project 2665, project 2662, and improvements to come

Moderators: Site Moderators, FAHC Science Team

kasson
Pande Group Member
Posts: 1459
Joined: Thu Nov 29, 2007 9:37 pm

project 2665, project 2662, and improvements to come

Post by kasson »

We don't normally post regarding projects that are still in beta testing, but we wanted to make an exception in this case:

The SMP client has been yielding a number of important results for us, but its speed has been limited by less-than-ideal scaling (and less-than-optimal CPU utilization) on 4-core and higher systems. We have spent a lot of time working on this and have a new version of the A2 core that does *much* better in this regard. The first versions of this new core are almost ready for release.

Running project 2662 on this new core, we see CPU utilization of >95% on both 4-core and 8-core machines.
In conjunction with the release of project 2662, we have decided to increase the points value of its sister project (using the A1 core), project 2665. New assignments of both projects will now be worth 1920 points. This make the 2665 points yield more similar to most A1 projects; it will also substantially increase the 2662 points yield.

Because the SMP portion of our project does not operate in isolation and because we value donors to all portions of the project, in the long term we will have to re-examine the normalization of SMP points yield relative to standard and GPU clients. We will also be introducing some additional bonus schemes. But these are things to come.

Thanks for folding!
VijayPande
Pande Group Member
Posts: 2058
Joined: Fri Nov 30, 2007 6:25 am
Location: Stanford

Re: project 2665, project 2662, and improvements to come

Post by VijayPande »

I should add that with the new high performance cores starting to mature, I am going to look to rebalance points project wide, including *all* clients. There are some disparities that need to be resolved in some uniform way, with the goal of pushing what's best for the science that's running right now.
Foxery
Posts: 118
Joined: Mon Mar 03, 2008 3:11 am
Hardware configuration: Intel Core2 Quad Q9300 (Intel P35 chipset)
Radeon 3850, 512MB model (Catalyst 8.10)
Windows XP, SP2
Location: Syracuse, NY

Re: project 2665, project 2662, and improvements to come

Post by Foxery »

/lick kasson

:mrgreen:
Core2 Quad/Q9300, Radeon 3850/512MB (WinXP SP2)
jevans64
Posts: 22
Joined: Mon Apr 21, 2008 3:44 am

Re: project 2665, project 2662, and improvements to come

Post by jevans64 »

I was wondering where the 1920 points came from. :lol: I thought I picked up a new project.

Will all projects be able to use the A2 core or does it start with p2662? I am guessing that p2662 will introduce the new core.
Image
Ivoshiee
Site Moderator
Posts: 822
Joined: Sun Dec 02, 2007 12:05 am
Location: Estonia

Re: project 2665, project 2662, and improvements to come

Post by Ivoshiee »

jevans64 wrote:I was wondering where the 1920 points came from. :lol: I thought I picked up a new project.

Will all projects be able to use the A2 core or does it start with p2662? I am guessing that p2662 will introduce the new core.
Normally all old projects have been using the core it is being released with and do not change to the new core mid way.
Mitsimonsta
Posts: 30
Joined: Tue Feb 12, 2008 1:53 am

Re: project 2665, project 2662, and improvements to come

Post by Mitsimonsta »

Thanks for the upgrading of points Kasson. I think this one was needed.

While it is still not in line with the others, I am grateful, and I will shut up now.
Image
jevans64
Posts: 22
Joined: Mon Apr 21, 2008 3:44 am

Re: project 2665, project 2662, and improvements to come

Post by jevans64 »

I know VJ and Kasson will want to stick-whip me for this, but I am running 2 SMP clients on 4 of my quads. It would be great if I could set a flag, like -advmethods, to allow fetching of just A2 units vs. A1 units. That way I could run 2 SMP clients on 2 quads and work A1 units and run 1 SMP client on 4 quads for A2 units. I could also use said flag to set my Duos to catch only A1 units.

Was this a special adjustment or just a re-bench using the same 1760 * completion time?
Image
P5-133XL
Posts: 2948
Joined: Sun Dec 02, 2007 4:36 am
Hardware configuration: Machine #1:

Intel Q9450; 2x2GB=8GB Ram; Gigabyte GA-X48-DS4 Motherboard; PC Power and Cooling Q750 PS; 2x GTX 460; Windows Server 2008 X64 (SP1).

