First-time build: I've picked out the parts, but I'd really appreciate some input

[LMF5000]

LOL, nice. Benefits of living on a small island I guess.

Yeah, there are about 400 boinc users who appear under "Malta" in all the boinc lists. Bit low considering the island's population is almost 500,000 people this year (including sister island Gozo). Perhaps many crunchers don't bother to select their country when installing boinc :hmm:

 

[LMF5000]

Did a few power consumption tests. I don't have a kill-a-watt meter, but I do have an AC clampmeter. However since it measures current by electromagnetic induction it's not very accurate. In my opinion it underestimates the true value, but anyway, here are the results:

State, Watts

Startup (i.e. during boot) 104.5 W
Idle 58.52 W
CPU only at full power (BOINC) 108.68 W
CPU only at full power (IBT normal) 125.4 W
CPU only at full power (IBT maximum) 123.31 W
GPU only at full power (BOINC) 206.91 W

CPU & GPU at full power (BOINC) 244.53 W

So with the CPU and GPU both at full power, it draws almost 250W from the plug. Makes an 850W power supply seem a little excessive . Though like I said, I feel the meter is underestimating it.

Anyway, in Malta, running this at full power 24/7 costs €35 a month in electricity bills (assuming €0.20 per kWh).

Edit - "IBT" means Intel Burn Test.

 

[krnmastersgt]

Is your electricity provided at a fixed rate regardless of consumption? My electricity is pretty cheap until I hit too high of a consumption for any given month (lower during the summer months since everyone blasts their AC) and boy when I hit that next bracket it was painful for the wallet.

Also it was said that 750W would be more than enough even for an SLI'd 670 config + drives. With 750W you have a LOT of room to breathe with though, and you do waste a bit of energy in the conversion process, so it is less efficient than could be, but unless you built this rig PURELY to crunch it shouldn't be much of a problem (I'm hoping your gaming outweighs DC by a bit ).

 

[LMF5000]

Is your electricity provided at a fixed rate regardless of consumption? My electricity is pretty cheap until I hit too high of a consumption for any given month (lower during the summer months since everyone blasts their AC) and boy when I hit that next bracket it was painful for the wallet.

Also it was said that 750W would be more than enough even for an SLI'd 670 config + drives. With 750W you have a LOT of room to breathe with though, and you do waste a bit of energy in the conversion process, so it is less efficient than could be, but unless you built this rig PURELY to crunch it shouldn't be much of a problem (I'm hoping your gaming outweighs DC by a bit ).

It's also staggered here, but in a complicated way. See: http://www.enemalta.com.mt/index.aspx?cat=2&art=5&art1=9

If I can keep it to below 1750 units per person per year, they chop 25% off the bill. The instant you overshoot that by a little, you get the full bill.

Price per unit is also divided into consumption bands. It gets very expensive to exceed 10,000 units per year (or the pro-rata equivalent, eg. 2500 units if the bills are delivered every 3 months, 5000 units if bills are 6-monthly etc.). It also factors the number of people living in the house, whether the house is residential (i.e. you live in it), domestic (i.e. you own it but rent it out to other people), or non-residential (i.e. you work in it), whether you're single or triple phase, whether you prefer being charged in kVA or kWh, what your favourite brand of porridge is and what song is playing on the radio on bill delivery day ...

So I just ignore everything and assume an average of €0.20. Haha . Sometimes it pays to be an engineer - because it means you have the right to simplify things by taking reasonable assumptions like this one (as long as you err on the pessimistic side)!

