Client performance - CPU/GPU usage - summary

Case number:671071-990207
Topic:Game: Other
Opened by:Rav3n_pl
Status:Open
Type:Suggestion
Opened on:Sunday, August 21, 2011 - 22:13
Last modified:Wednesday, October 11, 2017 - 23:52

To reduce amount of feedback threads I try to summary some of them into one.
There is list of feedbacks about CPU and GPU usage.

http://fold.it/portal/node/728230 (1)
http://fold.it/portal/node/986056 (1,3)
http://fold.it/portal/node/988114 (2)
http://fold.it/portal/node/988181 (1)
http://fold.it/portal/node/987964 (1)
http://fold.it/portal/node/988293 (4)
http://fold.it/portal/node/990133 (5)

Marked all of them as duplicate to this thread.

Major issues:
1. GUI eats one full CPU core even if we do nothing.
2. Client is not using multi cores - when minimized it is one 100% core thread + up to 15% second thread "flashing"
3. Proteins over 100 residues are almost unplayable on older machines
4. Use GPU acceleration for wiggle/shake operations.
5. Sound thread are working even if we disable all sounds/music

(Sun, 08/21/2011 - 22:13  |  6 comments)


S0ckrates's picture
User offline. Last seen 13 hours 40 min ago. Offline
Joined: 05/19/2017
Groups: None

This would help immensely with helping me to spread Foldit's influence by streaming on Twitch. With Foldit and Open Broadcaster Software eating away at my CPU, it gets laggy trying to wiggle large puzzles. By leveraging my GPU, my gaming PC can really make the game look smooth and ease the burden on my CPU so that it can keep the stream stable while I wiggle. Smoother stream means more professionalism, faster wiggles, and less dead air for me to fill. Plus if other gaming inclined players wanted to make the cross into science, they will have their gaming PCs to leverage while playing.

Joined: 09/24/2012
Groups: Go Science

Memory accumulates over time.

For big puzzles (like drug design), memory usage is a limiting factor.

On Linux Ubuntu, my 8Gb ram gets full after 2-3 days. Then, extra clients close without notice (client crash).

The only way, on all platform, to recover RAM is to restart the computer. then the memory usages starts again from bottom.

Disk usage accumulates over time.

After several years playing, the disk usage of Foldit Folder becomes to be huge.
This seems to slow down the Foldit behaviour.
It's possible to correct it manually (deleting old puzzles).

Is there any way to save disk place by deleting any kind of temporary files at the end of each puzzle? Which files could we delete (manually) in order to save place and still retain some old puzzle results ?

jflat06's picture
User offline. Last seen 1 day 19 hours ago. Offline
Joined: 09/29/2010
Groups: Window Group

For the disk usage issue, we like to let players keep around their old puzzle files so that they can revisit them if they want to. Would maybe a "clear old puzzles" button be helpful?

For files you can delete - anything in the Foldit directory (same folder as the log.txt) of the form 0000XXXXXXX.ir_puzzle, or similar, where XXXXXX is the puzzle ID of an old puzzle. I would not delete anything under
XXX = 800000, as these are the files for the tutorials, and it will probably break the program if you delete them.

For the RAM issue, thanks for the input. It is likely we have some memory leaks somewhere, and would be good to go over them.

However, if closing the program does not return the memory to your system, then this is an operating system bug. The OS is responsible for reclaiming all memory that a program was using when that program halts. Again, that's not to say we don't have memory leaks that need to be fixed.

Joined: 09/24/2012
Groups: Go Science

Disk usage:
Thanks for the tip. I'm interested to keep old puzzle local save solutions.

A button "clean old puzzle temps" would be helpful.

Biggest archive files seem to be IR_SOLUTION files but I suppose these are needed to open old solutions.

Other maps seem to take disk place:
cmp-binary-XXXX
cmp-database-XXXX
cmp-resources-XXXX
puzzles

In the latest, I suppose we can delete suicksave99.ir.solution files without problem.
I suppose that keeping autosave-creditbest.ir_solution is enough.

LociOiling's picture
User offline. Last seen 1 hour 10 min ago. Offline
Joined: 12/27/2012
Groups: Beta Folders

On 1439, I'm seeing the client consistently using more than one logical processor. The 1439 on my four-processor laptop is using about 32% CPU, so around 1.25% logical processors. (On Windows 10. Other platforms might report this as 125% CPU.)

Using Processor Explorer, there's one busy thread as usual, using up to 25% CPU. A second thread is consistently at 5% to 7% CPU. I'm guessing this is the "flashing" thread rav3n_pl mentioned in the original post. No clue what the second active thread is doing, maybe saving and uploading solutions. So instead of "flashing", it's more of a slowly pulsating glow.

Some players complain about large puzzles being unplayable, as rav3n_pl mentioned. Puzzle 1439 still seems playable to me, despite having 301 segments. (The backbone of segments 109 to 301 is frozen, which may help a bit.) The puzzle is of course much slower than usual, so wiggle and shake take a long time, as I'd expect. But I haven't seen any freezes or crashes so far.

I'll try 1439 on my two-processor former Windows XP (now Windows 10) laptop next.

LociOiling's picture
User offline. Last seen 1 hour 10 min ago. Offline
Joined: 12/27/2012
Groups: Beta Folders

On the two-processor laptop with only 2 GB memory, puzzle 1439 is still playable, and even got me a quick rank up running EDRW. This is despite the fact that EDRW had been running for several hours from the exact same start on my "fast" desktop.

The performance aspect is another thing. On the two-processor system, Foldit is using between 60% and 85% of CPU. Again, I'm see one thread that's using close to 100% of a logical processor (or 50% of CPU in Windows terms), and a second, less busy thread. So one client is OK, two clients would be pretty slow.

Sitemap

Developed by: UW Center for Game Science, UW Institute for Protein Design, Northeastern University, Vanderbilt University Meiler Lab, UC Davis
Supported by: DARPA, NSF, NIH, HHMI, Microsoft, Adobe, RosettaCommons