My imagination ~~~ time dilation ~~~

[orion456]

So is it my imagination or are the times necessary to complete WUs getting longer but the points staying the same?

[Adak]

I'm getting some project 30xx WU's on my SMP rig, and they certainly are the slowest I've ever folded on my C2D. Ppd drops about 300, when they're on board.

[Shiggity]

Not your imagination, the 3000 WU's own me too and don't give as many points as they should imo. They should be 5-10% more because everyone is saying they take longer.

[ChasR]

Unless you have a machine in the same family as the benchmark machine, dual Woodcrest @ 2.33 GHz, running Linux, you can't really say the whether the WUs are running slower than they should. Perhaps the p2605s and p2653s run faster on your machine than on the benchmark machine. A WU like p3065 that runs on your machine at the same speed as the benchmark machine would be slower than p2605/2653 even though benchmarked perfectly. I checked the p3065 production on a Q6600 and found running a single instance it matched a single instance of p2653 pretty closely. Running two instances of each, p2653 is much faster.
Pande Group's Peter Kasson has been working very hard to bring cpu utilization of the SMP client as close to 100% on a single instance as possible on quads. You can imagine the effect his success will have on DC rigs and quads running two instances.

[Shiggity]

It's not just that, it's the fact that they take so long to complete with nonstop folding. I usually complete a 2653 after 16-18 hours, the 3xxx units were taking me over a day (that's almost nonstop, with me doing very light things, does not include gaming time). Any WU that takes that long should have a premium on it imo. So the people who only care about PPD aren't mad and it makes people more willing to crunch the larger sized WU's.

Yeah it sounds selfish to care about PPD, but everyone does to some degree. But I mean c'mon, if it takes me 24+ hours to complete one big 3xxx, it should have some kind of premium.

my 2 cents

[WarriorII]

The highest point WU's were very desirable.

People would build Brand Specific machines just for the high points they could get.

Big bucks were spent for the 600 points a WU.

Man ! 2 of those a day ! You were a Big Boy.

And there was absolutely no way an AMD guy could get one!

You HAD to have an Intel to get any.

I still want those back.

[ChasR]

A q6600 @ 3.3 GHz with Linux as the OS, running a single instance makes about 3600 ppd on p306x and it makes about 3650 on p2653. On the benchmark machine they both make 1760 ppd. Why should p306x be valued higher than they are now? p3065 does indeed take longer to fold, but it's worth 2144 points. The SMP WUs were designed for quad core processors. If you aren't running one, You Mileage May Vary and it is unreasonable to expect it not to. Anyway adding to the points value of a WU to accomodate any percieved lack of production on a C2D or AMD rig will provide anyone with a quad a windfall bonus. The realtive performance of the of the different machines can't be changed by adjusting the point values.

[jerseytomato]

Actually it's me causing the time dilations on your WU's :p I'll have to postpone my catching up with you till after the chimp challenge.

[ChasR]

Big bucks were spent for the 600 points a WU.

Man ! 2 of those a day ! You were a Big Boy.

And there was absolutely no way an AMD guy could get one!

You HAD to have an Intel to get any.

I still want those back.

Actually, the QMDs were 450 points and had a peak memory useage of 350 MB. They were Intel only and on the really fast rigs like Joe Camel's
920D @ 4480 MHz made 1170 ppd, running two instances. If you had a P4 with HT and got two of them, your machine became unuseable. I was thrilled at the time to make over 700 ppd on my $5000 Xeon Nocona duallie but could only run one QMD instance because of ram bandwidth bottlenecking. Now days, with points inflation and hardware improvements, that $5000 would buy me 30,000 ppd. The real old timers folded for 1 to 5 points per day. In that light it's hard to fathom why SMP WUs are valued such that they make over 1000 ppd/core on quads and we shouldn't complain about the SMP points being too low.

[WarriorII]

I would sneakernet WU's home from work. :D
Good weeks, I was up to just under 6k (@week)

[orion456]

Actually it's me causing the time dilations on your WU's :p I'll have to postpone my catching up with you till after the chimp challenge.


YAY chimps!! :soda: I'll take all the help I can get! :eek:

Mr. Tomato, you have been squishing at my heels for some time and I'm starting to get that red juice all over. So I to warn you, I still have Linux as a last resort. :confused: God forbid you make me go there!

[deadlysyn]

Actually, the QMDs were 450 points and had a peak memory useage of 350 MB. They were Intel only and on the really fast rigs like Joe Camel's
920D @ 4480 MHz made 1170 ppd, running two instances. If you had a P4 with HT and got two of them, your machine became unuseable. I was thrilled at the time to make over 700 ppd on my $5000 Xeon Nocona duallie but could only run one QMD instance because of ram bandwidth bottlenecking. Now days, with points inflation and hardware improvements, that $5000 would buy me 30,000 ppd. The real old timers folded for 1 to 5 points per day. In that light it's hard to fathom why SMP WUs are valued such that they make over 1000 ppd/core on quads and we shouldn't complain about the SMP points being too low.

Sorry for the rant, but I don't really see how anyone can complain about SMP WU PPD being too low. It seems like it wasn't too long ago that SMP clients did not even exist, and everyone here was using the 5.02 and 5.04 clients, averaging much less PPD out of each rig. The only big daddy WU's of that time that I remember is the Ribo's, each valued at 600 points.

[orion456]

but I don't really see how anyone can complain about SMP WU PPD being too low.

As usual this is one of those relative things. Specifically I have a p4 940 folding and it usually makes over 1000 points per day but recently I felt that number was going down. Today it got a 3064 and it is making only 650 ppd which is significantly down from 1000; just seemed strange to have point deflation.

[ChasR]

Ribos (600 pointers), QMDs (450 pointers), and the best by far for AMD rigs the Tets (364 pointers). My FX-53 @ 2.54 made 340 ppd on the ribos and 530 ppd on the Tets. The p4 640 at 3.76 GHz made 560 ppd on the QMDs. Those were the days! When million point milestones took years instead of weeks.

[ChasR]

It's not points deflation, it's just less points inflation.:)

[muddocktor]

It's not points deflation, it's just less points inflation.:)

Yeah, you got that right, ChasR. When I first started folding the big point wu's were worth 5 points and took almost a day to finish with a fast Tbird machine. :eek:

[deadlysyn]

Ribos (600 pointers), QMDs (450 pointers), and the best by far for AMD rigs the Tets (364 pointers). My FX-53 @ 2.54 made 340 ppd on the ribos and 530 ppd on the Tets. The p4 640 at 3.76 GHz made 560 ppd on the QMDs. Those were the days! When million point milestones took years instead of weeks.

And then they would occasionally throw the Patty Melts in to really mess with us.:p I am still folding with an old trusty P4 530J, just got a 3907 getting around 503 PPD.:eek:

Yeah, you got that right, ChasR. When I first started folding the big point wu's were worth 5 points and took almost a day to finish with a fast Tbird machine. :eek:

If I was around folding back then, I would be severely :mad::temper::argue: if someone was complaining about getting 200 PPD less because the WU mix changed. To me the points mean nothing, just something to keep the excitement. I only fold for the science.