days ago by Chuck
Thank you to our moderators who stepped in on our behalf and thanks to everyone for speaking up.
My credit values have returned as well


C.
7) Message boards : Proth Prime Search : PPS Sieve credit nerf (Message 37743)
Posted 3058 days ago by Chuck
Gang,
I have double checked my results also and I have a nearly 500 credit reduction for WUs returned after 30 Jun 2011 approx 1200 UTC. This impacts my dual core Intel, my tri core AMD and the Phenom II X6 desktop. Tasks returned on 30 Jun 2011 before this time post the standard 4523 credit for CPU while they now post 4070 credit for the exact same CPUs. There have been no changes in either machine (hardware changes / software updates)

My runtimes are within a few seconds of each other and it appears to be a simple change in credit awarded....


*HOWEVER*

I did notice in the tasks now currently enqueue are showing an estimated time to completion of GREATER THAN 2000 days on the Intel, while similar tasks show the estimated time to completion of 43 days. Regardless of a C2D V2, Phenom II N850 or Phenom II X6 1045T, all return within 24 hrs. The X6 is obviously the fastest at 17.5 hrs, N850 at 21 hrs and the C2D at approx 23 hrs. The X6 is a desktop machine w/ DDR3. The N850 and C2D are laptops with DDR3 and DDR2 respectively.

IF there is anything I can do to help, ****PLEASE**** don't hesitate to ask me.

Tnx,
C.



This has been the normal for me for both Linux laptops.... BOINC 6.10.58 and FC-14 64-bit now for over 3 months.
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2019 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 1.43, 1.50, 1.54
Generated 14 Nov 2019 | 15:23:54 UTC