New core = significant production drop GPU

[mmonnin]

From what I've read around here, science from 2.25 is still good.

[Grandpa_01]

From what I've read around here, science from 2.25 is still good.

I believe the science from core 2.25 is good they do internal testing before releasing a core. And there was a purpose for releasing it so that Kepler cards could fold, also from what PG members have said before there were some environmental variables added to help with screen lag. But for me the core is no where near as stable or fast as the 2.22 core is for Fermin cards as I posted the other day I lost 70% of the WU's I ran with it with a normally stable OC so it was double penalising me (The core has a 30% slower tpf and I had to drop the OC from 920 to 900 core clock and lost another 15%). It is really no big deal it is there project, my cards and 1 of the 4P's are doing other thing now where I feel they are being utilised to there full potential and getting adequate rewards for doing them.

[mmonnin]

Some have mentioned the p7626 WUs run several degrees hotter than other WUs. Maybe close to that cliff for your card at the OC

[kiore]

Some have mentioned the p7626 WUs run several degrees hotter than other WUs. Maybe close to that cliff for your card at the OC

In my experience they do run hotter, but I think grandpa is referring to all units even 'cooler' ones and it is pretty clear that TPF and PPD on 2.22 was a lot faster on the same units than 2.25 is. Prof Pande has said in relation to the memtest errors that 2.25 is 'stricter' on memory, this may be the better science they are looking for. Really we have no clear indication why the project has preferred 2.25 over 2.22 other than it looks like it is to be a unified core, works on Kepler cards, has environmental changes to reduce some previous lag issues and the more recent indication that it is designed to improve science by being 'stricter' on memory faults.
Me, I personally preferred the 2.22 as a donator it gave me better results (eg points), but the project seems to prefer2.25 and I assume this is for the same reason.
Perhaps this weekends issues with the rollout of 2.25 and the G80 and ATI cards, will result in a clearer understanding of these things.

[Grandpa_01]

What would be really nice is PG comes out and tells us why the forced 2.25, is it doing better science ? is it so people folding Kepler's will feel better ? is it to deal with the screen lag issues some are seeing ? s it that they need it for the future WU's because Kepler can not fold without it and 2.22 does not work properly on the new Kepler designed WU's. It is not that hard to give the reason if they are ligament reason's and would go along way toward PR after all we do give our time, $$$ and equipment to PG, and a little info and accountability to the donors is a very small price to pay for what is being given.

[mdk777]

well, Grandpa, we say the same thing over and over and don't get any response.

I suggest you take it up with your team representative...and organize action accordingly if you know what I mean.

[k1wi]

In the wider charity/non profit environment there is a recognition that due to the tightening economy donors are becoming increasingly savvy with who they donate to and how try distribute their donations, in order to feel that they are doing the most good with their limited donations.

This has forced all charities to up their accountability, transparency and efficiency. The good charities continue to do well while many hit the wall.

I think folders are really becoming increasingly savvy, in particular much more knowledgeable about the mechanics behind folding. Which is a really positive thing for the wider scientific field/s. Therefore communication is becoming increasingly important. Communication that is timely and comprehensive.

For what it's worth, I think PG has lifted it's game latey and reconition needs to be given where it is due. but all communication is good communication and there is always room for improvement - that's why people like grandpa are important.

[mmonnin]

Some have mentioned the p7626 WUs run several degrees hotter than other WUs. Maybe close to that cliff for your card at the OC

In my experience they do run hotter, but I think grandpa is referring to all units even 'cooler' ones and it is pretty clear that TPF and PPD on 2.22 was a lot faster on the same units than 2.25 is. Prof Pande has said in relation to the memtest errors that 2.25 is 'stricter' on memory, this may be the better science they are looking for. Really we have no clear indication why the project has preferred 2.25 over 2.22 other than it looks like it is to be a unified core, works on Kepler cards, has environmental changes to reduce some previous lag issues and the more recent indication that it is designed to improve science by being 'stricter' on memory faults.
Me, I personally preferred the 2.22 as a donator it gave me better results (eg points), but the project seems to prefer2.25 and I assume this is for the same reason.
Perhaps this weekends issues with the rollout of 2.25 and the G80 and ATI cards, will result in a clearer understanding of these things.

