Posts by RFGuy_KCCO
log in
1) Message boards : News : Sieve Application Testing (Message 20641)
Posted 942 days ago by RFGuy_KCCO
NOTE FOR Slicker: The way Nvidia implements open_cl you NEED TO RESERVE A FULL CORE for high GPU use. Any way you can modify the Settings on the server so when the task is sent out it RESERVES A FULL CORE on Nvidia GPU's?


Just use an app_config.xml file to specify 1 CPU core per WU. That is the best way to accomplish what you want on your machine.

Here is what I am currently using on all of my boxes:


<app_config>
<app>
<name>solo_collatz</name>
<gpu_versions>
<gpu_usage>1.0</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
<app>
<name>large_collatz</name>
<gpu_versions>
<gpu_usage>1.0</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
<app>
<name>mini_collatz</name>
<gpu_versions>
<gpu_usage>1.0</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
<app>
<name>micro_collatz</name>
<gpu_versions>
<gpu_usage>1.0</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
<app>
<name>collatz_sieve</name>
<gpu_versions>
<gpu_usage>1.0</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
</app_config>
2) Message boards : News : Sieve Application Testing (Message 20570)
Posted 949 days ago by RFGuy_KCCO
Credit for these WU's will definitely need to be adjusted. I am only receiving ~30K points for each of these, yet they run for ~12 hours. I usually get ~240K points per Large WU and they only run for ~6 hours.

When do you plan to adjust the credits granted per WU, Slicker? It isn't currently worth my time to crunch the Sieve WU's at twice the crunching time and only 1/8th the points.
3) Message boards : News : Sieve Application Testing (Message 20559)
Posted 950 days ago by RFGuy_KCCO
I have completed two of these new WU's so far and neither one validated. I have no problems with validation for the original WU's, so something is funky with these sieve WU's. If I have another one fail, I'll be shutting off sieve tasks.
4) Message boards : News : Sieve Application Testing (Message 20550)
Posted 951 days ago by RFGuy_KCCO
Slicker - Do the same .cfg file optimizations apply with the sieve WU's?


The options are not all the same. I created a thread for optimizing the sieve app. http://boinc.thesonntags.com/collatz/forum_thread.php?id=1273


Awesome, thanks!
5) Message boards : News : Sieve Application Testing (Message 20546)
Posted 951 days ago by RFGuy_KCCO
Slicker - Do the same .cfg file optimizations apply with the sieve WU's?
6) Message boards : Number crunching : Credits from Collatz Large, Solo, Mini, Micro (Message 20382)
Posted 1007 days ago by RFGuy_KCCO
The bonus is dependent on run time, not WU type or size. Any WU that takes more than one half hour to complete gets the bonus.
7) Message boards : Number crunching : What is the best GPU for Points Per Day. (Message 20373)
Posted 1011 days ago by RFGuy_KCCO
By the way, the reason I run Large WU's is because Collatz give a point bonus on all WU's that take longer than 1/2 hour to complete. On a 7970, the only WU's that should take longer than 1/2 hour are Large WU's. The Solo WU's take only about 20 minutes on a 7970.


I dropped about 100 seconds off my times for each unit, they are now completing in half the time they were, due to your tweaks...THANK YOU VERY MUCH!!!

I will now have to rethink the mini units I am doing.


I know I am replying to myself, kinda sorta, but I've had to undo the tweak settings for now, the gpu was running too hot, ie 97C!! Right now it is back down to 67C and is back to the old times again. I use MSIafterburner to monitor the fans and it was running at 100% for extended times while crunching.



Yeah, that is one downside to optimizing the app, which increases the GPU usage.

Sounds like you could possibly use some more cooling in your case. Have you checked the fans and cooling on your GPU, to ensure it isn't covered in dust? How much airflow do you have in your case? Which 7970 do you have and what are the clocks?

Collatz will definitely heat up a GPU, especially when running optimized, but 97deg is still pretty high. Mine generally run right around 75deg max, but I have them in cases that have lots of airflow.
8) Message boards : Number crunching : What is the best GPU for Points Per Day. (Message 20366)
Posted 1012 days ago by RFGuy_KCCO
Which GPU video cards produce the most points per day. Nvidia or AMD(ATI)?


