Cross-project credit equality took a giant step forward with CreditNew; it makes it possible, but it would still take cross-project adaption, and that cannot really happen until there is GPU to CPU calibration. At present there probably isn't enough projects to make this step solid, but I still expect a method to turn up this year.
Even then there is still questions about app performance (OpenCL vs ATI(14) vs CUDA) and considerations such as GPU Utilization, power usage, and non-complicated GPU projects (relatively simple maths) vs complex apps and mixed CPU/GPU apps.

The badges will help keep things real; give a different perspective, especially the project badges which will show your relative contribution to each completed project. Each completed project will of course have resulted in publications, presentations, displays and will often have facilitated both new GPUGrid research and other research projects...

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

By my point of wiev; GPUGRID is not in a credit war. Compare to other project (GTX560Ti factory OC 900 MHz)
1. PG PSA tpsieving - theoretical RAC ~ 800 k, 4-6% CPU consumption, 99% GPU load
2. DistrRTGen theoretical RAC ~ 500 k (process tammed to realtime resp. -20), 98 % CPU consumption, 99% GPU load
3. PG PPSE Sieving theoretical RAC ~ 250 k (november 2011), 4-6 % CPU consumption, 99% GPU load
4. PG GWC Sieving theoretical RAC ~ 125 k (november 2011), 4-6% CPU consumption, 99% GPU load
5. GPUGRID longrun tasks, SwanSync=0, + 50% time bonus <24 hours
-- theoretical RAC max 105 k with NATHAN_CB/TONI_FAAEAC tasks only, 100% CPU consumption, 99% GPU load
-- theoretical RAC < 95 k with other longrun tasks, 100% CPU consumption, 86% GPU load
Where can you see the overcrediting of GPUGRID?
Do not compare to Einstein or Seti (apps of both projets runs fine on CC 1.1 GPU, not on CC 1.3+ GPU). To crunch that projects by CC 1.3+ GPU (without using app_info) is not a good idea. From the other side - try to crunch NATHAN_CB or TONI_FAAEAC by a 9600GT (suitable GPU for Seti, Albert & Einstein).....have you tried it? Now you can see overcredited is Seti, Albert & Einstein, as any credit > zero credit.
I did not compare GPUGRID to ATI GPU efficient projects (MW, Collatz, Moo!), as i mean that it would be waste of eletricity power and GPU cycles to crunch these project by nV GPU. Only for info (much cheaper and low power consumption HD 4770)
1. Moo! - theoretical RAC ~ 86 k, 100% CPU consumption, 99% GPU load
2. MW - theoretical RAC ~ 68 k, 5% CPU consumption, 99% GPU load
3. Collatz - theoretical RAC ~ 65 k, 1% CPU consumption, 99% GPU load
1-4. POEM++ ATIOpenCL, using CreditNew that looks like a joke, for near the same run time (about one hour) was granted credit between 2,5 k - 10 k.
By the way - DistrRTgen used the CrediNew for a time, but has been left as unusable for app of the project.

You put a bit of work into that analysis, and it's sound.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Yes, very useful. Thanks.

Just had a GIANNI wu completed and returned within 24 hrs but have been cedited with 45000

There seems to be a very large difference in the time taken on the "GIANNI" wu's on the same machine. As low as 17hrs and up to double that. No changes or use on that machine.

Windows 7 Ultimate, i7 920 running @ 3002MHz in an Asus P6T motherboard running a cool GTX 285 (60-65C) in prefer maximum performance mode and 6G memory.
Any ideas as to why there should be such a time range?

Any ideas as to why there should be such a time range?

Some. Compare time per step of these tasks. Maybe admins will answer you better as your hosts are hidden.

Hosts should now be showing

Hosts are now showing.

Hosts are now showing.

Not anymore. Don't understand why one would have any desire to hide them anyhow.

That aside I have noticed that different work units in this project give vastly different credit even for a given run time. Does the amount of computation done per second actually vary that much from one task type to another? (b/c gpu utilization seems to be fairly similar) Don't know if i've seen much variation in credit within a specific task name however I suppose I've seen time variations. Perhaps the calculatons can vary substantially within a task type?

BTW I haven't seen any task type use more than ~90% gpu often only using ~80%. Is this normal for a 460?

---

Thanks for the response but I think it was simply the cards downclocking ... now sorted.

