Bruno - please see the discussion in these feedbacks, there is proof that something is wrong for many computers.
http://fold.it/portal/node/997299
http://fold.it/portal/node/997291
And please add your CPU chip number to Timo's feedback with a note that you are one who has no problems, it will help diagnose faster!
Closing this it can be accessed via search and reopened if necessary.
Since the last update (Begin 2018), I observed a reduction of the number of possible clients on the same computer. This could be linked to the size of the proteins (aflatoxin, symmetric …) and/or the filters.
-reduction from 7 to 2-3 on a Linux platform (I suspect bad memory management problem - the number decreases over time).
-small reduction on a Windows platform
-no apparent reduction on a Mac platform
(this is no true science).
Suggestion for symmetric h Bond filtered puzzles: add a few days (I had no CPU time enough to finish the 1499c one) and/or give us some more days for the "share to scientists" (and some more number of puzzles - I had to delete 2 candidates).
I suspect 1503 could reignite this issue, much slower for me then prior afavatoxin puzzles on multiple computers. May have to reduce clients by half again, below current ( 1 client/physical core ) and yes, I keep my computers clean. I hear new major windows 10 updates in 2 days or so, could really make for some error prone machines well before 1503 finishes.
We should have probably opened a new feedback for the big 2018 update, but here goes.
Using Process Explorer on Windows, I'm seeing Foldit consistently using more than one logical processor.
For example, on my laptop, which has four logical processors, Foldit is often using 30% or more of total CPU processing power, so over 100% of one processor. With three clients running, the laptop is showing 100% CPU most of the time. I'm seeing clients taking nearly 50% of total CPU for at least brief periods.
All three puzzles are relatively large. Even before the big update, large puzzles used more CPU. With smaller puzzles, one thread is normally doing most of the working, taking around 100% of one logical processor. A second thread was active once in a while.
With larger puzzles, the second thread is much busier. I'm currently seeing a third thread using a significant amount of CPU on at least some clients.
The good news is that even with three clients running and Process Explorer showing 100% CPU, I'm still able to use the laptop for other things, and everything is mostly fine. Some things are of course a bit slower.
But I'll probably scale back to two clients on the laptop, as long as we have large puzzles running. Previously, three clients was the rule for this laptop.
I'm also running the aflatoxin puzzle on an older laptop with two logical processors. Several days in, running EDRW on low wiggle power, and no problems. I'll post more about that later.
See Number of processors (Windows) for more on Process Explorer and how to determine the number of logical processors your system has.
"Threads" are mini-programs that are started by a main program. Each Foldit client typically has around 15 threads running. Each thread can run on only one logical processor at a time. When running a recipe or even just a wiggle, there's normally one Foldit thread using about 100% of one logical processor. The problem with the current puzzle mix is that there are two or more busy threads, so one Foldit client is using over 25% of total processing power on my laptop.
Just for a recap of the changes that happened in the latest update from the Rosetta software -
There are 2 big changes that have impacted performance:
1) The score function of the game got slower, as the more accurate computations unfortunately also take more time. But the work they do should be more accurate and valuable.
2) We also split up the computation that happens during a Wiggle and some other tools so that the meat of the work is being done in its own thread, separate from the code that copies those updates into the UI. This means that the 'worker' thread is able to work a lot more efficiently. These threads will automatically get split up into your available logical processors.
The net effect of the update is that the computations are slower because of (1), but you should still see good performance and responsiveness because of change (2). However, if you run multiple clients, the advantage of change (2) becomes diminished, since those other clients will already be using up the other cores' processing power.
Your explanation jflat06 would seem a good reason for removing tracks from the client since it is highly unlikely that the average size of puzzles will be getting smaller. Loci, I use Process Lasso and ALL clients are run at below normal priorities. Been that way virtually from my roots in Foldit and about the only complaints Ive ever had was the increasing folder sizes several years back. I strongly support client diversity as each client creates its own potential. The more the merrier, I'm just a monitor and I like to rag the devs about the over-complexity of the client that actually started with tracks, a looong time before new chapter.
It's a good strategy. Thanks jflat for the explanation.
It's not very important to be able to run more clients, because parallel folding on the same puzzle is exploratory and ultimately a waste of resource (real double blinds are done by different players).
I don't think parallel folding on the same puzzle is wasteful.
For example, on different clients you can try different recipes
(or the same recipe with different input parameters) from the same
starting structure, then when one client gets far ahead of the others,
you can stop the laggards and copy the best structure from the leader
to the rest. Then you can start all clients from this new structure
with different recipes on each client. Kind of like having your own
team of evolvers.
Some recipes use random numbers, so having 2 clients run the exact
same recipe from the exact same starting structure can give very
different results.
Another example is for puzzles known to contain disulfide bonds,
one can have each client try a different combination of cysteine
pairs. Usually some combinations work better than others.
Totally agree here! More clients trying different thing equals more exploration of the solutions landscape.