For the maximum performance at the best price, I believe the HD7970/280-X is the best choice at Collatz. They also kick ass at Moo!, PrimeGrid (Genefer), Poem, and Milkyway.

You really can't go wrong with a 7970/280-X on Boinc, except for those projects that are Nvidia-only (GPUGrid, BURP, and Asteroids), or perform far better on Nvidia (PrimeGrid-PPS Sieve, and SETI).
9) Message boards : Number crunching : What is the best GPU for Points Per Day. (Message 20365)
Posted 1012 days ago by RFGuy_KCCO
By the way, the reason I run Large WU's is because Collatz give a point bonus on all WU's that take longer than 1/2 hour to complete. On a 7970, the only WU's that should take longer than 1/2 hour are Large WU's. The Solo WU's take only about 20 minutes on a 7970.
10) Message boards : Number crunching : What is the best GPU for Points Per Day. (Message 20363)
Posted 1014 days ago by RFGuy_KCCO
1) Go here:

C:\ProgramData\BOINC\projects\boinc.thesonntags.com_collatz


2) Right-click-->Edit on the file in that directory named:

mini_collatz_6.04_windows_intelx86__opencl_amd_gpu.config

3) Paste the following text exactly as shown into that file, then Save and close the file:

verbose=1
items_per_kernel=22
kernels_per_reduction=9
threads=8
sleep=1

4) Exit and restart BOINC and enjoy much faster crunching times!
11) Message boards : Number crunching : What is the best GPU for Points Per Day. (Message 20361)
Posted 1015 days ago by RFGuy_KCCO
So can anybody update this issue ?

What are the best choices to maximise WU per day ?

TIA

dunx


My AMD 7970 is getting about 300k per day of RAC while running the mini units like this:
200.68 50.33 788.69 Mini Collatz Conjecture v6.04 (opencl_amd_gpu)



You are wasting time and leaving tons of points on the table by crunching Mini WU's. You'd do much better crunching Large WU's. My 7970's average about 1 million PPD each when I dedicate them to Large WU's.


Thanks, those are taking almost 7 hours per unit on your gpu's, mine are taking in the 3 to 4 minute range for each unit, I kind of like the little ones but it seems I may be paying for it. I will check it out.


Actually, most Large WU's take about 5.5 hours on my 7970's. Sometimes longer(still can't figure out why), but mostly in the 5.5 to 6 hour range.

Are you using an optimized config file? Your performance on that 7970 seems slow. For example, my HD7870 completes Mini WU's in 3 minutes almost exactly. I would expect a 7970 to do it closer to 2 minutes. If yours are taking 3 to 4 minutes, I suspect you aren't using an optimized config file.
12) Message boards : Number crunching : What is the best GPU for Points Per Day. (Message 20359)
Posted 1016 days ago by RFGuy_KCCO
So can anybody update this issue ?

What are the best choices to maximise WU per day ?

TIA

dunx


My AMD 7970 is getting about 300k per day of RAC while running the mini units like this:
200.68 50.33 788.69 Mini Collatz Conjecture v6.04 (opencl_amd_gpu)



You are wasting time and leaving tons of points on the table by crunching Mini WU's. You'd do much better crunching Large WU's. My 7970's average about 1 million PPD each when I dedicate them to Large WU's.
13) Message boards : Number crunching : Collatz server seems to be ignoring my resource settings (Message 20280)
Posted 1068 days ago by RFGuy_KCCO
If you are not happy with the way Collatz gives credits then why do you stay there must be a reason. or perhaps could it be the so call inflated credits.


It's a backup project because of when, during Christmas, several projects were down.

Still has nothing to do with why Collatz gave out work when four other projects were online and available for work units.


