Message boards :
Number crunching :
Why do tasks with more GFLOPS required get unequal credits?
Message board moderation
Author | Message |
---|---|
NavyDT Send message Joined: 16 Jun 15 Posts: 1 Credit: 1,498,584 RAC: 0 |
Hello all, i had a quick question. im using a NVIDIA GeForce GTX 780 and a Radeon HD 6790 with the lunatics optimized apps. Why is it that the nvidia gets assigned tasks that are 12,000 gflops long and the ATI is getting tasks that are 120,000 gflops long. Further more, why does a task that takes 10 times more work get nearly the same credit reward? maybe some one can shed some light on this, i will post an example below. .......................................................run time..... CPU time..... Credits Anonymous platform (NVIDIA GPU)..934.12..... 275.50..... 74.42 Anonymous platform (ATI GPU) ......3,939.56..... 372.19..... 82.32 you can see that the ATI card took 3 times as long, but only got 8 more credits, that does not really make sense to me. Thnaks for taking the time to consider my post, Very Respectfully, Bryan F. |
Donald L. Johnson Send message Joined: 5 Aug 02 Posts: 8240 Credit: 14,654,533 RAC: 20 |
Hello all, Do a Keyword search on "CreditNew". There have been extensive discussions on the credit-granting system and how inconsistent it is. One aspect of the CreditNew system that screws up comparisons such as you are trying to do is this: Credit for any given Work Unit is based not just on your machine, but also the computer you were partnered with. So every Work Unit is different. There is NO single credit/flops or credit/runtime standard you can apply. Donald Infernal Optimist / Submariner, retired |
Cosmic_Ocean Send message Joined: 23 Dec 00 Posts: 3027 Credit: 13,516,867 RAC: 13 |
Welcome to the forum. Every task (especially the 'setiathome' tasks, which are known as "Multi-beam" or MB for short) is going to have a variation in how much processing is actually done/needed. For MB tasks, that largely depends on the 'true angle range' which can often be determined if you look in the stderr section for the actual taskID for that WU. For Astropulse (AP), the percentage of the task that was blanked has a lot to do with how much processing time is needed, and therefore, how many FLOPs were needed to go through it. All of that being said, there's been a joke here on the forum for years that "CreditNew" is referred to as a Random Number Generator. There have been a few people who have actually walked through parts of, or the entire code for it and have a good read on why it works the way it does, and the consensus from them is that it could be pretty robust, but a few of the factors involved need some adjustment and fine-tuning to get it there. When that will actually happen... *shrug* Your guess is as good as mine. Linux laptop: record uptime: 1511d 20h 19m (ended due to the power brick giving-up) |
rob smith Send message Joined: 7 Mar 03 Posts: 22216 Credit: 416,307,556 RAC: 380 |
As Donald has said CreditNew is to blame, crudely it looks at the matching pair of results for a task and takes the lower CPU-equivalent flops for the task as being the "right answer" and Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.