A 650W supply would've been great, but OcUK didn't have any in stock from a brand I liked, so I went for a 750W. But then they apparently made a mistake at the warehouse as well and didn't actually have them in stock either, so they gave me the 850W. So my supply runs at a very comfortable (but inefficient) load of 25%. Wonder how that happened... ()

I also hoped my gaming would outweigh my DC by a bit, but that hope was quickly bludgeoned by the baseball bat of reality - in the form of my supervisor requesting me to write a scientific paper about my research within just two-weeks. And on the day the PC arrived. I've only had time to build the PC, set up all my programs, and get 5 GTAIV missions done since the boxes arrived on the 5th of July :'(. The most used program after BOINC right now is Microsoft Word, closely followed by excel and adobe reader. Ironically the SSD is really useful for speeding up the opening of 70 txt files and converting them to excel using macros.

Deadline's tomorrow though - so hopefully this weekend will make up for it .

 

[mfenn]

Did a few power consumption tests. I don't have a kill-a-watt meter, but I do have an AC clampmeter. However since it measures current by electromagnetic induction it's not very accurate. In my opinion it underestimates the true value, but anyway, here are the results:

State, Watts

Startup (i.e. during boot) 104.5 W
Idle 58.52 W
CPU only at full power (BOINC) 108.68 W
CPU only at full power (IBT normal) 125.4 W
CPU only at full power (IBT maximum) 123.31 W
GPU only at full power (BOINC) 206.91 W

CPU & GPU at full power (BOINC) 244.53 W

So with the CPU and GPU both at full power, it draws almost 250W from the plug. Makes an 850W power supply seem a little excessive . Though like I said, I feel the meter is underestimating it.

Anyway, in Malta, running this at full power 24/7 costs €35 a month in electricity bills (assuming €0.20 per kWh).

Edit - "IBT" means Intel Burn Test.

Actually, those numbers seem fairly reasonable to me. The CPU is 95W, so that maxed out leaves 30W for everything else idling, which is quite normal. The GPU maxed out could theoretically add another 170W, and you're adding 136W which is not unexpected for a CUDA program (they're generally not efficient enough to achieve full residency like a graphics app will).

 

[LMF5000]

Actually, those numbers seem fairly reasonable to me. The CPU is 95W, so that maxed out leaves 30W for everything else idling, which is quite normal. The GPU maxed out could theoretically add another 170W, and you're adding 136W which is not unexpected for a CUDA program (they're generally not efficient enough to achieve full residency like a graphics app will).

4am in malta right now. Still trying to finish that paper. Let's see if I can still think clearly enough to produce some good maths. :\

At full load:

CPU = 77W TDP. Probably more because overclocked to 4.2GHz
GPU = 170W TDP
HDD = 10W
SSD = 5W
8 fans = 20W approx

Total so far = 282W, ignoring consumption of motherboard components like chipset, RAM, wifi card and USB devices. And yet my clampmeter measured 250 watts, at the input of a unit with about 80% efficiency, meaning that only 250*0.8 = 200W got to the electronics. Now, GPU-Z says that I consume 70% of TDP while running BOINC, so that's 119W instead of 170. Dunno if the CPU has a similar behaviour. So they don't look as wrong as I originally thought.

I'd measure the reading of a known load (like a 60W fan) with the clampmeter and report the readings, but our electricity supply is way beyound spec. It should be 240V RMS, but when I measure actual electrical outlets with the meter they only have 209V. And it falls to 204V when the kitchen A/C is turned on D:. So none of the simple devices in the house (lightbulbs, fans, fridges etc.) draw their rated power - though I could probably calculate the actual predicted power at 209V using ohm's law (P = V^2/R) for a resistive load like a kettle or a heating element. Knowing rated P at 240V (just read it from the label) it's a simple matter of multiplying rated P by (209/240)^2. Then I could measure the actual load with the clampmeter and see if there's a discrepancy between readings.

Won't work for lightbulbs though because they're non ohmic and don't scale linearly with supply voltage. Things with motors (fans and so on) are even more complicated. I think my best bet is to measure the power draw of the kettle cause it's a nice ohmic load. I feel that I'm starting to ramble a little. Time for sleep... Stay tuned!

P.S. for the numbers above, I used Power = 209V * Amp reading on clampmeter. So should be accurate in that regard...