NumberFields now updated to v3 app

Good at maths? Then why not figure on helping these projects - Gerasim, GPUGrid, Moo! Wrapper, NFS, NumberFields, ODLK, PrimeGrid, Rakesearch, SRBase
Post Reply
UBT - Timbo
UBT Forum Admin
Posts: 9673
Joined: Mon Mar 13, 2006 12:00 am
Location: NW Midlands
Contact:

NumberFields now updated to v3 app

Post by UBT - Timbo »

Hi all

It looks like NumberFields have developed a new version CPU app, (v3.0) that runs 10x faster.

For now the fixed credits per task has been dispensed with and credit will be issued on a run time basis (for a while).

More info here:

https://numberfields.asu.edu/NumberFiel ... d=366#2360

I am hoping that this might mean that more credits could be earned per task - I'll download some work and see how it goes.

regards
Tim
UBT - Timbo
UBT Forum Admin
Posts: 9673
Joined: Mon Mar 13, 2006 12:00 am
Location: NW Midlands
Contact:

Re: NumberFields now updated to v3 app

Post by UBT - Timbo »

Hi all

I've now submitted just under 300 NumberFields tasks and can report as follows:

1) The earlier tasks (v2.xx) ran for around 4 hours (abt 240 mins) (in my case and depending on CPU speed) and earned approx 370 credits per completed task. That's about 1.5 credits per minute.

2) The new tasks (v3.0) are supposed to run for around 20 mins (so, about 10x faster, as they thought) and are now awarded credits based on the actual CPU time taken. I have 2 PCs running NF now - an older Quad core (using 3 cores, with one reserved for GPU) and a 16-core Xeon (using 15 cores + 1 for GPU).

The Quad core takes between 700-2,000 seconds and earns about 0.667 credits per minute, while the Xeon takes between 1,000 and 2,800 seconds but earns 0.835 credits per minute.

So, it seems less credit is earned per task but the upside is that every task takes far less time.

This could have a negative impact on the project as their servers will be under greater strain having to send and receive more tasks in less time.

regards
Tim
Woodles
UBT Contributor
Posts: 11757
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: NumberFields now updated to v3 app

Post by Woodles »

I have issues with
1. Deploy new cpu executables. Since it's 10x faster, I will need to drop the credit by a factor of 10. (Credits/hour will remain the same for the cpu but will obviously drop for the GPU)
Why? Ten times the work is being done but there's no extra reward for the user?

Later on in the discussion Michael says that he's going back to CreditNew,
So there has been rampant abuse again by GPUs, so the credit system is in need of modification again.

I feel like history is repeating itself here, but I am going back to CreditNew.
That's always a good step in my book! :doh: :(

The upshot seems to be that if you use a GPU on Numberfields then you'll now use extra power to do ten times as much work as previously but get less recognition.

Mark
UBT - Timbo
UBT Forum Admin
Posts: 9673
Joined: Mon Mar 13, 2006 12:00 am
Location: NW Midlands
Contact:

Re: NumberFields now updated to v3 app

Post by UBT - Timbo »

Hi Mark

I can never understand the reasoning why some projects offer significantly fewer credits than others.

Surely, as a project, you want as many successfully completed tasks to be returned and (one assumes) in as short a period of time as possible. That then gives the project real data to work on.

And offering a high level of credits should ensure a lot of members help the project, because the payback will be worthwhile.

I'm not sure who now controls the BOINC credit system - one assumes that David Anderson has stepped back now from the development side of the project and whoever has taken over has just kept things ticking over.

But as we know, with Bitcoin Utopia and Collatz, (who both award significantly higher credit per tasks than other projects) the credits per task are set by the individual project...so, it should be very simple for a project to give members as many credits as it wants.

So, cutting back on the credits, just because they've used a different algorithm within the application, just seems to be a slap in the face.

regards
Tim
Woodles
UBT Contributor
Posts: 11757
Joined: Thu Dec 20, 2007 12:00 am
Location: Cambridgeshire

Re: NumberFields now updated to v3 app

Post by Woodles »

Hi Tim,

I don't feel that every project should align perfectly with every other project, that's impossible, but once you've started a project and set a credit level for the work done it should be kept reasonably constant.

What I was querying was the fact that Numberfields had been giving a certain credit value for each completed parcel of work and were now going to drop that to 10% of the initial value for future identical parcels of work. The point isn't to keep credits per hour constant, it's to keep credits per work unit constant. At present, a new user can join the project but would have to do ten times as much work as an existing one to catch them.

The irony is that Collatz and BU probably match(ed) the initial definition of what a credit was worth better than some other projects, they just produced a lot more useful work than other less efficient projects. BU in particular went through a long drawn out process with DA about their credits distribution and even he stated that it was the correct value for the work done.

I can't see me doing any more Numberfields work unless it's for a FB sprint and certainly not the GPU work (it helps that Nick hasn't updated his GDPR setting so I appear to be the team #1 :D )

Mark
ChelseaOilman
TeAm AnandTech team member
Posts: 807
Joined: Sat May 02, 2020 6:14 pm
Location: Texas/Colorado

Re: NumberFields now updated to v3 app

Post by ChelseaOilman »

https://numberfields.asu.edu/NumberFiel ... d=458#2776
I have replaced CreditNew (or CreditScrew as some like to call it) with a fixed credit system.

The fixed value is based on the average runtime from many samples from a single machine in a well controlled environment (one of my own computers). Obviously, the fixed value will change with each batch, since each batch has a different average runtime.

This should be a pretty fair system, especially after a host has processed enough WUs for the average to kick in. The main downside is the potential for cherry pickers - hopefully there won't be too many of them.
Post Reply