Folding Forum

As a CPU folder only, I was surprised a few weeks back when I got a project that gave 50,000 base points plus, but where the atom count was only above 20,000 plus atoms (I recall being from the batch of projects 16982 - 16994, as seen here https://apps.foldingathome.org/psummary ) I was surprised at the numbers, especially as my average ppd is around 55,000. I got project 16994 yesterday, which is what reminded me to ask here.
I saw someone on this forum mentioned "how big these projects are", based on how long they take to complete compared to the majority of the others. So i'm interested in understanding why it doesn't seem to matter how many atoms there are in a given work unit compared to how long the ETA is. For example, projects 18206, 18210, 18211 and 18212 all have atoms over 300,000, but the base credit is only 2470. What is it being worked out that makes a project a big one?

There are several things that can make a project "big". The most common ones are the number of atoms involved in the simulation and the number of time steps taken by each WU. Less commonly mentioned is the file size.

The number of atoms is fairly straightforward, more atoms results in more calculations of the forces and motion between all of them. These need to be done for avery possible atom to atom interaction, but for distant pairs that can be approximated.

The number of steps varies between WUs from different projects. The lowest I recall from recent WUs has been 125,000, and the highest is 10,000,000. Each time step is 2 or 4 femtoseconds. All of the CPU projects I know of use 2 fs, some GPU projects use 2 and others use 4.

When it comes to file size there is a fairly direct relationship between the number of atoms and the download file size. More atoms require more entries in the downloaded WU to describe where and what type of atoms is located.. The upload file size though varies more due to differing data being required to be returned from the simulation runs. Some need more detailed and frequent samples from the calculated data than others.

The default "Normal" is for WUs of up to 25 MB, there is a "Small", and "Big" is over 25 MB.

50,000 is not a big CPU project, it's medium at best. There is a 205,000 CPU project live right now with more in the pipeline. https://stats.foldingathome.org/project

I suppose there's essentially two things that go into determining the size of the project - length and breadth.
Length = the length of time being simulated, usually measured in nanoseconds/picoseconds and broken down into steps. The current big projects are simulating 10 million steps. Prior to that they could be from thousands to 2.5 million steps.
Breadth = number of atoms being used to simulate the protein. More atoms mean more calculations for each step.

You can have a large protein with lots of atoms and a short simulation time be worth the same as a small protein with few atoms but a long simulation time because they're both using roughly the same number of calculations.

aetch wrote:50,000 is not a big CPU project, it's medium at best. There is a 205,000 CPU project live right now with more in the pipeline. https://stats.foldingathome.org/project

Oh it is, there are several GPU projects with fewer base points. The 200k one is simply huge.

A few weeks ago I thought it was big viewtopic.php?f=72&t=37374

Not any more.

Base credit is not really a good indicator of the WU size ... short deadline inflates the base credit actificially ...

There are only two concept of big WUs :
- the original one is based on the number of atoms ... also, more atoms requires bigger data files to describe it, so it was the original concept of a big WU (the max-packet-size small/normal/big parameter) and was created when connections where slower and limited ... WUs are considered as big when the number of atoms is above 150-200k ...
- the other concept should be called long WUs ... these are usually simulating longer timesacles ...

The first case is usually easy to handle, since the number of atoms is what drives how a WU will behave on a given hardware (more atoms fits better on GPUs with more SPs or CPUs with more cores) ...

The second case is problematic, because a small WU can be long ... which won't scale well with more SPs or threads but still require a huge amount of compute power ...

Got the picture now, thanks everyone. Just found where "steps" is mentioned in the logs. Which makes me wonder, if displaying steps in the control panel would be another useful bit of information to have. In-consequential I know.

Now currently running 16990. TPF is over 15mins. Looking in the logs, I notice the steps written as
What does the first line represent if the total amount steps to be completed is 10,000,000?

The first step of your WU is step 40,000,000 or after the 4 previous Gens of the particular Run and Clone of the project. Your system will process 10,000,000 steps, and when returned will be used to create the starting conditions for the next Gen WU in the series.

So Gen 0 of the WU will do from 0 to 10,000,000, Gen 1 is created from the returned data and does the next 10,000,000 steps. This repeats until as many Gens are done that are needed or the series reaches a stopping condition.

How do I set my GPU to give preference to big wu? is it "client-type" "big"? I can't find it in the help areas or faqs

If you set max-packet-size extra slot option to a value of big the download will be bigger, but that only very loosely corresponds to the completion time expected.

JimboPalmer wrote:If you set max-packet-size extra slot option to a value of big the download will be bigger, but that only very loosely corresponds to the completion time expected.

Would that be set as (for example): max-packet-size, with a value = big? And also, would it be done in Extra Slot Options or Extra Core Options?

Paul

name max-packet-size value big

I think it should be a slot option, the core doesn't do the upload or download.

I would think it would be either a slot or client option. The client handles the downloads while the core just processes what has been downloaded.

Best way is to test and report back.

I just added it to "Extra client options", I'm hoping it will affect all slots.

I'll have to give it a try on Tuesday, to get my 3950X rolling again once the new PSU is in.

Paul