GPU2 Suggestions Thread

Moderators: Site Moderators, PandeGroup

GPU2 Suggestions Thread

Postby Mechromancer » Thu Apr 10, 2008 9:42 pm

Please try and keep this thread on topic i.e. Suggestions. Use another thread for other discussions (about HT etc.)

I'm making this thread so we can separate all the suggestions/wishes we have for the final release from the main thread. I'm hoping this will keep the main thread clean of anything not directly relating to client or core issues. I would be especially guilty of that. Anyways here goes:

1) PS3-style GUI Display (because its so sexy and I can show it off to my friends)

2) Multi-threaded GPU client. Catalyst drivers support multiple threads, btw. This would help keep the GPUs 100% fed, even on slower machines like my Opty 185@2.8Ghz. Right now my 2900XT is only at ~50% utilization. I already make more PPD with the GPU2 client than I did with the SMP client so this would make sense for a lot of Dual Core machines.

3) Instead of typing flags when executing the EXE, the options for multiple GPUs and other flags could be added in the "configure" screen under a new tab. Flags under tabs would benefit all the other clients as well and make it easier for us noobs.
Mechromancer
 
Posts: 31
Joined: Thu Feb 21, 2008 4:47 pm

GPU2 Suggestions Thread

Postby Mechromancer » Thu Apr 10, 2008 10:03 pm

I was wonder what the time frame on a PS3 style GUI for the new GPU client is. The PS3's display is so nice I could sit back and watch it all day long. I can't wait for the desktop version of that to hit. With that said, is it days or weeks ahead? Or are there plans for one in the first place?


Thread merged. -7im
Mechromancer
 
Posts: 31
Joined: Thu Feb 21, 2008 4:47 pm

Re: PS3-style GUI for GPU2 Client?

Postby 7im » Thu Apr 10, 2008 10:06 pm

Mechromancer wrote:I was wonder what the time frame on a PS3 style GUI for the new GPU client is. The PS3's display is so nice I could sit back and watch it all day long. I can't wait for the desktop version of that to hit. With that said, is it days or weeks ahead? Or are there plans for one in the first place?


Like any development with the FAH project, ETAs are not given, as they almost always change.

GPU2 FAQ: Hey, where did all of the GPU client data files go?

The new GPU client is a SysTray client, similar to the new v6.x CPU SysTray clients, and follows a standardized Windows installation procedure. This new client type is similar in nature to the previous GUI style client, but with notable changes, and a separate visualizations module (to come later).



BTW, I think #3 is a great suggestion, if GUI style clients are the direction the FAH clients are headed.
How to provide enough information to get helpful support
Tell me and I forget. Teach me and I remember. Involve me and I learn.
User avatar
7im
 
Posts: 14648
Joined: Thu Nov 29, 2007 4:30 pm
Location: Arizona

Re: GPU2 Suggestions Thread

Postby Mechromancer » Thu Apr 10, 2008 10:59 pm

Thank you for merging the thread. It makes things a lot cleaner this way. I'll go read the GPU FAQ because they did a great job of answering all my earlier questions, and probably questions I haven't even thought of yet. They covered all the bases.
Mechromancer
 
Posts: 31
Joined: Thu Feb 21, 2008 4:47 pm

Re: GPU2 Suggestions Thread

Postby Mechromancer » Fri Apr 11, 2008 3:12 am

4) I'd also like to see a PPD increase. Why? To make the PPD/watt worthwhile. A lot of quad core users are complaining that the PPD is not worth the wattage compared to their quad cores. Intel quads don't use anywhere near as much power as a GPU, even when overclocked. There has to be an adjustment of PPD to make GPU folding equal to CPU folding with regard to PPD/watt.

The thread was split here as it diverged into a discussion about PPD/W. For the continuation, see here: viewtopic.php?f=10&t=2048 -UF
Mechromancer
 
Posts: 31
Joined: Thu Feb 21, 2008 4:47 pm

Re: GPU2 Suggestions Thread

Postby Galahad » Fri Apr 11, 2008 4:00 am

Mechromancer wrote:There has to be an adjustment of PPD to make GPU folding equal to CPU folding with regard to PPD/watt.


