If it's a power consumption competition then my netbook would rule the roost here. Just 20W at the socket which is amazing

Sounds very efficient and it is for what it does, which is 5 hours worth of flat out crunching while I'm away from the mains out and about on trains etc but.....

The Atom N270 "inside" at 1.6Ghz provides 749 floating point MIPS which is 37.45 floating point MIPS/Watt

Which means the huge monster cruncher I just built is over 2.8 times more efficient at using electricity to produce results!!

Incidentally switching on HT on the Atom still results in 749 floating point MIPS per CPU which suggests to me that the benchmark is single threaded in BOINC Work Units just take twice as long to run with HT

So why choose the not so easy to say floating point MIPS as the measure?

Well the WCG needs to do a lot of maths and usually this will involve a few decimal places - this is floating point maths. Integer MIPS are when you add 1+1 and get 2 (unless you still run the original Pentium ) which is great for finding the worlds largest prime number but not so great when it comes to the accuracy of the shape of a folded protein molecule which I think might turn out to be more important in the grand scheme of things.

The MIP bit is Million Instructions Per Second so you might expect 1000 floating point MIPS to be 1 Gflop which is a Giga (1,000,000,000) floating point operations per second

but it isn't

aarrgghh

It's down to the way benchmarks are designed and the way CPU's perform the floating point maths inside their various arrangements of silicon which means that 1 Flop is not a Milli floating point MIP

So to make this challenge easy when it comes to the benchmark I thought let's use BOINC's built in one and then apart from the vagaries of Hyperthreading we can compare apples with apples and exclude those banana's


106.5 Floating point MIPS/Watt at the socket

I can hear it now........There's no way that that Dave Autumns can claim to have the Greenest Cruncher on the WCG well..well.. it's a travesty



I'm here to provoke, cajole, and irritate the life out of some folk into action, in the drive towards more effective use of our Planet's resources through progress.

Yours

"Clean Green" Dave

Yeah, the Atoms are not the most efficient; they dont do much work. I had an Ion 330 (dual core with HT; 4 threads). Even though it overclocked to 2GHz (and not many do) it did far less work than an average dual core system. The system also used about 50 or 60W so it did not save anything compared to say my laptop, which is an old and power greedy 2.2GHz IC2D. It also uses 60W. The laptop would get more credit per day, by a long way - and that does not even get to crunch anymore. Unless the system has a good GPU I don'€™t use it to crunch. Any CPU with a very small instruction set is going to struggle for performance, unless you find a niche project that it does especially well on. So Atoms don'€™t cut it, and the HT is just a con.

I might post a few results of my other systems later, when I get a chance to measure them. For now just the one,
Phenom II 940 on an old Microstar K9A2 Platinum motherboard (AMD 790FX chipset, ATI southbridge). 2x2GB PC2-6400 DDRII, Vista Ult x64. 4xGT240 (shaders up from 1340 to 1600MHz)
27/06/2010 18:08:49 Processor: 4 AuthenticAMD AMD Phenom(tm) II X4 940 Processor [Family 16 Model 4 Stepping 2]
27/06/2010 18:08:49 Processor: 512.00 KB cache
27/06/2010 18:08:49 Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 htt pni cx16 syscall nx lm svm sse4a osvw ibs skinit wdt page1gb rdtscp 3dnowext 3dnow
27/06/2010 18:08:49 OS: Microsoft Windows Vista: Ultimate x64 Edition, Service Pack 1, (06.00.6001.00)
27/06/2010 18:08:49 Memory: 4.00 GB physical, 8.17 GB virtual
27/06/2010 18:08:49 Disk: 931.51 GB total, 635.90 GB free
27/06/2010 18:08:49 NVIDIA GPU 0: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 307 GFLOPS peak)
27/06/2010 18:08:49 NVIDIA GPU 1: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 307 GFLOPS peak)
27/06/2010 18:08:49 NVIDIA GPU 2: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 307 GFLOPS peak)
27/06/2010 18:08:49 NVIDIA GPU 3: GeForce GT 240 (driver version 19621, CUDA version 3000, compute capability 1.2, 512MB, 307 GFLOPS peak)
06/07/2010 11:37:30 2472 floating point MIPS (Whetstone) per CPU
06/07/2010 11:37:30 7690 integer MIPS (Dhrystone) per CPU