Pande group doesnt have the bandwidth to make separate cores just for Kepler or just for Fermi. Kepler requires v2.25 so thas where the project is going. That is that. I can't believe all the crying over this. QRB will be back and make all this irrelevant.

The temperature thing I was referencing had nothing to do with 2.22 or 2.25. They run the same. The particular set of WUs I mentioned run hotter, and on an OC card it might now be too much.

Vijay is also taking time out of his weekend to work on today's issues. On the day of his child's birthday party.

[mdk777]

Pande group doesnt have the bandwidth to make separate cores just for Kepler or just for Fermi. Kepler requires v2.25 so thas where the project is going. That is that. I can't believe all the crying over this.

Your presumption.

All Grandpa, (and several dozen others), did was ask for a confirmation of this, if it is in fact the case.

The complaint is never about the action, it is always about how it is not explained.

I for one am tired of asking...so I will be dropping out again...

I had hoped that interesting things were going to happen on the GPU front...but after the usual weeks drawing out to months of non comment and non commit...

well, I guess I'll just check in again in a year or two and see if anything has changed.

PS. I have not seen for example, any response to this excellent post:

viewtopic.php?f=16&t=23074&start=30#p230056

Perhaps Grandpa and I are ignored because of our consistent and perhaps redundant requests.... but when someone (who has donated 15 million points) goes to the trouble of writing such a cogent and well organized proposal, he deserves a reply.
That he never got one says something important.

[mmonnin]

Big time spenders like Grandpa come and go. Its a fact of any DC project. It doesn't give him or anyone else special treatment because they spent money to get their competitive fix on.

Pretty sure Vijay has some actual science to do instead of answering our questions all the time. That's what the site admins are here to do. Just compare the post count between bruce and any of the PG staff.

I started folding back when WUs were single digits. I can remember 1.2 and 2 point WUs and was once in the top 100. Now a 580 running a p8057 did what I did my first few months in a matter of minutes. PG group have seen people complain about points for years and years and hardware not matching up to others. This is nothing new. The ones that stay do it for the science, not because their hardware isn't optimized due to PG changing something not to their liking.

[mdk777]

PS. I have not seen for example, any response to this excellent post:

viewtopic.php?f=16&t=23074&start=30#p230056

didn't bother to read the post I take it.

Nothing in it complaining about points.

"science" doesn't happen in a vacuum. It takes place in society...esp. a DC project as you mention.

That's what the site admins are here to do.

Except of course that they cannot answer questions that they have no information on.

It doesn't give him or anyone else special treatment

Nope, good communication benefits ALL...

[sswilson]

Big time spenders like Grandpa come and go. Its a fact of any DC project. It doesn't give him or anyone else special treatment because they spent money to get their competitive fix on.

Pretty sure Vijay has some actual science to do instead of answering our questions all the time. That's what the site admins are here to do. Just compare the post count between bruce and any of the PG staff.

I started folding back when WUs were single digits. I can remember 1.2 and 2 point WUs and was once in the top 100. Now a 580 running a p8057 did what I did my first few months in a matter of minutes. PG group have seen people complain about points for years and years and hardware not matching up to others. This is nothing new. The ones that stay do it for the science, not because their hardware isn't optimized due to PG changing something not to their liking.

Unfortunately the forum admins are exactly that, they're moderators for the forums and aren't in the loop so the best they can offer is educated guesses.

I'm sorry you feel that way about the so called "big time spenders", but our desire for better information about what we're investing time, money, and hardware into sounds to me like a pretty reasonable request. We're not asking for (or demanding) changes to the project, we're asking for clear and timely information about what it is our hardware is doing. We're also not asking for special treatment, just the same common respect I'd hope anybody involved in a project deserves, let alone those who donate their time and money freely.