It is because you don't understand how BOINC works, apparently. The projects don't push work to you, your BOINC client requests specific amounts of work from each project. Don't blame Collatz for how this works, blame BOINC.
14) Message boards : Number crunching : Collatz server seems to be ignoring my resource settings (Message 20279)
Posted 1068 days ago by RFGuy_KCCO
Collatz is set to 100 resource (the base) and the other 8 projects I'm attached to on this machine go up by multiples of 4x from 100. Collatz is set to run on GPU's only as a last resort if Einstein, SETI, Moo! and Prime (the other 4 projects that will give out GPU work units for this pre-Fermi, NVidia Quadro FX 3700 GPU) have no work or their servers are down.

So here is a chart of the resource shares allotted:
Project Resource Credit
Einstein - 409,600 112,000 SETI - 102,400 494,000 PrimeGrid - 1,600 101,777 Moo! - 400 20,566 Collatz - 100 272,988


Now, Collatz gives out extremely inflated credit for WU and is way ahead of almost all the other projects, even though my machines have only given it approx. 96 total hours of work over the Christmas break. It's at half the level of SETI even though at least one of my older machines worked on it for 2 years and SETI is allotted 1,000 times the resource share.

I made sure the servers of the other 4 projects were not down after seeing Collatz download new work. So, why did Collatz flood my GPU with 4 days (my buffer setting because of the typical, expected holiday server downtimes) worth of new WU's that I had to mostly abort (I left the minis run)?


Because Boinc doesn't work like you think it does, Boinc works off the rac of each project and then combined with your percentage setting tries to even them all out. The formula is VERY complicated and has been evolving for a LONG time, and is still being worked on even today.

As for the credits each project gives, they are NOT comparable between projects, each project sets their own amount of granted per workunit credits. They set that number based on a TON of things, more credits means more users hang around but also means a bigger load on the Projects Server. Some projects even get paid for their results, so in some cases that figures into the equation too.

You also cannot compare cpu credits to gpu credits as a gpu can do about 10 times, or more, as much work in the same amount of time a single core of a cpu
can.

The easiest thing for you to do would be to get more machines and try and even out the high paying projects on one machine, or set of machines, and the lower paying projects on other machines. Another option would be to turn the work fetch off for some projects for a given time period, say 30 days, that way the other projects would get exclusivity to your machine for that time period, letting you decide who sends you work and who doesn't. You could also reduce the percentages to zero for some projects, meaning don't get ANY work unless the other projects don't have any. Your setting of 100 only kinda sorta does that, but obviously not to your satisfaction.


OK, you are making some poor assumptions. I have 7 machines running with four of the fastest connected to Collatz and 8 other projects. They have been setup this way since September and yet Collatz still is WAY out of control with credit approaching the second and first highest projects that have 1000+ times the resource share and this was done in under 96 hours. The top two projects were worked on for a couple years with an Intel t7500, among older CPU's.

That is some RIDICULOUS WU inflation for Collatz!!!
GPU efficiencies over CPU on those calculations can not account for that inflation.

Yes, I understand each project sets it's own WU payout and I'm saying Collatz is ridiculously high for a mathematics problem that has no apparent real world usefulness as opposed to say, mapping asteroids and their orbits or folding proteins.

*** That all has nothing to do with the point that Collatz downloaded a lot of work when it's resource share is 1/4000th that of Asteroids@Home WHILE Asteroids@Home was still up and running. ***


Edit: Previous hot-headed response has been deleted after I calmed down. :)
15) Message boards : News : Database Is Back Online (Message 20184)
Posted 1090 days ago by RFGuy_KCCO
Thanks for the updates, Slicker!

I apologize if I came off harshly in the thread I started about the outage over the holidays. I meant no disrespect and absolutely appreciate all that you do running this project. Thank you for your time and efforts!
16) Message boards : Number crunching : Umm... So... What happened over the holidays? (Message 20171)
Posted 1104 days ago by RFGuy_KCCO
At least we deserve a warning of impending down time so we can adjust our
work loads.


The project might've been working fine when he went hunting/went out-of-town and dropped out while he was away.

Kind of hard to warn of "impending down time" when you don't know his impending.

Give the guy a break.