By my point of wiev; GPUGRID is not in a credit war. Compare to other project (GTX560Ti factory OC 900 MHz)
1. PG PSA tpsieving - theoretical RAC ~ 800 k, 4-6% CPU consumption, 99% GPU load
2. DistrRTGen theoretical RAC ~ 500 k (process tammed to realtime resp. -20), 98 % CPU consumption, 99% GPU load
3. PG PPSE Sieving theoretical RAC ~ 250 k (november 2011), 4-6 % CPU consumption, 99% GPU load
4. PG GWC Sieving theoretical RAC ~ 125 k (november 2011), 4-6% CPU consumption, 99% GPU load
5. GPUGRID longrun tasks, SwanSync=0, + 50% time bonus <24 hours
-- theoretical RAC max 105 k with NATHAN_CB/TONI_FAAEAC tasks only, 100% CPU consumption, 99% GPU load
-- theoretical RAC < 95 k with other longrun tasks, 100% CPU consumption, 86% GPU load
Where can you see the overcrediting of GPUGRID?
Do not compare to Einstein or Seti (apps of both projets runs fine on CC 1.1 GPU, not on CC 1.3+ GPU). To crunch that projects by CC 1.3+ GPU (without using app_info) is not a good idea. From the other side - try to crunch NATHAN_CB or TONI_FAAEAC by a 9600GT (suitable GPU for Seti, Albert & Einstein).....have you tried it? Now you can see overcredited is Seti, Albert & Einstein, as any credit > zero credit.
I did not compare GPUGRID to ATI GPU efficient projects (MW, Collatz, Moo!), as i mean that it would be waste of eletricity power and GPU cycles to crunch these project by nV GPU. Only for info (much cheaper and low power consumption HD 4770)
1. Moo! - theoretical RAC ~ 86 k, 100% CPU consumption, 99% GPU load
2. MW - theoretical RAC ~ 68 k, 5% CPU consumption, 99% GPU load
3. Collatz - theoretical RAC ~ 65 k, 1% CPU consumption, 99% GPU load
1-4. POEM++ ATIOpenCL, using CreditNew that looks like a joke, for near the same run time (about one hour) was granted credit between 2,5 k - 10 k.
By the way - DistrRTgen used the CrediNew for a time, but has been left as unusable for app of the project.

Just a question on this one. you refer to the GPuGrid longruns with 100% cpu consumption.
Are you referring there to not having any other tasks running on your computer at all apart from running the GpuGrid task?

I ask because I currently run an i7-2600k@4.5 under win7 64bit that crunches 8 WCG tasks and 2 gpugrid tasks (560ti2gb @ 860Mhz and a 460 1gb @800Mhz) and have no trouble completing Nathans in around 32k for the 460 and 26k for the 560.

both normally return 35811points, though I've just noticed a couple of 31500. not sure what that's about.

Are you referring there to not having any other tasks running on your computer at all apart from running the GpuGrid task?

I am running 3 CPU tasks and 1 GPUGRID task (4CPU Xeon, set "use at most 99% of the processors").

both normally return 35811points, though I've just noticed a couple of 31500. not sure what that's about.

....NATHAN_CB1... gives 35,811 credits, ...NATHAN_FA5...gives 31,500 credits, both with 50% time bonus.

ah. ok thanks.

Im currently running my 470 at 800 core and I have swan_sync=0 and in boinc manager settings I have set on multiprocessor systems use at most 75%.
I got 3 primegrid pps llravx running and they take all other three cores and usually all long runs hog completely the spare core I give them. Although the task says its running 0.44 cpu + 1.0 gpu.
Cpu is 2500K@4.5Ghz

All tasks of all types and size.

So how do we know if the credits are doubled or not? People here are talking about getting up to 67,500 credits for one units yet I am getting a fairly consistent 35,000 for per unit. Here is one of my units:
http://www.gpugrid.net/result.php?resultid=5048734

I KNOW there are MANY reasons for different credits being awarded to different people, I just don't know how to tell if I am getting double credits for returning them on time.

There is no credit breakdown, or formula, just "Credit 35,811.00".
The old site gave more info; claimed credit and awarded credit.

So you would need to accept the bonus system is 50% for <24h return, 25% between 24h and 48h. Then you can see from the result that it returned inside 24h so would have been granted full credit.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help