Folding Forum

bruce wrote:

bcavnaugh wrote:

bruce wrote:When was the last time a p8018 was downloaded? There was a note a few hours ago that they changed something this afternoon.

Please restart one of the clients and post the first few pages of the log.

Last Download was 24 March before that 14 March. Current Download 27 March 8:38 AM MST

OK so your are saying to Kill the 3 I now have running and delete my work and cores and then restart my computer and this might fix my below issue?

I absolutely did NOT say to delete any work units. I did say post your log and you have not done that. I guess it's time for me to stop trying to helping you.

I am sorry, I did not say that you did, I had to leave to work as you do and had to go to work today as well.

I took your restart as starting a new project I don't no how to stop one of 3 running GPU's and then restarting them without starting them over.
Call me dumb but I don't know all the ins and outs of your software.
I pay out $450 to $460 a month of my own money to Fold. I am not one of your corporate partners and can't write off the cost off Folding out of taxes. Yes you have one more mad single donor.

Highlight the gpu slot in V7
right click on it - should bring up a drop down list
then select stop
then select start
this will restart a WU from last position or checkpoint

Core 17 - wu 9401 seems to be down again - http://fah-web.stanford.edu/logs/171.67.108.31.log.html

Core 15 and 8018 is back on

kiore wrote:

jimerickson wrote:fold what you get.

I agree, these units need to be completed and people dumping just mean they come around again.. and things are delayed, for the sake of a a few days lower PPD.

Admittedly, PPD is the only (a priori) metric of scientific usefulness which even the most altruistic of us have access to; I myself switched over to a Linux OS for folding since
(1) Linux tends to be better than windows at not wasting computational power on overhead such as a GUI.
(2) (at least momentarily, and as mentioned earlier) Linux is assigned only core_17 work units.

Many of us just want to be helpful, and Stanford's PPD figures are the most direct feedback about how helpful we're being.

Granted, standardizing PPD is a nightmarish task for any distributed computing project, and there will necessarily be inconsistencies and errors which the Pande Group has made heroic efforts to minimize, but getting core_15s and the attendant PPD drop made me personally feel like I wasn't being as helpful to the Pande Group as I was when folding core_17s.

1 is a myth.
2 is a limited offer, only if you have NV cards.

Linux is better at SMP than Windows. That's why Linux performs better. GUI has no impact on SSE or floating point operations, where 90% of FAH does it's work.

PPD on core_15 hasn't changed. Doesn't make me feel like I contribute any less when I was folding the same WUs for the same points a year ago.

I do feel differently about my contributions compared to a year ago. Getting an 8018 WU is like stepping back in time. Both time and technology move forward. A year ago I ran mostly Fermi GPUs. I retired those systems and now run the most advanced offerings from NV and AMD. I have upgraded to improve my effective contributed value to FAH. That value is informed by the points system. When an 8018 sucks up more power and delivers fewer points, it's not just annoying, it's shameful because the value of the potential contributions lost by my GPUs to time spent on substantially less valuable (yes, fewer points means less value - not an arguable point) WUs cannot ever be recovered. Given the substantial point difference multiple 780ti systems between Core 15 and Core 17, the opportunity cost for FAH is enormous. And disappointing from a donor perspective.

I think that the fact that FAH is running a project at all proves that that project is valuable. They would not run projects that are not valuable. It seems that some of the older projects that generated less points have not completed yet, and that is why they are still around, not that they are not valuable.

Only 8018 on core 15 again, back server with core 17 online! Not a joke!

Core15 announced "end of life" four months ago "! Clip from:
https://folding.stanford.edu/home/blog/page/5/


Changes in Core17: update and move to full fah
November 7, 2013 by Vijay Pande ·

With FAH GPU Core17 taking over the lion share of our calculations, with the older cores (especially core15 and core16) phasing out (as previously announced), we’re rolling out core 17 to full fah (not just adv). This should hopefully resolve the issue that many GPU donors have been seeing in terms of not getting any work, now that Core15 and Core16 jobs are at end of life.

Yes, but EOL does not mean there is no Core_15 work to finish up. EOL means new research projects using GPU folding will be using Core_17. As I recall there has only been one project released since the announcement to use Core_15, and that was just a variation on an existing project for that research group. In the meantime the projects already released for 15 still need to be finished. Given that some folding projects can take a year or more to finish due to the serial nature of the WU's, 4 months is just a fraction of the time that could be needed to clear them out.

