Quo vadis QRB

Moderators: Site Moderators, FAHC Science Team

Post Reply
foldinghomealone2
Posts: 148
Joined: Sun Jul 30, 2017 8:40 pm

Quo vadis QRB

Post by foldinghomealone2 »

There have been many discussions about the QRB, its advantages and disadvantages.
I was always in favor for the current QRB, but nowadays the current calculation of the QRB seems a bit ridiculous.

Due to the current calculation method of QRB, when processed time for a WU is reduced to 50% (which might occur with anticipated Ampere compared to Pascal GPUs) the reward is almost tripled (factor is 2,83 to be more accurate).

Need some real examples?
GTX 980 Ti and 970 compared to a 2080 Ti (same Project P11711). - Data taken from our own HFM readings.

980 Ti needs 1,96x longer to process a WU but a 2080 ti gets 2,7x more PPD.
Our previous workhorse, the 970 needs 3,4x longer to process a WU but a 2080 ti gets 6,2x more PPD.

I would prefer a more linear calculation.
There are a few questions:
- why is the k-factor not changed? It should be lowered with each new project to stop the inflation of PPD
- with TPF of around 1minute, why there is still a deadline of 10 days?

I understand that the QRB is important for investments in new and much faster hardware but like in the real world, (high) inflation of PPD is no good in the long term.
Apart from the problem that old rewards are really meaningless compared to current rewards (which offends and offended many veteran folders), one question is raised:
Why should I invest now in new hardware when I can get exponential more points when I wait just one more GPU generation (probably with even less power draw)?
foldy
Posts: 2061
Joined: Sat Dec 01, 2012 3:43 pm
Hardware configuration: Folding@Home Client 7.6.13 (1 GPU slots)
Windows 7 64bit
Intel Core i5 2500k@4Ghz
Nvidia gtx 1080ti driver 441

Re: Quo vadis QRB

Post by foldy »

foldinghomealone2 wrote: Why should I invest now in new hardware when I can get exponential more points when I wait just one more GPU generation (probably with even less power draw)?
If you wait 5 more years then GPUs are even more faster. But if you want to get more points and accelerate FAH now then buy GPU now.

Maybe a good way is to buy GPU now for FAH and sell it again in one year when new GPU generation is released. So you get more FAH points now and still in future can accelerate again with then newest GPU. There are gamers out there which already wait to get your current GPU reselled. Because gamers often do not need the newest GPU with less power draw or efficiency but only a fast GPU. While FAH users want an efficient GPU because it will run 24/7.

But I like the question: "Why is QRB not linear?" Half TPF = Double PPD?
foldinghomealone2
Posts: 148
Joined: Sun Jul 30, 2017 8:40 pm

Re: Quo vadis QRB

Post by foldinghomealone2 »

I'm aware of the dilemma, that if you don't start walking you will never get somewhere.
But currently this system is just unfair to the results of veteran folders.
Veteran folders invested so much money and effort to folding that their 'heritage' shouldn't be destroyed by over-inflation of PPD.
foldy wrote:But I like the question: "Why is QRB not linear?" Half TPF = Double PPD?
Yes, something like this.

Or as a second idea, with a capped PPD. Maybe to 2Mio PPD.
So maybe a GPU xyz is the fasted GPU available then you get 2Mio PPD when folding with it.
Also, double TPF means half PPD.
This would avoid the inflation of PPD.

-

Top 450 folders currently make more than 1Mio PPD (per day) and more than 550 folders have more than 1bn points.
They aren't normal gamers that use their gaming PCs for folding.
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: Quo vadis QRB

Post by bruce »

Every project is benchmarked with a goal of avoiding unnecessary points inflation EXCEPT due to hardware improvements. You are suggesting that PPD increases specifically due to hardware improvements should also be limited in some way.

As an "old-timer" I'm well aware of the fact that early projects earned me a total of one or two points and those points don't really insignificant when compared to what I can earn with a modern GPU. (I accumulated a lot of points which are essentially worthless today.

The first question I'd like to interject is what about older GPUs (or CPUs) which are still folding? Should the PPD of a GTX750 Ti (a formerly "sweet-spot" GPU from ~5 years ago) start earning less PPD than it did in it's prime? (I also have a LowProfile machine containing a GT710.)

