F@H PPD, can this be correct?

[Xpl1c1t]

Before highschool I had my Athlon XP 2100+ running F@H for years. During highschool, an Athlon 64 3200+ for ~4 years. Back then, I was even installing F@H clients on the unadministrated P4 machines in the school library and linking them to my account!

In total I racked up ~220 WU and about 28,000 points.

Since them glory days I've basically had shit hardware for years and had decided against dissipating heat for a low return.

Now I have a marginal GTX 750 Ti and have more than doubled my points in less than one day... The graphics card cranks ~70K PPD, and fact right now i'm working on a single WU which awards 23.6K.

Can someone elucidate how the points gained from all those years and years of cranking single threaded F@H WUs can be bested by a cheap card in one day? Seems like inflation to me, and not only in hardware performance...

 

[waderunner]

What CPU are you using? I believe F@H awards points based on the work that a baseline "benchmark" machine or GPU can accomplish. Depending on what hardware you used back in high school, relative to the benchmark back then, and the hardware you use now, relative to the current benchmark hardware, you may be accumulating points based on the use of stronger hardware, relative to the benchmark hardware of the time.

Also, I suppose point inflation might also be at play (maybe they do simply award more points now), though I don't know anything specific about that.

 

[Xpl1c1t]

The CPU is a Xeon X3430 which is essentially a i5 750 Lynnfield quad.

 

[Xpl1c1t]

And here is my statistics page at the Stanford site.
http://fah-web2.stanford.edu/cgi-bin/main.py?qtype=userpage&username=xpl1c1t&teamnum=33

 

[jfb9301]

It probably is normal, I am not as familiar with the lower end of the GTX7xx series. 70K is on target with what I am familiar with. A GTX570 will get 25-40KPPD, and a GTX970 can get 290-340KPPD. GPU points have gotten completely out of control. There were many arguments that the points are inflated, and that GPUs produce less science than CPU. As routine with Stanford all this has fallen on deaf ears, and has a lot to do with why so many giants of our team have stopped folding. As to your question of why, I have watched the history of it all unfolding, and could not even begin to explain.

 

[Nathan_P]

70k PPD is perfectly normal for a 750ti, Most of the points these days come from the Quick Return Bonus. This encourages people to run their machines 24/7 and get more points in return. It was first applied to SMP projects about 5 years ago and has since been applied to the gpu WU.

The theory is that equal work get equal points but as jfb says GPU wu are nothing like the CPU wu in terms of size of complexity

 

[bman212121]

The thing to keep in mind is that technology gets faster. You were folding on a single core CPU of modest clock speed that was released almost 12 years ago. The GPU you would have paired with that cpu probably would have been a Ti4200 or an FX5600 as the 6 series wasn't even out yet. (It could have actually been a red card by the name of the 9700 PRO) I'm not even sure how many times faster the 750 Ti is than one of those cards, but just throwing out something like 50x or 100x wouldn't be unimaginable.

On top of the generation leaps you also need to take into account the floating point operations. A GPU is designed for floating point where a CPU is more for arithmetic. The top CPU an i7 4970K can do about 100Gflops, where the 750 Ti is rated for about 1,400 Gflops. Folding tends to like floating point operations so that plays well with the GPU.

So yes I do agree that the GPU points are a bit overstated, but at the same time it's not unjustified to say that the work a new GPU can do in a short amount of time is equal to the same amount of work that an old CPU did during all of it's time folding. The shiny new GPU could really be doing the work 1,000 times faster so 1 day of a new GPU is the same amount of computation as 3 years on that old CPU.

The theory is that equal work get equal points but as jfb says GPU wu are nothing like the CPU wu in terms of size of complexity

Yea that's where you can argue the value of the work versus the amount of work done. The amount of operations a second is still much higher on the GPU, but in terms of what you get from the extra work might not be as tangible.

 

[Gilthanis]

And just think how crazy it would be if someone foot the bill to get an ASIC designed to run this work. They would completely blow the GPu's out of the water. But the cost to design, manufacture, and then try to reclaim the investment would just be too tough. So yeah, utilizing more specialized hardware has been the trend at many DC projects. I personally don't see how anyone can put a value on the "science done". However, you can imagine that the faster it is done or the more work units done will certainly be easy to calculate.

Think of it this way. A cure for cancer would be a wonderful thing. Finding cheap, good quality, and deployable water filter/treatment setups would be wonderful. The clean water projects would save millions more lives and improve living conditions far beyond a cure for cancer. (and lets face it, we are really talking about a cure for many cancers. Not just any cancer) Which science has more value? Now, take that rationale and put it to a smaller comparison like at FAH. How do you truly compare the sciences they are doing? Who tells you what science has more value? And what do the point payouts tell you? It is hard to look back at the points you made when you started and not feel like it was a huge waste of electricity. On the same note, things will only continue in that direction. What you crunch today will be blown away by DC gear 5 years from now. The difference is that we could find a cure today instead of 5 years from now. It would be dumb to not run it on the gear we have available. I hear a lot of guys say that it isn't worth running CPU work units on projects that have GPU's. Those people aren't looking at it from the science value but just points. If it was worth running on CPU before someone coded GPU, then it should still be worth running on CPU. What were you an idiot for running your CPU's? The answer is no. Its value is the same now as it was then. You can just crank out more.