Message boards :
News :
Should the fixed credits be higher?
Message board moderation
Author | Message |
---|---|
Send message Joined: 8 Jul 11 Posts: 1341 Credit: 498,424,986 RAC: 578,352 |
I have had a couple people mention that the fixed credits should be higher. I tried to set the credits to be in line with the cobblestone definition, but it's possible I missed the mark. I will wait for feedback before making any changes. Please share your credit opinions. |
Send message Joined: 23 Jan 20 Posts: 3 Credit: 181,944 RAC: 0 |
G'Day Eric, I think they should definitely be higher. The credits seemed so low I've stopped running tasks. LHC@home give approximately 1 per second on my CPU. Milkyway at home give even more 227.51 for tasks that take between 137 and 165 seconds on a GPU. Cheers Ken |
Send message Joined: 16 Jun 20 Posts: 1 Credit: 46,222 RAC: 0 |
i dont do this for credit. its not a contest to me. i run boinc out of the kindness of my heart and to further humanities knowledge. I would do it even if there was no credit system. i will double up on my lhc processes. |
Send message Joined: 19 Aug 11 Posts: 31 Credit: 67,893,366 RAC: 40,662 |
Short answer: Yes, credits are too small here. Longer answer: Try doing come crunching with your own machines on other projects. PrimeGrid is a good baseline, a good place to start. Run your CPUs there against one of their LLR sub-projects, and run your GPUs on their PPS sub-project. That should give you a good benchmark to compare to what you generate against this project. Then consider offering bonus rates for unusual things that you may require for your project, for example large memory requirements, or long running times without checkpoints, etc. Reno, NV Team: SETI.USA |
Send message Joined: 3 Jun 12 Posts: 9 Credit: 3,136,701 RAC: 0 |
I agree with kraDen and Ron and Zombie at the same time :) Many people don't crunch for credit BUT credits are the only "reward" that a boinc project can give to its volunteers. So unless you are a "big, old and well established" boinc project, and can basically "not care too much" about credit, it is still an important incentive to get new crunchers, and keep the one you already have :) I don't like over-crediting for that same purpose, it's not a good philosophy, but under-crunching is worse, IMHO. |
Send message Joined: 7 Oct 11 Posts: 17 Credit: 17,116,951 RAC: 41 |
Short answer: Yes, credits are too small here. I agree 100%!!! Too few credits and people will come tocheck things out and leave, too many credits and you will get the hackers trying to cheat the system. PrimeGrid seems to have found a happy medium by not giving out too many nor too few credits. They also have tasks that can run from 10 minutes to over 100 hundred hours. To me the idea of a project is to attract enough users to get thru the work without attracting so many users that you are constantly spending time and money to keep up with all of their needs, credits can and do affect that. |
Send message Joined: 24 Nov 17 Posts: 1 Credit: 3,713,394 RAC: 0 |
In the past I did not worry about credits per day but recently I started working with Gridcoin in which this project is one of the whitelisted projects. Most of the whitelisted projects are similar in the RAC/Magnitude that develops over time but does also increase with lower user counts. I have a personal quest to get projects to 1M points and I did that with this project but then it fell lower on the list of priorities to get to 5M because the RAC was much lower than others in Gridcoin. As previously suggested, you may have to spend some time running other projects on your machines and put the tasks into a spreadsheet to see what they are averaging at the task level per second or minute. Then see what changes you could make to line up with that. Someone suggested PrimeGrid or there is also Asteroids and Amicable Numbers to compare to. |
Send message Joined: 11 Oct 18 Posts: 12 Credit: 1,013,277 RAC: 0 |
Higher credit. If not higher maybe some sort of bonus for crunchers with high turnover. |
Send message Joined: 8 Jul 11 Posts: 1341 Credit: 498,424,986 RAC: 578,352 |
Thanks all for the comments. I will run some other projects to get an idea of their credits... not sure why I didn't think of that before. |
Send message Joined: 3 Sep 11 Posts: 30 Credit: 9,825,481 RAC: 18,486 |
The short runners (those under 8,000 seconds or 2.2 hours) give good to reasonable credit numbers with 74 awarded credits for each work unit. The longer ones don't give as good, dropping quite low, particularly the ones that run for over 3 hours, of which I get a number. If the work units utilize hard drive storage, heavy load on the CPU or other resources, then a bonus as zombie mentioned should be awarded. If there are not many long ones, then 74 to 100 (about 90 avg) would be alright. If there are a lot of long ones then a more reasonable amount maybe 110 to 130 (120 avg). How you get that to average out I am not sure. Conan |
Send message Joined: 8 Jul 11 Posts: 1341 Credit: 498,424,986 RAC: 578,352 |
