Very Interesting.
There is no CPU limitation, but there may be a core number limitation.
If you go into the "My Grid" selection and then "Device Profiles" and select the profile of your extra large server (dafault, school, home, or work) then "Custom Profile" button you will see that you have two parameters in the "Advanced Options" "Processor Usage" paragraph of the page that says:

On multiprocessors at most use "XX" processors
On multiprocessors use "100%" of processors

The "XX" parameter means the number of cores and in fact the number of threads. On my machine this number is set to 24 which is enough as I have an hexacore with hyperthreading that means 12 threads.

You have 8 CPUs. But how many cores per CPU ? and is there HT or not. If you have 8 cores per CPU and 8 CPUs then you have 64 cores and if they are hyperthreaded you have 128 threads. You must then set the parameter to 128. I have no idea what is the absolute maximum number of threads than is allowed by WCG.

It is important to know what type of processors (CPUs) you have and how many threads they can manage in parallel.

Okay, a bit more experimentation, and I've decided to terminate the project.

I did get it to use all 8 cores. It was a benchmarking issue. I built the machine on a single CPU, Single Core virtual machine, then migrated it to a dual CPU, quad core per CPU virtual machine. So the previous update sent down only 5 work units. When it suddenly saw 7 more cores added, it simply picked up any work unit it could find, but didn't ask for any more. By re-benchmarking and forcing an update, it got it working.

The machine is actually 2 x 4 Core processors, though I'm not sure of what type. It had an estimation of about 70 minutes per HCC work unit, which is faster than any other computer I own.

The way I set it up isn't going to be a reliable way to price it. $35/week was based on the average price per hour of computing over the past 3 months. But as I set mine only 2c about the present set price, I got priced out and shut down very quickly. I'm guessing if I had priced it with a margin of 10c/hour more than the average, I would have had 100% run time.

The problem with getting shut down, is that it didn't work how I thought it woudl work. What I had been expecting was that it would do a graceful shutdown, and restore to the same point it was when I was kicked out. Instead, it deletes that virtual machine and when the price is right again, it rolls back to the image I booted off, losing all the checkpoints (in fact, after the reboot, it went back to using only 5 cores again).

There is a way around this, because you can provision a drive and mount it as a separate permanent drive. But then I'd have to manually build BOINC, instead of installing it from RPMs, unless I can figure out how to get the RPMs to install to the permanent mount point, rather than just /usr/bin

I'm too busy at work at the moment to spend any more time on there, but if I can figure out how to get it running on their 8 x Cluster Computer, which averages 55c/hour (About $80/week? - But it claims to have slightly more than 4 times the amount of computing power as the single "extra large" server I built.) then I will probably get it up and running again, and let it run for a week or two, as a bit of a "cash" donation to WCG. See how I go. Maybe will have some time over Christmas.

I checked with Amazon Web Services (AWS) and here is how things look.

I will do my analysis on the base of hardware that has to run fully 24/7 365/year dedicated to WCG.

You can rent an "Extra Large Quadruple" hardware instance. It is a two CPU unit were each CPU has 4 cores for a total of 8 physical cores. The CPU's are Xeon 5570 "Nehalem". They are HT capable. The speed is 3 GHz. We have a performance 8x3 = 24 GHz.

We will consider a three year period and consider amortization over this period. The reason is that AWS has a pricing for one or three years. So I took three years.

Renting such a unit for three years under the conditions mentioned above (running 100% exclusive WCG under Windows) will cost: 21'612 US$

This makes the AWS CPU cost at 900 US$ /GHz over 3 years.

If I compare now to my hardware. All my CPU's dedicated exclusively to WCG are i7 980X with six physical cores and run at 4.0 GHz.
The performance of 8 cores is 8X4= 32 GHz.

The acquisition cost was 2'000 US$ for a six core rig. The equivalent 8 core rig costs 2'666 US$.
Consumption is 0.0475 kWhr/core/hr that is for 8 cores 0.380 kWhr/hr.

Running cost are for the three year period are 2'100 US$ for eight cores. That makes it a total of 4'766 US$ if we consider that the hardare is completely amortized.

This brings the Hypernova CPU cost to 148 US$/Ghz over 3 years.

The AWS solution is 600% more expensive.

If I decide to amortize my hardware for four years, which would be acceptable for such high perf CPU like the 980X then the picture is even more horrible for AWS.

To rent the "Extra Large Quadruple" for one additional year has a cost of 8'800 US$. The cost is then 366 US$/GHz/year.
The total four year cost is then 900 + 366 = 1'266 US$/GHz

Whereas my solution stands at a total four year cost of 5'466 US$ which boils down to 171 US$/GHz for the whole four year period.

The ratio is now worse. The AWS solution is 740% more expensive.

Conclusion forget it !

Thanks for the info KodeX.
I will give a look.

I should put it in perspective:

1. In Australia, electricity is 27c per kw/h. My current computer uses about 300 on idle, about 450 on CPU full, and about 650 with both GPUs running. Let's say 450W.

In Australia, that's nearly $20/week in power alone.

2. We don't need heating... We need cooling! Even as late as late Autumn!

I also wasn't planning on using it to save money. As I said, I was only going to run it for a week for fun, and to learn about the AWS services, and how they are instantiated, terminated, and perform.

My actual plan is to get an Ivy Bridge core when it comes out next year. With a 22nm fabrication, expected TDP should be a little lower than a 2600, but with extra cores, you should also get a lot more processing done. I will probably then run it at work for a while to save money at home ;)