Unfortunately, there's no reason to anyone in Stanford to care about PPD/watt. They care about PPD/science. And they are right.
I know the pieces fit
User avatar
Galahad
 
Posts: 82
Joined: Sun Dec 02, 2007 6:46 am
Location: Russia

Re: GPU2 Suggestions Thread

Postby Belboz99 » Fri Apr 11, 2008 4:23 am

Unfortunately, there's no reason to anyone in Stanford to care about PPD/watt. They care about PPD/science. And they are right.


Unfortuanately, that's where the problem is even more apparent.

The GPU client crunches numbers far faster than even Quad-Core CPU.


Have you taken a look at the TFLOPs generated by the GPU cleint, compared to how many clients it has, vs the SMP client, compared to how many clients it has?



The GPU client is much better for the Science, it's a damn shame I'm going to shut it down after a few WU's tonight because it's not worth the 500ppd drop in production to have to keep the machine running Windows all day. I'd much rather run Linux and keep it folding along at 1,500ppd. ;)

But really, the GPU client is better for science.

Dan
Belboz99
 
Posts: 6
Joined: Thu Mar 06, 2008 4:43 am

Re: GPU2 Suggestions Thread

Postby legoman666 » Fri Apr 11, 2008 5:05 am

Get rid (or have an option to) of the tray icon and allow console viewing/configing (like the SMP client). I don't need more crap in my system tray.
Image
legoman666
 
Posts: 296
Joined: Sat Dec 22, 2007 6:26 pm

Re: GPU2 Suggestions Thread

Postby 7im » Fri Apr 11, 2008 5:09 am

legoman666 wrote:Get rid (or have an option to) of the tray icon and allow console viewing/configing (like the SMP client). I don't need more crap in my system tray.


GPU2 FAQ: What's different between the old and the new FAH GPU client?

Scientifically, the new client introduces several new advances...

Initially, this new client will be a SysTray style client only. A console version may follow later.
User avatar
7im
 
Posts: 14648
Joined: Thu Nov 29, 2007 4:30 pm
Location: Arizona

Re: GPU2 Suggestions Thread

Postby 7im » Fri Apr 11, 2008 5:20 am

Belboz99 wrote:...
Have you taken a look at the TFLOPs generated by the GPU cleint, compared to how many clients it has, vs the SMP client, compared to how many clients it has?

The GPU client is much better for the Science, it's a damn shame I'm going to shut it down after a few WU's tonight because it's not worth the 500ppd drop in production to have to keep the machine running Windows all day. I'd much rather run Linux and keep it folding along at 1,500ppd. ;)


Vijay has posted before that the FLOPS count is disceiving because of the architectural differences between CPUs and GPUs. CPUs can store interim results across multiple cycles. GPUs can not store data across mutliple cycles, therefore the GPU client has to recalculate that interim value each cycle. The GPU may be 20x faster to do calculations, but if the GPU has to do 4x the number of calculations just to complete 1 stup, then the GPU is only really 5 times faster to produce results. 4 steps forward, 1 step back, etc.

GPUs are like top fuel dragsters. They go really fast, for a short distance, but can't carry much. CPUs are minivans. They can calculate forever, carry lots of data with them, but are very slow by comparison. The SMP client sits in the middle, with good speed from 4 CPU motors, and can carry almost as much as CPU clients.

On your other comment, is the PPD difference between the Linux and Win SMP clients really 500 PPD for your system?
User avatar
7im
 
Posts: 14648
Joined: Thu Nov 29, 2007 4:30 pm
Location: Arizona

Re: GPU2 Suggestions Thread

Postby Rattledagger » Fri Apr 11, 2008 7:01 am

The following comment belongs in the PPD/W discussion thread, but since the rest of the post relates to this topic it was not moved -UF
legoman666 wrote:So call it 105w. Assuming 1600PPD, that is 15.2PPD/W. With 3400PPD on a Q6600 @ 105w, that's 32.4PPD/W. Q6600 is still 2.1x more efficient in terms of PPD per Watt.

Is the 25 W or whatever it is for letting one cpu-core run at 100% feeding the GPU included in these numbers or not?


