Hi Dave, I can't be specific as to the benefit (points-wise) of BOINC, but there is definitely an uplift. I run a range of machines, from P3's (Coppermine) to P4's (Presscott) and AMD's (Barton, Venice and Palermo).

I saw uplifts of 10 - 25% on the P3's, all of which were over-RAMmed so to speak; whereas I saw approximately 120% on my fastest machine, which runs only 512MB RAM. BOINC looks at processor only, which I think is correct.

I'm sure this subject has already been covered, but I think the main reason for the large uplift is that the multiplication factor is based on a Linux machine. BOINC benchmarks under a Linux OS are about half what they would be for Windows on the same machine.

Over half of my systems are multi-processor systems (2, 4, 6, and 8 way). The UD client only uses 1 CPU. That big was a waste of good CPU cycles. When I was running the UD client, not only were a whole lotta CPU's completely unused, but I noticed that the CPU in each system that was being used wasn't running at 100% utilization consistently. Each would spike at 100%.. and maybe run at 100% for a short time, but usually they would bounce around a lot between 80% and 90% for awhile... then back up to 100% again. Sometimes they would dip down to 60% for a short bit.

Once I switched my multi-processor systems to BOINC, not only am I using all the CPUs in those systems, but my CPU utilization is a stable and consistent 100% across all CPUs in those systems. Looking at one 2 processor system I get about a 2x improvement from switching from UD to BOINC. On another it's about 2.5x. improvement. On another, it's closer to 2.7x. From what I see, it's not just that I'm using more than 1 CPU in a system, it's that I'm using my processors more efficiently with the BOINC client. It boils down to a whole lot more work is getting done.

I haven't switched my existing single processor systems from UD... not yet anyway. I am running it on any new installs though, even if they do have only a single little underpowered and outdated CPU that has been shoved in a cabinet and been collecting dust for several years.

I <3 BOINC.
I <3 SMP systems.


My systems run a wide range of processor types and speeds. I've got old Slot-2 PIII Xeons from 550 MHz up to Xeon 3.4 GHz systems. I've got a few Opterons (242, 248, and 850). I've got some old slow Slot-1 Coppermine systems. Most systems have either 256 or 512 MB RAM per CPU. A few systems have 1, 2, or 3 GB of RAM per CPU.

The real reason I'm here is to fight some nasty diseases. I just want to do that as efficiently as possible. BOINC is allowing me to do that.

-j

Just to throw my 2 cents worth in again - I have 3 PC's running Boinc, and they all give more points than the UD Agent (they are all beyond the maximum benchmarking that UD allows)
However I also have 2 PC's running the UD Agent as it gives more points on these than boinc. One is a 1.7Ghz Pentium with 512MB RAM and on average gives about 15% less points with boinc, and the other is a 950 Duron with 384MB RAM, and this gives 50% less points with Boinc!!
So if I was to go solely on the performance of the 2 UD machines, I would be campaigning for the conversion factor to be increased, rather than decreased!!

As I stated over at the MOT thread:

I for one accept that there has to be an average score, we cannot gear it just to a few particular spec machines.

Ady

I had a spike yesterday, but it's too early for an average. I just validated more results than usual.

And David - welcome to the team. I do keep an eye on the new members, even if I can't welcome them all properly. Are you guest crunching, or are you here to stay? You're welcome either way.

Just to throw in my two-penny worth. Since this thread started I have been running one of my machines with both the UD and Boinc installed (both crunching FAAH work units) as opposed to just UD. By the way Windows XP seems to be doing a reasonable job of balancing the two.

Here we come to the interesting bit: whilst my results returned rate remains pretty much unchanged, indicating that the Boinc throughput is not perceptably different, my points returned rate appears to have gone up, indicating that for what ever reason Boinc is getting more points awarded.

The CPU in this machine is an AMD AthlonXP 2000+, the UD rating for it is 110