Folding Forum

This is only regarding SMP and GPU-Clients.

I have been folding for about 1 year and have spend quite some time reading this forum about people happy/unhappy about the PPD's they are getting, heat-problems and so on...... I'm folding 100% for the science (family reasons) and see the point's as a nice way to compete. but I also understand the people that are here only for the points. I have a suggestion for this and would be happy to get some comments/feed-back.

Stanford know we are not living in a black/white world (Intel/AMD, ATI/nVidia, Small/Medium/Large/SuperSize Wu's & Stabil/Beta WU/Core's), so please let the donors see this by adding more options to the client/cores/WU's. I know the following suggestions will give Stanford a bit extra work, but I think it's in the best interest of the science.

1) Build a more advanced client that have advanced detection and options, so it can detect the donors hardware and make 2) possible.

2) Build a few more benchmarking-machines with different configurations (Intel & AMD with ATI & nVidia cards inside) and benchmark the WU's on fx. 4 different machines and 4 different graphic-cards (eg. Intel E7200/Q9300 & AMD Athlon64 X2-4800/Phenom 9850 with nVidia 9600GSO/260-192 & AMD HD 3850/HD4850) - Then rank the machines on how efficient they do a WU and then give the donors the WU based on the donors hardware. If there are no more "1-ranking" WU's to a specific donor (most efficient hardware to do the WU), then give a "2-ranking", "3-ranking" or worst case a "4-ranking WU - This way science win every time, Stanford have a interest in having WU's ready that fully uses the donors hardware (more science) and the donors have a interest in newer hardware (more science/PPS´s).

3) Change the Point-system so fast returns of a WU get a bonus
a) If a WU is returned within 200% of the time it takes the benchmark-machine to complete the WU, the donor get 100% of the point.
b) If a WU is returned within 100% of the time it takes the benchmark-machine to complete the WU, the donor get 150% of the point.
c) If a WU is returned within 50% of the time it takes the benchmark-machine to complete the WU, the donor get 200% of the point.

4) Change the Point-system so non-stabile beta cores/WU's get a bonus (fx. 33% extra in basis-points), because new cores and/or WU's might overheat, give EUE's, and in general need more "babysitting"..... - Any one accepting non-stabile beta cores/WU's also accept EUE's, overheating, "babysitting"... Benchmark-machines always run on stock-spec. in a 20C/68F room, so if the donors machine is in a warm place and have heat-problems, like I might have (living in south Spain), don't complain to Stanford.

I hope we can discuss like adults in this tread and accept that everybody might have another view, so please be open to other opinions.

Excellent idea and presentation.

Number three will put an end to the excuses and rationalizations of those contributors who are running their clients in conflict with the policy of The Pande Group.

All good ideas, much of which is already in progress...

1. Already in the works. The v6 client has added hardware detection code. However, some back end server coding is still need to take advantage of this.

2. Doesn't work, and only changes the flame war from how much PPD to waring over which hardware to use. And even if Stanford could afford the extra hardware, extra support staff to maintain it, and new researchers to do all the extra benchmarking work, we would still be right back where we started arguing about PPD, but instead of just debating about one benchmark, we'd be watching flame wars about setting the PPD on 4 systems instead. Think this one through a bit more, and it should become more clear how this doesn't help, and could actually hurt. However, the second part of the suggestion is already in use to a degree, but could use additional work. IIRC, they hired a dedicated CS student to update a lot of the server code to make this specifically work better, amoung other things (like #1).

3. PG already announced this change was coming many months ago. About how a future points system revision might help in this regards. Still waiting, and without more detail than just the announcement. Oh well.

4. This has already been done in the past as well, on a case by case basis. And may happen again. However, predicting EUEs, let along handling them well, is a difficult task. I doubt anything in the way of a full time automatic feature like this has a workable solution. The return on investment isn't there.

Keep the ideas coming...