As for suggestions, in random order:
1; Maybe I've become short-sighted, but there is the option to install to non-defalt directory, as mentioned in the FAQ?
2; Option to install for "all users", so regardless of who logs-in they can continue running the same wu and not download a new one if they folds.
3; In my opinion a bug, just putting a shortcut in startup-folder without asking user if this is allowed or not.
4; Maybe goes by my eye-sight again, but how to accurately set to example 95% cpu or "checkpoint every 20 minutes", since AFAIK there's no good indication of anything except "0%-100%" and "3 minute - 30 minute"...
5; Since it's impossible to edit the shortcut, let user set "verbosity 9" and so on in client instead. Also, good idea to let user change this "on the fly", without needing to re-start client.
6; Another bug, Passkey is AFAIK only numbers 0-9 and letters a, b, c, d, e, f, but still accepts whatever you type-in, as long as it's 32 characters. Meaning, user has no idea if he's made a typing-errro or not.
7; A way so don't need to sacrifize a cpu-core to feed the GPU.
8; General FAH-client improvements and bug-fixes, one example is a method so client don't sit idle waiting for the much larger than wu-download and therefore much slower upload to finish.
Rattledagger
 
Posts: 182
Joined: Thu Dec 06, 2007 9:48 pm
Location: Norway

Re: GPU2 Suggestions Thread

Postby Sunin » Fri Apr 11, 2008 4:00 pm

1. I would like to see the ability to use multiple cores to feed the GPU. Currently my 3850 is seeing 50% usage because 1 core of my P4 HT @ 3.6Ghz is not feeding it enough information fast enough! If I could set both the CPU and HT to feed this vid card and could see 100% usage my points would double making it very attractive for those old machines that are folding! Currently my system is churning 2799's at just under 2min per 1%.

2. I would like to see the ability to throttle the GPU %. In some cases it would be nice for heat purposes to limit it to say 80% Similar to how you can limit CPU usage which is a great way to manage temps since GPUs are less likely to have aftermarket HSF installed. Currently I have 6 processors with Aftermarket HSF and only have 1 SLI rig that is under water.

3. Please please please include nvidia in this soon!

4. I don't know if it is planned, but the fact that you can't dictate where to install the client is not fun. It took a little bit to figure out what folder I needed to share so I could properly monitor it with FahMon

5. Have your standard 2.4Ghz CPU process a 2799 and see what PPD it would get. This seems more apples to apples than using a 3850 to bench with.

Still working with someone, I think from here, on benching my system for them and providing some behind the scene info for enhancements for older AGP systems.

Right now my old 1950 Pro AGP = 622 PPD, my new Sapphire 3850 AGP = 722 PPD. Look forward to this new client!

The thread was split here as it diverged into a discussion about hyperthreading. For the continuation of that discussion see here: viewtopic.php?f=10&t=2050&p=17913#p17913 -UF
Sunin
 
Posts: 68
Joined: Fri Apr 11, 2008 3:50 pm

Re: GPU2 Suggestions Thread

Postby 7im » Fri Apr 11, 2008 4:56 pm

Rattledagger wrote:As for suggestions, in random order:


My suggestion is to RTFree FAQ before posting your suggestions, as most were answered in the FAQ.
User avatar
7im
 
Posts: 14648
Joined: Thu Nov 29, 2007 4:30 pm
Location: Arizona

Re: GPU2 Suggestions Thread

Postby 7im » Fri Apr 11, 2008 5:00 pm

Sunin wrote:4. I don't know if it is planned, but the fact that you can't dictate where to install the client is not fun. It took a little bit to figure out what folder I needed to share so I could properly monitor it with FahMon


As noted in the FAQ, you can install the client to run from anywhere you want. However, the data files are anchored per Microsoft specs so that upgrades etc do not affect the data.

And if you wanted, you could move everything to one location, and create a new shortcut, also described in the FAQ.
User avatar
7im
 
Posts: 14648
Joined: Thu Nov 29, 2007 4:30 pm
Location: Arizona

Re: GPU2 Suggestions Thread

Postby Foxery » Fri Apr 11, 2008 6:27 pm

One small thing... Suggest rebooting at the end of the installation instructions. In some cases, the new client is quirky until you reboot.

In my case, the system hard-crashed immediately when I tried to play a video in Media Player. Since rebooting, I've had no further troubles. I skimmed through a few other examples in another thread.
Core2 Quad/Q9300, Radeon 3850/512MB (WinXP SP2)
User avatar
Foxery
 
Posts: 322
Joined: Mon Mar 03, 2008 3:11 am
Location: Syracuse, NY

Next

Return to General GPU client issues

Who is online

Users browsing this forum: No registered users and 1 guest

cron