What often gets lost in the shuffle is that there are two major components to this project... one is the scientific number crunching, and the other is the method of getting that crunching done which is a distributed computing model. Like it or not, DC models require that attention be paid to the social engineering component of keeping your volunteers engaged, and currently that requirement is not being met. The question that really needs to be asked is if you're interested in keeping the folders who dedicate time and resources 24/7 to the project, or if you're going to be satisfied with a smaller base of volunteers who will try FAH out for a week or so and then uninstall it because it isn't doing anything for them.

As far as folders coming and going... I've got a few years worth of folding under my belt as well, but as has been stated by other members of the forums here, I've got other options to dedicate my hardware to if this project can't get it's head around the fact that volunteers need timely communication.

[codysluder]

What would be really nice is PG comes out and tells us why the forced 2.25, is it doing better science ? is it so people folding Kepler's will feel better ? is it to deal with the screen lag issues some are seeing ? s it that they need it for the future WU's because Kepler can not fold without it and 2.22 does not work properly on the new Kepler designed WU's. It is not that hard to give the reason if they are ligament reason's and would go along way toward PR after all we do give our time, $$$ and equipment to PG, and a little info and accountability to the donors is a very small price to pay for what is being given.

It seems pretty obvious to me. They're forcing 2.25 because they have to support Kepler.

How is Kepler being supported? NVidia came out with a new version of drivers and a new version of CUDA. You have to install the new drivers and Stanford has to start using the new CUDA. Version 2.22 certainly uses the old version of CUDA and 2.25 must use the new version of CUDA. Can anybody honestly say they know that something else was changed? Why are asking Stanford what they changed. Is anybody complaining to NVidia? If not, why not?

Stanford respects their partners and isn't likely to deflect the to blame to NV even if my assumptions are 100% correct. It's better just to say nothing and wait for NV to fix the performances losses encountered in the current version of CUDA.

[Nathan_P]

Big time spenders like Grandpa come and go. Its a fact of any DC project. It doesn't give him or anyone else special treatment because they spent money to get their competitive fix on.

Thats a bit unfair, Grandpa has been with the project nearly 5 years now has 700m+ points and 29.5k WU under his belt, not exactly someone who comes and goes. Actually most of the big spenders stay with the project through thick and thin as they have invested a lot of money because the believe in the project's goals.

[Grandpa_01]

What would be really nice is PG comes out and tells us why the forced 2.25, is it doing better science ? is it so people folding Kepler's will feel better ? is it to deal with the screen lag issues some are seeing ? s it that they need it for the future WU's because Kepler can not fold without it and 2.22 does not work properly on the new Kepler designed WU's. It is not that hard to give the reason if they are ligament reason's and would go along way toward PR after all we do give our time, $$$ and equipment to PG, and a little info and accountability to the donors is a very small price to pay for what is being given.

It seems pretty obvious to me. They're forcing 2.25 because they have to support Kepler.

How is Kepler being supported? NVidia came out with a new version of drivers and a new version of CUDA. You have to install the new drivers and Stanford has to start using the new CUDA. Version 2.22 certainly uses the old version of CUDA and 2.25 must use the new version of CUDA. Can anybody honestly say they know that something else was changed? Why are asking Stanford what they changed. Is anybody complaining to NVidia? If not, why not?

Stanford respects their partners and isn't likely to deflect the to blame to NV even if my assumptions are 100% correct. It's better just to say nothing and wait for NV to fix the performances losses encountered in the current version of CUDA.

What we do know is yes it is used to support Kepler and the environment variable was changed viewtopic.php?f=66&t=21571 also we know the cuda optimizations we never implemented viewtopic.php?f=66&t=22197&p=221635#p221635 as far as the nVidia statement goes it is possible, Fermin being faster than Kepler at running Gromacs certainly can not be good for nVidias sells, but I would not think PG would cripple Fermin for nVidia.