Presently Using 3 cores for CPU crunching, with one set aside to support the 4 GPUs. 355W at the wall. 550W Corsair 80+ PSU. The CPU did overclock by 10% with the same CPU voltage, but I would not try that with 4 PCIE powered (75W each) GPUs in it!
With Boinc Snoozed the power usage is 147W. That is not bad considering there are 4 GPUs in it. Each uses about 10W idle, so if I took 3 out it would be 117W.
Running 4 CPU tasks it uses 210W (with 4 idle GPUs). So that would be 180W crunching 4 CPU tasks at 3.0GHz (with only one GPU in the system, and sitting idle).
Dave'€™s machine is about twice as efficient as my Phenome II 940!

My Phenom II system is knocking on a bit now. I do have an i7-920, but who knows what that really does. There is some sort of adjustment factor used when awarding credit (because of the HT). So if the system asks for 100credits it tends to get about 83 credits* - I'€™ll go by that, even though I know it changes with different task types.
Anyway, when it is idle, the i7 is 125W at the wall with a Fermi GTX470 installed. I recon the Fermi uses about 45W when idle, so if I stuck in a more efficient GPU I could have a system that when idle used 90W. The thing is though, that is almost the power consumption of your system when its running!
I crunch with the Fermi, so it goes up from here.
With the GPU idle (GPU Snooze) and the i7-920 pegged at 2.67GHz (no turbo boost running keeps the CPU Voltage down) the system uses 198W at the wall. It is actually the same with 8 threads crunching as with 6 threads crunching!
Again, if I allow for the GPU (replacing it with one that would only use 10W), the system would in theory use 163W. Reading around that is about right.
Boinc says 2542 Whetstone and 6369 Dhrystone. Although there are 8 threads, we know Boinc measures one thread while not using the others, so it is not an accurate measurement – hence the downward WCG credit adjustments. Say its 17% over rated per thread* (much better HT performance than an Atom), then that is a total of 2542 x 8 x 0.83 = 16878.88 in total, for 163W. 87. That's actually a competitive 103.5 Floating Point MIPS/W, but still, short of your 106.5 FP MIPS/W.
So much for a £700 i7 system! You would have thought paying more money up front would save in the long run, but not so. Although the LGA 1156 i7 systems are more efficient than the 1366 systems, they are still much more expensive than a system based on Phenom II 905e Quad.

I expect the 45W AMD Athlon II X4 605e 2.3GHz 2MB, might be another efficient CPU.
Also spotted this one, 910e 2.6GHz 2MB 6MB socket AM3 65W 45nm SOI

* The ratios of 15 granted credits to asked for credits by my i7-920 (All HPF2 WUs):
0.89010989, 0.921438083, 0.819235226, 0.857277883, 0.80831643, 0.874405328, 0.798742138, 0.787587413, 0.790178571, 0.84717608, 0.811837456, 0.851590106, 0.839717742, 0.789686552, 0.883610451

I have 250 Watts at the wall socket.
The boinc benchmark gives me 3'742 FLOPS per core. It mentions also 12 cores. I have effectively 6 physical cores hyperthreaded. I do not know if the benchmark measures a total global perf value and then splits it per 12 or how it is done. So I will take the two versions:

1) 12 cores : 3'742 X 12 = 44'904 / 250 = 179.6 FLOPS/Watt

2) 6 HT cores (HT adds 15%) 3'742 X 6 X 1.15 = 25'819 / 250 = 103.2 FLOPS/Watt

Yes.
Regarding the fans I would not touch it first because I run at 4Ghz and I want to keep low temps for the cores. In this summer heat ventilation has to be optimal. In the mid of winter when my room temps in the basement go down then I can try to lower the fan speeds.
I will do a check with HT off. I have to check the GPU and maybe the memory settings. The values I give is for stable 24/7 crunching and not a kind of an absolute one shot record.