I'm still waiting for validation on LHC and Milkyway, but results are in for PrimeGrid: https://www.primegrid.com/results.php?hostid=1009160 Converting to credits/hour this gives the following for the CPU: 174, 122, 622, 643, 140, 173 NumberFields is only giving 52 credits/hour (on average). To be in line with PrimeGrid I would have to increase credits by about a factor of 3. I will wait on the results from Milkyway and LHC before making any final decision. |
Send message Joined: 22 Apr 20 Posts: 1 Credit: 858,951 RAC: 0 |
I am actually satisfied with the credit awards. Assuming the BOINC recommendation is 200 credits per GFLOP/day and that is what you're aiming to award, then I'm averaging more credits than that. My average CPU task runs for about 120 min at a peak of 3.8 GFLOPs. Some tasks are as short as 80 min, and some as long as 330 min. [(3.8 GFLOPS*200 CREDITS / GFLOP / DAY)/ 24 HOURS/DAY]*2 HOURS = 63 credits. So 74 credits seems more than enough for 90-120 min. If I run a task for 5 hours, I feel shorted a little sure, but I feel that's more than made up for by the tasks completed in 90 min. So my vote is that it's fine how it is. |
Send message Joined: 8 Jul 11 Posts: 1341 Credit: 498,424,986 RAC: 578,352 |
G'Day Eric, Did you mean LHC gave 1 credit per minute on the CPU? Because that's what I'm seeing, almost 60 per hour. This is in line with NumberFields. |
Send message Joined: 8 Jul 11 Posts: 1341 Credit: 498,424,986 RAC: 578,352 |
So LHC awards about the same credits as NumberFields, at least on the CPU. Milkyway was inconclusive. I had 1 task which gave low credits (~ 50/hr). I compared with other hosts that were slower but got over twice the credit, so I'm not sure what's up with that. I also tested a GPU task - it had a huge payout. Maybe they have an extremely efficient GPU app which can justify such a payout. So I have decided to double the credits. This will get it closer to the PrimeGrid credits and I think this will make most people happier. If necessary I can always put it back later. |
Send message Joined: 26 Jun 13 Posts: 11 Credit: 8,735,592 RAC: 0 |
If you want to compare credits/hour with other projects you may need to take how well the programs utilize CPU into account. Physic related projects sometimes yields lower credits because they are not pushing CPUs into limit like math program does, especially Primegrid -- prime detection algorithms are well developed and well optimized for performance. As a long time user here I guess I will just crunch some over the time regardless of the credits... |
Send message Joined: 15 Jun 15 Posts: 2 Credit: 1,194,710 RAC: 0 |
My GTX1060 daily gets about: 7k on NumberFields (CreditNew on May) 560k on PrimeGrid PPS-SV 150k on PrimeGrid AP27 260k on SRBase TF 270k on AmicableNumbers Besides, CPU subprojects pay my i5-4590 20-30k on PrimeGrid/SRBase, about 5k on NumberFields. So, yes, you could increase fixed credits by a factor 3 or 4, at least. LLR apps are very efficient though. Maybe you are right to only double fixed credits. |
Send message Joined: 26 Jun 13 Posts: 11 Credit: 8,735,592 RAC: 0 |
It is addressed that the CPU algorithm is a lot more efficient than the GPU one (soon after the GPU task was introduced) so you do not see the GPU credit being far superior than the CPU one (comparing to all those prime number projects). |
Send message Joined: 8 Jul 11 Posts: 1341 Credit: 498,424,986 RAC: 578,352 |
It is addressed that the CPU algorithm is a lot more efficient than the GPU one (soon after the GPU task was introduced) so you do not see the GPU credit being far superior than the CPU one (comparing to all those prime number projects). Technically, the CPU and GPU use the same algorithm. It's how it's implemented that needs improvement (i.e. the GPU threads need to stay in lockstep). Note even with the inefficiency, the GPU is still about 10 times faster than the CPU (depending on the GPU and CPU of course). |
Send message Joined: 14 Mar 12 Posts: 2 Credit: 10,071,839 RAC: 0 |
1. Set the credit to what you want. If you want to get more results increase the points. If you want to get many more results increase the points many times. If you want to still be doing this when you are 90 years old then continue as you are. If you want to see things sooner increase the points. 2. If you want people to use their GPU on your project do not keep the points the same for CPU and GPU. A newer GPU needs to get at least 200,000 points per day. The projects that are doing well are giving more. 3. Credit Screw gives about 0.4% to 0.8% of the number of seconds the WU takes to finish. I just ran 11 of your CPU WUs and got about 1.4%, 61 points divided by 4,200 seconds to finish the WU. 4. Years ago, when I ran Number Fields up to 8,000,000, I was getting at least 5%! Thanks for keeping your project going and good luck! |
Send message Joined: 19 Aug 11 Posts: 31 Credit: 67,893,366 RAC: 40,662 |
So I have decided to double the credits. This will get it closer to the PrimeGrid credits and I think this will make most people happier. If necessary I can always put it back later. Has the change in credits been implemented now? Reno, NV Team: SETI.USA |