And what about the typical mobile GPU from folks who have a single computer which happens to be a laptop?

Any proposed changes to the points system will stir up a lot of controversy -- mostly from people looking to preserve their self-interest -- so IF changes ever reach the point of potentially being introduced, you're going to have all your ducks in order and FAH has to be ready for the resulting uproar. (I'm not advocating for or against a change ... just warning everybody.)
foldinghomealone2
Posts: 148
Joined: Sun Jul 30, 2017 8:40 pm

Re: Quo vadis QRB

Post by foldinghomealone2 »

I'm totally aware that all changes to the QRB would result in a big discussion and there won't be the perfect solution.
bruce wrote:...avoiding unnecessary points inflation EXCEPT due to hardware improvements...
If you translated your statement into a formular it would be like foldy stated: "Half TPF = Double PPD?"

And I showed before there is points inflation APART from hardware improvements.
You cut TPF in half and you get almost triple PPD. That's a big, unnecessary inflation.

You could easily counteract by reducing k-factor a little bit for each new project to get inflation in control again.

With the current system you will see GPUs in the near future that will make 10Mio PPD.
An this doesn't come because they will be 5 or 6 times faster than a current top notch GPU. It also comes because of the non-linear QRB.
Is this really what everyone wants?

And I think if you change the system slowly and stepwise the uproar won't be that big.
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: Quo vadis QRB

Post by bruce »

Faster is defined as better. Running a single WU twice as fast (half TPF) is scientifically better than running two WUs at the designated TPF although they use about the same amount of resources. This is an ancient policy which was adopted long ago when Intel invented HyperThreading and people could choose between running a single WU on a pair of threads (aka virtual CPUs) or run a pair of WUs, each at about 140% of the TPF. I don't expect FAH to accept your logic or change that policy.
- with TPF of around 1minute, why there is still a deadline of 10 days?
Well, what GPU were you using to get that TPF? There is a very wide variation in GPUs and no matter what you actually have I'll bet there are a lot of people with slower GPUs which would potentially not meet the 7 day timeout, especially if they don't run them 24x7. Should those people be limited to a zero bonus?
foldinghomealone2
Posts: 148
Joined: Sun Jul 30, 2017 8:40 pm

Re: Quo vadis QRB

Post by foldinghomealone2 »

I'm not sure if the influence of folders with slow GPUs is really high. At least measured with PPD.
According to current EOC stats - only active folders, 24/7 average PPD considered:
Top 500 out of 8000 active folders (~>890kPPD) make 62,7% of all points
Top 1000 (~>455kPPD) make 77,9%
Top 1500 (~>259kPPD) make 86,2%

If there are still GPUs used that can't meet the 7 day timeout, then I would guess that those GPUs - while adding some FLOPS to the network - slow down the project massively.

And then the question would be if it wouldn't be better to discontinue support for slow GPUs.
If only some of the discontinued slow-GPU-folders would upgrade to faster GPUs it would even benefit the project
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: Quo vadis QRB

Post by bruce »

That's not going to be a popular idea with Donors that have been running slower hardware and FAH is not going to exclude them from the project on your say-so. A better plan is to find a way to use the slower GPUs productively.
foldinghomealone2
Posts: 148
Joined: Sun Jul 30, 2017 8:40 pm

Re: Quo vadis QRB

Post by foldinghomealone2 »

Like in real life, many necessary and good ideas aren't popular and therefore avoided or suppressed.
And I'm not demanding that FAH changes something on my say-so.
HaloJones
Posts: 920
Joined: Thu Jul 24, 2008 10:16 am

Re: Quo vadis QRB

Post by HaloJones »

Interesting subject because it arguably shows that there is an efficiency gain to be made by having two (or more) divisions of units. There's one division allocated to those who specify that they are 24/7 folders with hardware that meets a particular standard; these units have short deadlines of no more than 1-2 days. Then there is "the rest" who get different units that are limited only to a deadline of ten days.

The units being allocated to the faster folders would never be delayed by a slower folder but the slower folders would still be able to contribute.

It's just having a new "client-type". Since only dedicated folders even know what attributes like that even exist it would have no impact on the bottom 20% of contributors but would benefit the science being produced by the top 1500 producing 86% of the points.
single 1070

Image
Post Reply