So, it is impossible to simply post something like the following as a courtesy? You know, just like Slicker has already done in the past, AS CAN BE SEEN RIGHT ON THE FRONT PAGE OF THE SITE...

From June 5-15, the temporary Collatz system admin will be the cat sitter. That doesn't sound very reassuring does it? Sorry, but if the server goes down and requires anything other than a simple reboot, there may be an extended outage.

Unfortunately, the chances of that happening are higher than normal since Commonwealth Edison (a.k.a. ComEd, the electrical company in the area) failed to notify me on their original attempt to replace my power meter last week but were unable to do so since it evidently takes at least two people to do it. Since they didn't call before the last attempt, Murphy's Law suggests that they will not call before the next attempt and that it will take place while I am on vacation. The cat sitter doesn't know how to switch over to the generator if the outage exceeds the UPS battery so if the power is out for an extended period, Collatz may remain down until I get back.


Look, many of us invest thousands of dollars to support these projects. All I expect in return is the courtesy of a notice. I think if a person or entity begins a project like this, it is incumbent upon that person or entity to keep the people crunching for that project informed of these things. I am not asking for compensation or anything special. A simple sentence or two on the front page would suffice. Also, I understand that sometimes things go down without warning. It happens. In those cases, a sentence or two stating what happened after the site comes back up would be nice.

Some of you are acting like I am asking for the moon, or something. Remember, when we crunch for any of these projects, we are essentially doing a favor for the project administrator(s). We spend money on hardware and electricity, after all, and they get work completed on their project. This is a two way street... Well, it should be, anyway...
17) Message boards : Number crunching : Umm... So... What happened over the holidays? (Message 20166)
Posted 1105 days ago by RFGuy_KCCO
I understand Slicker runs this project from home, etc. I guess I am just disappointed that this was an extended outage and it wasn't even acknowledged. This has happened frequently lately (though not for nearly this long), compounding the issue. I fully understand stuff happens; all I ask is for some communication to let us know what happened. Perhaps my expectations are too high?
18) Message boards : Number crunching : Mixed nvidia and AMD drivers (Message 20162)
Posted 1106 days ago by RFGuy_KCCO
I have an AMD 7870 and an Nvidia 780ti humming along nicely in the box I am currently typing this on, which runs Windows 8.1. No issues whatsoever.
19) Message boards : Number crunching : Umm... So... What happened over the holidays? (Message 20159)
Posted 1106 days ago by RFGuy_KCCO
So the project is back up, but without a peep from Slicker to explain the extended outage. This fact alone makes me reconsider my support for Collatz. I'm sorry if anyone disagrees, but I feel all the people who contribute their hard earned dollars to this project at least deserve the courtesy of being informed of what happened. I am hopeful an explanation is forthcoming.
20) Message boards : Number crunching : AMD GPU reduce clock under load (Message 19853)
Posted 1223 days ago by RFGuy_KCCO
Ah, I understand now - if you don't reserve a CPU core for the AMD OpenCL apps, they will run very slowly because the GPU is starved for work from the CPU.

I use an app_config.xml to allocate one full CPU core to each AMD OpenCL GPU task. It isn't necessary for the CUDA apps, as they run just fine with <25% of a core allocated, usually only using <2% of a core on both my GTX770's and my GTX780Ti. However, to get the best completion times out of the AMD OpenCL apps, a full core is absolutely essential. I find these apps will use ~95% of a core when they run. This includes all of my AMD GPU's - HD7870, HD7950, HD7970, R9 280 (same as an HD7950), and an R9 280X (same as an HD7970). All of them are CPU hogs. Since you want to focus on CPU projects, reserving a core isn't an option, which is understandable.

When choosing which sized WU's you want to run, remember this: 30 minutes of processing time is key. You want to pick a size that requires >30 minutes to complete, that way you get bonus points for each WU. Also, keep in mind that each size up from the Micro WU's is a 16X increase in WU size (and processing time, of course). So, Mini = 16X Micro, Solo = 16X Mini, and Large = 16X Solo.

Happy crunching!


Next 20

Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.