GPU assignment of big WUs to slow GPUs

Moderators: Site Moderators, FAHC Science Team

Re: GPU assignment of big WUs to slow GPUs

Postby bruce » Wed Jul 01, 2020 6:29 am

How many hours per day? (0.3 days to complete = 7.2 hours per day.)

If you fold a couple of hours and shut off your computer you really can't expect to get a bonus. That's what baseline points are for.
bruce
 
Posts: 19984
Joined: Thu Nov 29, 2007 11:13 pm
Location: So. Cal.

Re: GPU assignment of big WUs to slow GPUs

Postby aetch » Wed Jul 01, 2020 7:00 am

bruce wrote:If you fold a couple of hours and shut off your computer


I think that's the issue. He can only donate a few hours at a time and wants work units small enough that he can complete, return and provide some benefit.
1). Ryzen 9 3900x, RTX 2070 Super, 16GB, Win 10, F@H 7.6.13
2). i7-4770K, GTX 1080 Ti, 16GB, Win 7, F@H 7.6.13
aetch
 
Posts: 47
Joined: Thu Jun 25, 2020 4:04 pm

Re: GPU assignment of big WUs to slow GPUs

Postby bruce » Wed Jul 01, 2020 7:46 am

Understood, but scientifically speaking slow returns are costly to the projects ... which is why there's a Final Deadline as well as a preferred deadline. What probability would you estimate for the reissued WU to be returned by someone else before he returns the one his computer is sitting on?
bruce
 
Posts: 19984
Joined: Thu Nov 29, 2007 11:13 pm
Location: So. Cal.

Re: GPU assignment of big WUs to slow GPUs

Postby bruce » Wed Jul 01, 2020 7:50 am

rwh202 wrote:The old v6 client allowed you to 'request work units without deadlines' and there was greater use of the 'packet-size' flags to get bigger or smaller WUs (that was mostly concerned about data bandwidth, but had the same effect).


False. The projects without deadlines were discontinued long before that and they were actually not part of FAH though they shared the servers and points system. They were called Genome@home.

the packet-size flag is still operational, but most of the project owners don't bother to set it for their project.
bruce
 
Posts: 19984
Joined: Thu Nov 29, 2007 11:13 pm
Location: So. Cal.

Re: GPU assignment of big WUs to slow GPUs

Postby Joe_H » Wed Jul 01, 2020 3:50 pm

bruce wrote:
rwh202 wrote:The old v6 client allowed you to 'request work units without deadlines' and there was greater use of the 'packet-size' flags to get bigger or smaller WUs (that was mostly concerned about data bandwidth, but had the same effect).


False. The projects without deadlines were discontinued long before that and they were actually not part of FAH though they shared the servers and points system. They were called Genome@home.

the packet-size flag is still operational, but most of the project owners don't bother to set it for their project.


The last is true enough, I've sent off messages more than once during beta testing for projects whose return sizes were much larger than "Normal" so that a "Big" setting would be required to get them. A 130 MB return file takes over half an hour to upload on my DSL connection, the default setting of "Normal" for the client is limited to 25 MB.
Image

iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
Joe_H
Site Admin
 
Posts: 6584
Joined: Tue Apr 21, 2009 5:41 pm
Location: W. MA

Re: GPU assignment of big WUs to slow GPUs

Postby Foliant » Tue Jul 14, 2020 4:22 pm

Today, on my slowest card, i noticed a WU rendering at 54 Minutes per Frame.
Usually i get things done before or slightly after the Timeout.
Image

I have some idea / feature request that might either be verry dumb or smart:

After lets say 5% or 10% progress the client uses a formula involving Estimated TPF to decide if a WU can be calculated within the Expiration (or maybe Timeout +x)and if not can automaticaly dump the WU.
There also should be a Option like autodumpbigtpf v="false" if someone uses old hardware and knows it normally works within or short after the timeout.

I dont know how hard to code that idea is and also dont have the inside knowledge to evaluate if that would help at all.

Regards,
Patrick
1x Celeron J1900 @3Cores - 24/7
1x GTX 760 Ti (GK104) - nearly 24/7
1x GTX 770 (GK104) - nearly 24/7
1x GTX 660 (GK 106) - nearly 24/7
1x GTX 650 (GK 107) - nearly 24/7
1x GTX1660 Super (TU116) - if Idle
Foliant
 
Posts: 40
Joined: Wed May 13, 2020 5:39 pm

Re: GPU assignment of big WUs to slow GPUs

Postby bruce » Tue Jul 14, 2020 4:40 pm

You can post the enhancement request at https://github.com/FoldingAtHome/fah-issues. It might (or might not) get incorporated into the next version of the client. The question that's going to be asked is how big an increase will this produce in total WUs per day.

You probably should either add it to https://github.com/FoldingAtHome/fah-issues/issues/1479 or reference that suggestion.
bruce
 
Posts: 19984
Joined: Thu Nov 29, 2007 11:13 pm
Location: So. Cal.

Re: GPU assignment of big WUs to slow GPUs

Postby Foliant » Wed Jul 15, 2020 6:14 pm

I dont have a github account - could someone throw my idea into the pool for me?
Foliant
 
Posts: 40
Joined: Wed May 13, 2020 5:39 pm

Previous

Return to Discussions of General-FAH topics

Who is online

Users browsing this forum: No registered users and 2 guests

cron