Folding Forum

Those of you who have been here a while will have seen this with Core_a7 replacing Core_a4.

Core_22 is moving from Beta to Advanced. “Soon” it will be in general release.

Many of you will be frustrated and some of you very much so.

Core_22 does faster Science, and works with the latest AMD GPUs which could not fold with Core_21.

Both the folder volunteers and the researchers at the Universities like better, faster cores. You may think that as soon as there is a stable Core_22, there will be lots of work for it. There won’t be.

Just as we saw with Core_a7 replacing Core_a4, a great deal of WUs are continuations of research already started. Since no researcher is going to throw out his/her existing data, and data using different Cores is not comparable, we may see Core_21 WUs for 3 years.

At first, there will be a trickle of new Core_22 WUs as some researchers have all new Projects they wish to analyze. Over time, the ratio of new projects to continuing projects climbs and in about 3 years those folders who’s GPU does not support Core_22 will complain they have no work.

Again, this should sound familiar to experienced folders. I lost a GPU when Core_11 was replaced by Core_15.

So why will this time be so painful? Owners of AMD DRNA GPUs cannot use Core_21; they will be upset anytime they have no work.

Researchers will start new projects with massive proteins they could not analyze on Core_21. The fastest GPUs for this sort of work are currently by Nvidia. Researchers wanting fast response time, (and they all do) will not want to restrict those cards from using Core_22.

So the best science is going to be done on cards that could do Core_21 WUs, and get lower PPD, while GPUs that can’t do Core_21 will have no work and get 0 PPD. You are saying “Who wins in this scenario?” and the answer is The Researchers. They are in charge and any ‘solution’ that yields slower, less accurate Science will not be approved.

The good news? In 3 years the issue will go away. Meanwhile, do not blame the help on the website, they cannot change these facts.

Yes, this will be a mixed blessing as previous core advances have been, people who have had stable systems with no tweaks will have to readjust to the new environment. AMD gpu folders have had a bad run for quite a while and the core 22 opens a door to great performance, but as you pointed out core 21's will persist.

This has a fantastic 'conspiracy by the Researchers' tone to it and I am very thankful to JimboPalmer for the chance to enjoy this piece of writing. xD

Seriously though, I thought people would be annoyed at how long it took for core22 to get out rather than be annoyed about having a new core. I have a list of over 5 separate scientific questions to answer with the new core already, so just myself I can guarantee you there will be plenty of core22 work available, the beta projects I just put out have been waiting for nearly a year in internal testing for the new core to be ready. Sure, people might still use core21 to finish existing projects (and it's not true that "data using different Cores is not comparable" -- that would mean you could not compare new literature to older literature which used older versions of software, it can work faster but not differently), but we've had a lot of people keen to use core22 as the core21 OpenMM version is so old we're struggling to even make it work on our clusters, and I expect the switch to core22 to be rather smooth.

JimboPalmer touches on two separate problems: a) GPUs that can use core21 but not core22 -- there are no such GPUs, all GPUs currently running can also run core22 and that's one objective of the testing; b) GPUs that cannot run core21 but can run core22 -- that's AMD Navi GPUs -- i said there will now always be Core22 work available and this is not a problem.

All of Folding@home is by and for the Researchers, Core_22 is no different.

I am glad to hear you are thinking of the transition!

I am hopeful that Core_22 gracefully replaces Core_21. I confess I am less sure than you are that there will be no complaints! (You may find 32 bit Linux users with old graphic drivers that do not support Core_22, as an example. They cannot get newer drivers as 32 bit Linux is no longer supported)

I leave it to your judgement to delete the thread if you feel it is not a concern the folders should prepare for. In every previous transition, these issues have come up.

I don't think this transition is going to be as dismal as you're predicting. I don't see anything being taken away. The current projects that are running with Core_21 will gradually be replaced by WUs that run on Core_22, but as far as I can tell, NOTHING is being taken away. In the past, there were a few projects that would run on 32-bit Windows using GPUs that didn't support Double Precision, and those projects have been completed, so if your hardware happens to be that old, you've already experienced whatever loss has happened. Core_22 doesn't change any of that, it just adds support for one upper tier of AMD GPUs.

Similarly, 64-bit operating systems have supplanted 32-bit OSs which once replaced 16-bit OSs. That has nothing to do with FAH; the computer industry moves forward when it wants to and the researchers are not responsible for changes like that... nor are they in any way due to the research improvements in Core_22.

Can core 22 work be ran from core 21 or reverse?
Meaning, is there some sort of compatibility layer between the 2?

We have been waiting for Core22 for over a year. Old hardware has done its job over the years.

We're done with the steam engines. We will now stop diesel engines. We are waiting for electric cars. I don't like them, but they don't ask me about it. Progress.

MeeLee wrote:Can core 22 work be ran from core 21 or reverse?
Meaning, is there some sort of compatibility layer between the 2?

For most GPUs, your PC will have a Core_21 library and a Core_22 library. Any one WU is one or the other.

MeeLee wrote:Can core 22 work be ran from core 21 or reverse?
Meaning, is there some sort of compatibility layer between the 2?

No.

A WU requires a certain minimum revision of a specific FAHCore. A GPU that can run Core_21 can also run Core_22, but not the exact same WU. Certain internal changes have to be made to the project, itself, to upgrade it from running on FAHCore_21 to running on FAHCore_22 and in the process, it is assigned a new project number.

rafwiewiora wrote:This has a fantastic 'conspiracy by the Researchers' tone to it and I am very thankful to JimboPalmer for the chance to enjoy this piece of writing. xD

I am trying to figure out what the problem is. Apparently it is progress. Maybe it will just go away.

PS - All the cards I have tried thus far (GTX 980, 1060, 1070 and 1660 Ti are working great under Ubuntu. I will keep Core 22 while the others can do what they want.