Core_16 is a bit different. Only 2 projects were ever released to full folding, and they essentially were almost done when the announcement was made.

NookieBandit wrote: (yes, fewer points means less value - not an arguable point)

Yes, it IS arguable. It is known that Core_17 projects are assigned points on a different basis -- using a Quick Return Bonus that was not available on Core_15. So the ppd gained are simply not comparable. The only available comparison is the base ppd (without QRB). If you have that comparison, and it shows a substantial difference, you have a valid case. There is no case to be made by comparing total ppd.

I too wish that points were more easily comparable -- particularly since the Core_15 end of life is taking so long. But given the known difference between the points basis for the two cores, I am not concerned by a lack of ppd when I happen to be assigned Core_15 work. It does NOT reflect the relative scientific value of projects based on different cores.

David

David, you and I may simply have to agree to disagree. Here's why:

Core 15 does implicit calculations, while Core 17 does both explicit and implicit, and effectively can run the same calculations as an SMP WU. I'm sure one of the decision criteria on initial points assignment of Core 17 WU was the comparison of the value of the work performed against that performed by Core 15 and SMP cores. The conclusion had to be that the tremendous value provided by Core 17 WUs was worth a substantial premium. The points selection and QRB accelerator were not picked at random, but by some estimation of relative value to what was, at the time, currently known. Therefore, the point system used between Cores is a representation of a derived relative value and comparable across work units.

That being the case -- and the primary point of my post -- was the very real opportunity cost for the FAH program by running Core 15 work units on high-end Nvidia GPUs that are not optimized for the Core 15 architecture. Every 8018 received by a 780/ti is time that could have been more valuably deployed on a Core 17 work unit. Secondarily, from a donor perspective, it pains me to see my systems delivering 16% of the value they could otherwise be delivering (225,000 vs 37,000 ppd) by running Core 15 WUs, all the while consuming more energy in the process.

It seems reasonable that FAH should be able to detect the type of donor GPU and consistently assign WUs that most effectively utilize that hardware architecture. If that was the case, I would un-retire all my Fermi GPUs and have them happily crunch 8018's all day long, as that would be the optimal use of that hardware architecture. At the same time, IMHO, they are not at all optimal for running Core 17 which is why they stay retired. The gestation period for changes like that can take, charitably speaking, a very long time even though it would substantially optimize the FAH program relative to the opportunity cost of not doing so.

What would the "opportunity cost" be for a GPU that is not folding at all is the flip side of your question. Yes, PG is working on getting more Core_17 projects into production, but that takes time. Add that at the moment it appears one server's projects may have completed, and another server is down for the second time in a week, that leaves only one server available. They have greatly increased the number of available WU's on that server in the last week, but it can only supply so many systems at a time.

The rest of your suggestions would take a bunch of programming time to modify the assignment and work servers and how they handle requests. That resource, programming time, has often been stated to be in short supply by PG. As I understand it, much of it is currently being used to bring the NaCL beta folding client to full availability.

-alias- wrote:Core15 announced "end of life" four months ago "! Clip from:
https://folding.stanford.edu/home/blog/page/5/


Changes in Core17: update and move to full fah
November 7, 2013 by Vijay Pande ·

With FAH GPU Core17 taking over the lion share of our calculations, with the older cores (especially core15 and core16) phasing out (as previously announced), we’re rolling out core 17 to full fah (not just adv). This should hopefully resolve the issue that many GPU donors have been seeing in terms of not getting any work, now that Core15 and Core16 jobs are at end of life.

Yes my friend, when i reed this i buying 5 x evga 780ti sc. So when i see core 15 now i set my client to pause.

My pc with 2 Titan's and 2 AMD 6238's (24 total cores) (but downsized to 20 working cores) running Win 7 went from 300,000+ ppd to 83,000 ppd because of the p8018's, but that's the roll of the dice and I still can remember when I started of jumping for joy when I got a work unit that gave me 7 points when finishing it instead of the regular 5 points. LOL.

To the theme of Rawhide
Folding, Folding, Folding