Intel's next generation of CPUs actually looks like it might be not just competitive against AMDs current offerings, but possibly even better.

(Jim 1348 said)
We can hope so, but they have some sort of high speed/low speed combination of processors.
That might work for ordinary use, but how about BOINC projects? I don't feel at all comfortable about it until I see some tests.

Some early tests for you for the 'Intel Core i5-12600K CPU Up To 50% Faster Than Ryzen 5 5600X'

https://wccftech.com/intel-core-i5-12600k-cpu-faster-than-ryzen-5-5600x-blows-away-rocket-lake-core-i9-11900k-leaked-benchmarks/

Incentives are nice but I think it's important to make up for lost CPU time in the case where the BOINC scheduler hits a memory limit and reduces the number of active work units on a multi thread CPU hence under-utilising the hardware.

It would pay to check statistics of active computers and typical hardware utilization options for contributors to get a baseline system for project contributors but lets take a system with 8 "cores" as an example of a recent generation mid performance system:
A CPU that can process 8 work-units in parallel it would require 8GB of available RAM to allow full utilization of the CPU @1GB of RAM each. The same configuration could only support two concurrent work units @4GB of RAM each resulting in a CPU utilization as low as 25%.
A system with 2GB of RAM per processing core would still only result in 50% CPU utilisation.

If the project Admins can formulate an average "available GB RAM per processing core" ratio for contributors it could be used to calculate the lost %CPU utilization and a reward could be formulated to compensate for lost processing capability.
Alternatively, the Admins and/or community could nominate a ratio of RAM per processing core that would best suit the majority of contributors and the reward could be calculated using that reference system. eg: If we think 2GB per processing core is reasonable then a 50% credit bonus would be fair to make up for lost computation time.

Of course BOINC scheduling is never that straight forward unless you consider a user that is only crunching for Rosetta and all of the resources are dedicated simultaneously to large memory work units, but, unless we expect contributors to edit config files the worst case should be considered.

Personally I would rather see contributor hardware better utilised than receive bonus credits for poor performance, that may require stricter memory limits on the high RAM work units to reduce the likelihood of reaching memory limits in the first place.

I think you need to allow more than three days for completion.

I can do those, but at the expense of my other projects.

S. Gaber

Before you can grant any credit you need to have tasks available to distribute. Credit is always good you don't want to march or you will only get people that are hunting for credits. Those are my thoughts

---

Have a crunching good day!!

Hello, and sorry to give a late reply but this quesion is interesting for me, as I have the duty of maintaining my reaserch group omputing plateform for years and always have to justify our purchase to advisors or others.
In my oppinion, the extra credit for should be determind by the marginal cost of adding an extra memory compared with the marginal cost of adding some extra cpu costs.

The cost of CPU is estimated by comparing the computing power of popular CPU models with its amazon.com price, I here select 7 new models and think the cost of some extra cpu power are 0.46GFLOPS per dollar (1)

The cost of extra memory is estimated by the $40 price of ddr5 16g memory, so about 0.4GB per dolar.

A naive model is for every extra GB memory used, 1.15GFLOPS should be added to the total credit. A neglageble amount.

But personally I think most people will not have so many extra memory, according to steam, most popular configuration are 16 G, 6 cores, so only 2.67 GB/CORE on average. Having a core running 4 GB means a 6 core machine can only ran 3 (considering some memory reserved for system). So if consider an average system a times 2 compensation should be given to those who run large memory model.

So my personal suggestion is to compromise for a 1.5 times, simply taking the average of those 2 methods.

(1) considering cpu model i5 14500, i7 14700kf, i9 14900kf, ryzen 5 5600g, ryzen 7 9700x, ryzen 9 9900x, ryzen 9 9950x. one gets Computing_power[GFLOPS] = 0.460*Price[USD]-8.30 (R^2=0.91)[/img]