Posts by cenit
log in
1) Message boards : Number crunching : ATI: OS 10.7 vs Win7 (Message 13472)
Posted 2168 days ago by Profile cenit
With OpenCL, the kernel is compiled at runtime so there is no point where a human can step in and override the code created by the OpenCL compiler.

If you're interested, at milkyway the new opencl app loads the old IL kernel for supported cards and compiles at runtime for the 7xxx series.
If you want to learn that trick, you can maybe ask on the AMD forum or just look at matt's code:
2) Message boards : News : Checkpointing Fixed MAC OS X OpenCL Application (Message 13305)
Posted 2193 days ago by Profile cenit
Are you going to release a OpenCL app for Windows as well, the HD7970's do not have CAL on them.

At Milkyway they have found that running the Nvidia OpenCL app works fine on it though.

I released the OS X app first because there was no AMD support for OS X. That being said, I actually created the app on Windows first because I prefer VS 2010 over XCode as an IDE. So yes, there is a Windows OpenCL version that runs on both nVidia and AMD. Performance is lousy compared to the Brook+/CAL app though. So, I'll have to come up with a way so that if the GPU supports CAL, the server will send it the ATI app and if not, will send the OpenCL app.

well, it should be integrated into boinc. If an app is better on a host, newer boinc software releases should detect it (maybe after some test sending both types of app) and then send the best one!
3) Message boards : Number crunching : OpenCL Apps (Message 12834)
Posted 2285 days ago by Profile cenit
I spent a a week optimizing the code for OpenCL and still have issues with 100% CPU utilization at times (an issue since 11.6 version of the Catalyst drivers) and it still runs 40% slower that Gipsel's hand optimized kernels. He gained 15% in speed on 5xxx and 6xxx cards by tweaking the code, so even eithout that, it is stil 25% slower.

When run on nVidia GPUs on Windows or OS X, it brings the display to a grinding halt. If lessen the utilization, response is better but it takes twice as long to complete a WU compared with the CUDA apps. But, at least it isn't crashing the GPU driver any more.

Since Brook++ isn't supported by AMD any more, I'm thinking about learning PTX/CAL. OpenCL on CPUs is slower than OpenMP as well. But, neither plays well with GPU apps as they really hog the system resources. Most days, the R&D gets me nowhere, but at least I'm learning what doesn't work and we always learn more from our mistakes than our successes. At least that's what I keep telling myself. :-)

Collatz should be a relatively "simple" algorithm... It's a shame that you have so many problems.
Anyway, I don't think that learning CAL should be considered, since it's going to be a deprecated interface (from 2.5 SDK manual).....
4) Message boards : Science : Possible proof (Message 12454)
Posted 2416 days ago by Profile cenit
Well, its not been verified yet, thats `our` job so keep on crunching :¬)

Why do you say that this demonstration is not valid?

edit: just found someone else claiming the proof is incomplete
5) Message boards : Number crunching : OpenCL Apps (Message 12084)
Posted 2469 days ago by Profile cenit
I just read that CAL will be deprecated in the next SDK version, leaving only OpenCL.
Do you know what does it mean? Will it be impossible to use the "old" Gipsel app for everybody (yes I know in this thread you're mainly targeting MACs, on which CAL never existed, but let's talk opencl in general)?
I cannot understand it, and maybe I'll ask it directly to AMD. I cannot understand it because OpenCL runs on top of CAL... Do they want to stop exposing CAL api? Why such a move?!?
6) Message boards : Science : Largest number checked? (Message 8414)
Posted 2770 days ago by Profile cenit
The drawback is that you loose the records, so you don't know how long exactly it took to get to 1.

I do not think that Slicker keeps the records about each and every number checked. In fact, each wu checks millions of numbers but reports back about the steps of the longest one...
7) Message boards : Science : Largest number checked? (Message 6525)
Posted 2866 days ago by Profile cenit
still no one has any idea?
8) Message boards : Science : Largest number checked? (Message 6345)
Posted 2881 days ago by Profile cenit
is there any reason about 2,361,183,346,958,000,000,001 as the starting number?

The old boinc project "3x+1@home" finished at 2,361,495,584,454,253,581,311 and started at 2,361,183,241,434,822,688,255...
i'm not understanding the lower limits... are we sure that lower number are limited in this conjecture?
9) Message boards : Number crunching : Interesting Statistics (Message 5805)
Posted 2913 days ago by Profile cenit
what is the meaning of FLoating Operations Per Second when discussing a project that is doing integer ops?

in any case, these are impressing numbers!
10) Message boards : Number crunching : Are you ready? (Message 3968)
Posted 2973 days ago by Profile cenit
Said to be released "before 2010".

No date set yet.
11) Message boards : Number crunching : Are you ready? (Message 3731)
Posted 2981 days ago by Profile cenit
at the same time, we could say that collatz was not really ready to feed all those new gpus that came here. I cannot connect anymore to the project, apart from this precise moment...
12) Message boards : Number crunching : Are you ready? (Message 3552)
Posted 2987 days ago by Profile cenit
as far as I know, the validator now rejects 2.05 wu.

The problems come from 2.05 wu accepted previously, now that a wingman comes with his result the validator sometime crashes...
13) Message boards : Number crunching : Are you ready? (Message 3525)
Posted 2988 days ago by Profile cenit
I saw the news on the MW homepage today. Eeek! Well, I did a trial run at migrating the database to the new server but didn't go as far as pointing the projects at it.

I'm headed out on vacation for the next 10 days on Thursday a.m., but will hopefully have limited access at times to kick start the validator if needed. So... the big decision.... move the DB to the new server now to reduce the I/O on the current server and hope I can deal with any new issues remotely?

I'm leaning towards leaving it as is with the known issues and migrating over Thanksgiving weekend when I'll have lots of time to deal with any migration issues.

I would do the same. Do not change anything if you cannot babysit it for some days.
Maybe, if it is simple, you could erase any WU left that was done with 2.05 ati. I'm sure it would be appreciated even by those who will loose some pending credits, in the end this project is still at alpha stage and so it should be tolerated, to better manage the days in which you won't be at home and when maybe new people will come here.

(sorry for my english)
14) Message boards : Number crunching : Are you ready? (Message 3517)
Posted 2988 days ago by Profile cenit
it appears that Milkyway is down, and will be down for quite some time. Are you ready to feed all those who will come here? I hope so!
I hope Collatz won't crash on his feet because you don't have the new server already in place.

Oh, and good luck with your project, it doesn't even seem like a hobby. It is so well administered!

Main page · Your account · Message boards

Copyright © 2018 Jon Sonntag; All rights reserved.