Folding Forum

According to this:
http://folding.stanford.edu/home/core16 ... d-of-life/
Core 16 is done and Core 17 will soon (?) be the new standard.

Can anyone tell if those Core 17 WUs will be like those we get right now (project 8900, others?), or if there will be a way to get smaller WUs?

Because right now I don't think I could continue folding with Core 17. I can manage to get a WU through if I leave the PC running for several days, but that doesn't seem to be very userfriendly.
(Even then it's usually a close shave, my main card needs at least 24 hours - uninterrupted - to pull a single WU through. With core 16, about two hours a day was sufficent to get a WU through every few days.)

I'm not the only one not leaving the computer running all day, am I?

[edit]scratch that, best I remove the line entirely.

What card do you have ? In my rig with two NV cards a 8900 takes
On GTX 780 around 6 hours
On GTX 660Ti around 9 hours

In http://fah-web.stanford.edu/psummary.htm there are also smaller WU like 7810 and 7811; those around 2-3 hours on my 780 ...

There are different sized WU; but I also believe in total those WU will grow in size with the capabilities of hardware. More/faster science ...

As you can see on the Project Summary page, Projects 7810 and 7811 are half the size of P8900.

Threatened? That would imply malice, and that's simply not true.

ChristianVirtual wrote:What card do you have ?

Radeon HD 6970 and 6670 .

There are different sized WU; but I also believe in total those WU will grow in size with the capabilities of hardware. More/faster science ...

They shouldn't outgrow the power of the cards, though.

7im wrote:Threatened? That would imply malice, and that's simply not true.

I regretted the word soon after I posted it. I don't know how better to put it, so I removed the line.

The point I was trying to make is that I fear the new WUs will leave me unable to fold.

781x WU are shorter, though currently there is no setting that allows you to get short/small WUs only unfortunately. It doesn't bother me as I've retired all but my 7xxx AMDs and 6xx Nvidias and I run 24/7, but I certainly understand (and have heard from plenty of people) that the big Core 17 WUs can be a challenge if you can't run 24/7. To get something like this added would likely take some client and server work.

The size of the data returned from p8900s technically makes it a BigWU. It's just not programmed that way. If that's what people need to continue folding, this is where to speak up.

7im wrote:The size of the data returned from p8900s technically makes it a BigWU. It's just not programmed that way. If that's what people need to continue folding, this is where to speak up.

Three issues need examination
1. Do slow GPUs meet the deadlines?
2. Are WUs being classified for packet-size?
3. Can the WU be completed in a reasonable number of days.

1 CPU WUs are categorized and assigned by the number of cores with more/less difficult WUs designed for the hardware. GPUs need a similar concept.

2 . packet size needs to be changed on p8900. This would allow those with slow connections to opt - out of being assigned WUs they can't upload.

3. This isn't a serious issue if concern 1 is resolved. CPU WUs can take several days but as long as the WU has a corresponding deadline it's fine.

Please note that the last paragraph of the blog mentions that some AMD HD 5000 Series GPUs get low PPD and they are trying to figure out a solution to this issue. Once a viable solution is found, am sure that they can easily add other GPUs in it too. We have to wait and see what happens.

Yes, that blog mention does need clarification. It states:

One outstanding issue for Core17 is that while it has much greater performance for most AMD GPUs, it has been reported that PPD is slower for 5xxx class AMD GPUs.

which I think is an over-generalization. For instance, the 5800 series cards are quite capable of handling Core_17 and get better PPD than from Core_16. I suspect all or most of the 5700 cards will do okay as well. Conversely, the lower end 5000 series cards are probably too slow. But the same would hold for many of the lower end 6000 and 7000 series cards, not everyone with cards from those series are running 7800's or 7900's.

PantherX wrote:Please note that the last paragraph of the blog mentions that some AMD HD 5000 Series GPUs get low PPD and they are trying to figure out a solution to this issue.

The way I read it, they do try to find a solution, but it seems they'll continue with or without a solution - so far there isn't one.

Besides, the problem (well, my problem) with the WU size isn't so much PPD, as time to completion. If it takes the GPU a whole day to complete, some might say that a deadline of 6 days is reasonable.
However, that means you're expected to fold at least 4 hours a day. For me, 2 hours a day is realistic - less if the GPU is wholly used. Am I the only one?

[WHGT]Cyberman wrote: Am I the only one?

No you are not. The problem is real! PG knows of the issue. A solution has not yet been found.

What is a reasonable expectation for folding? 1 hour a day? 2 hours? 8 hours? Does 1 hour a day really help when the WU is returned so slowly? Not really, no.

You would be better off letting your computer fold overnight 1 day a week and not fold the rest of the week. The WU would get returned quicker (helps the project) and you would get more points (helps you), and your PC could stay off the rest of the time (less load on the PC). All positives. Fold wisely.

Just for comparison, older CPUs (PIIIs) had to fold at least 8 hours a day to make the deadlines for core_78 work units. Newer CPUs as little as 2 hours a day. But that older tech can't be supported forever, same as any tech. The crushing wheel of obsolescence continues to roll.

Warning: Bitter ranting ahead.

7im wrote:You would be better off letting your computer fold overnight 1 day a week and not fold the rest of the week.

I like folding while I use the computer for other stuff that doesn't need the GPU. With core 16 that was no problem.
It's one thing to have the GPU running in addition to the rest. It's another thing to have the computer running in addition to the GPU.

But that older tech can't be supported forever, same as any tech. The crushing wheel of obsolescence continues to roll.

What is "older tech"? Anything not currently advertised for? I have a hard time seeing a Radeon 6xxx as obsolete, although I agree it's on the lower end of the scale by now, but I would argue it's far from useless.

Much as I like FAH, too often I get the feeling that using anything that isn't top of the line is discouraged, that we should all buy the latest hardware and throw out those obsolete 3 year old graphics cards, even though they continue to bring enough power for purposes other than FAH.

I don't want to sound that bitter, but that core 17 doesn't sit right with me. Powerful as it is, the current project WU take forever to finish, it takes a long time to start up and every time I have to pause it, I lose at least a percent of progress, not to forget the ever-changing tpf estimates.
That may change if the WU size is adjusted, though.

[edit]I would have hoped that with OpenCL older cards would finally be useful for a longer time. Maybe I'm wrong and it's the opposite.

1. Fah encourages faster returns. Not the other way around. You get additional bonus points. Points get added to your total. You don't start with a lot of points and then have them taken away. You are not penalized like that.
2. The above bonuses encourage faster folding. That includes both using faster hardware AND folding more hours per day. Again, not the other way around.
3. I was talking about core 16 going end of life, not your midrange hardware. Your card folds well, just not enough hours per week, by a very small margin.

P.S. The move to bonus based GPU work units was announced a long time ago. This should not be a surprise to anyone. Admittedly, there is a bit of a trade-off here. As with the CPU work units moving to bonuses, the deadlines are shorter, but the potential to earn points goes up considerably.

http://folding.stanford.edu/home/change ... ork-units/ - GPU QRB mentioned (July 2, 2011).
http://folding.stanford.edu/home/unifie ... qual-work/
http://folding.stanford.edu/home/unifie ... plemented/
http://folding.stanford.edu/home/gpu-qrb-update/