Machine #2:

Intel Q6600; 2x2GB=4GB Ram; Gigabyte GA-X48-DS4 Motherboard; PC Power and Cooling Q750 PS; 2x GTX 460 video card; Windows 7 X64.

Machine 3:

Dell Dimension 8400, 3.2GHz P4 4x512GB Ram, Video card GTX 460, Windows 7 X32

I am currently folding just on the 5x GTX 460's for aprox. 70K PPD
Location: Salem. OR USA

Re: project 2665, project 2662, and improvements to come

Post by P5-133XL »

jevans64 wrote:I know VJ and Kasson will want to stick-whip me for this, but I am running 2 SMP clients on 4 of my quads. It would be great if I could set a flag, like -advmethods, to allow fetching of just A2 units vs. A1 units. That way I could run 2 SMP clients on 2 quads and work A1 units and run 1 SMP client on 4 quads for A2 units. I could also use said flag to set my Duos to catch only A1 units.

Was this a special adjustment or just a re-bench using the same 1760 * completion time?
Like all projects, it is unlikely, there will be a flag controlling what projects an individual is folding, and thereby which cores are used.
Image
toTOW
Site Moderator
Posts: 6359
Joined: Sun Dec 02, 2007 10:38 am
Location: Bordeaux, France
Contact:

Re: project 2665, project 2662, and improvements to come

Post by toTOW »

Especially since running two SMP on a Quad core is totally unsupported. (And remember that the latest linux SMP client supports the -smp X flag, where X>4 ... so the time will soon come that we'll only need one SMP client whatever the number of CPU we have)
Image

Folding@Home beta tester since 2002. Folding Forum moderator since July 2008.
jevans64
Posts: 22
Joined: Mon Apr 21, 2008 3:44 am

Re: project 2665, project 2662, and improvements to come

Post by jevans64 »

toTOW wrote:Especially since running two SMP on a Quad core is totally unsupported. (And remember that the latest linux SMP client supports the -smp X flag, where X>4 ... so the time will soon come that we'll only need one SMP client whatever the number of CPU we have)
I'm part of the 97% of the PC community that is running some form of Windows OS, so the Linux client isn't an option for me.
Image
toTOW
Site Moderator
Posts: 6359
Joined: Sun Dec 02, 2007 10:38 am
Location: Bordeaux, France
Contact:

Re: project 2665, project 2662, and improvements to come

Post by toTOW »

Don't worry, when the core/client will be stable under Linux, it will be ported to Windows ;)
Image

Folding@Home beta tester since 2002. Folding Forum moderator since July 2008.
Sunin
Posts: 22
Joined: Fri Apr 11, 2008 3:50 pm

Re: project 2665, project 2662, and improvements to come

Post by Sunin »

Any updates its been about 2 weeks since the A2 core was deemed coming soon. I'm anxious to see what this new client can do and to really maximize the use of my Quad farm! :)
Flathead74
Posts: 266
Joined: Sun Dec 02, 2007 6:08 pm
Location: Central New York
Contact:

Re: project 2665, project 2662, and improvements to come

Post by Flathead74 »

It has been reported that new work units that utilize the _a2 core should be released soon.
Sunin
Posts: 22
Joined: Fri Apr 11, 2008 3:50 pm

Re: project 2665, project 2662, and improvements to come

Post by Sunin »

Maybe I'm foggy then... is it a new client with the new A2 core WU's and to take advantage of it do you need the latest client? I'm running 5.91 on all my quads.
Xilikon
Posts: 155
Joined: Sun Dec 02, 2007 1:34 pm

Re: project 2665, project 2662, and improvements to come

Post by Xilikon »

Sunin wrote:Maybe I'm foggy then... is it a new client with the new A2 core WU's and to take advantage of it do you need the latest client? I'm running 5.91 on all my quads.
It's not the client which need updated, it's the FahCore_A2.exe which is automatically downloaded by the client to crunch new WU. Right now, it's still not ready for public release.
Image
Post Reply