Wattage question - 3x 280x drawing 1200w?

[Superdaan]

Hello all,

I'm new to mining, and have recently built a scrypt miner with 3 Sapphire Vapor-X R9 280x GPUs. Based on research (such as mining hardware comparison sites), I had estimated that these cards should draw approximately 300w each. I made a finger in the air estimate that the motherboard, RAM, CPU and USB harddrive together might draw another 100/150w, so the total would be around 1000/1050w.

My miner is Linux based with 4GB of RAM and the slowest CPU i could find.

I installed a kill-a-watt type of device (a power strip) just now and it measures that my miner is drawing 1200w. This seems extraordinarily high?

I have not yet undervolted these cards, this is something that I plan on doing later today, but I just wanted to pose this question here already because maybe there is a chance that the power strip is not displaying the wattage correctly.

Some extra info:
my cards' gpu engines have been overclocked to 1100 (factory default is 1000). This was the only way for me to get them to hash at ~750 kh/s each, which according to the same hardware comparison lists should be an attainable hashing speed with 300w power usage.

---------------
TL;DR
Does 1200w seem like an extraordinarily high power usage for a 3GPU 280x scrypt mining rig without undervolting the cards, and may the power strip be displaying incorrect watt information?

Thanks a lot

[vm1990]

im going to say the GPUs should be drawing around 1000w with the overclock in place the rest is most likely the CPU. the power the CPU uses is nothing to do with speed of it but its all related to the TDP of the CPU the lower it is the less power it uses old CPUs are horrible for power usage tats why most people here end up using some newer AMD semprons as there good and efficient little things.

[Superdaan]

Thanks for your reply, much appreciated.
The rig actually runs with an AMD Sempron (145) 2.8GHz processor. I do find the processor fairly noisy and it seems suspiciously active, so maybe there's something I can do to underclock or throttle the CPU.

Just a little update: I undervolted my GPUs and also changed the clockspeed from 1100 to 1080. The power strip now seems to suggest it's drawing about 950w, still more than I expected but better than 1200w. However it was showing 1100w with the current settings last night, it was only when I checked back this morning that it displayed 950w.

It seems strange to me that the rig would draw less power after it's been running for ~8hrs (with the new settings), is that expected behaviour? Still wondering if this cheap power strip is maybe a bit dodge.

[jameschase]

Thanks for your reply, much appreciated.
The rig actually runs with an AMD Sempron (145) 2.8GHz processor. I do find the processor fairly noisy and it seems suspiciously active, so maybe there's something I can do to underclock or throttle the CPU.

Just a little update: I undervolted my GPUs and also changed the clockspeed from 1100 to 1080. The power strip now seems to suggest it's drawing about 950w, still more than I expected but better than 1200w. However it was showing 1100w with the current settings last night, it was only when I checked back this morning that it displayed 950w.

It seems strange to me that the rig would draw less power after it's been running for ~8hrs (with the new settings), is that expected behaviour? Still wondering if this cheap power strip is maybe a bit dodge.

What kH/s are You getting by doing it? Well. maybe it needed more power to "kick" at start? And after some hours they are becoming stable?

[cryptocleaner]

3x ASUS R9 280x works at 950W Chieftek BPS-950C without any problem. 738kH per card, 1044/1500MHz

[pontiacg5]

New cards do seem to have some "burn in," but I'd imagine that you are just seeing normal temperature fluctuation. In my experience these cards draw more power the hotter they are.

You also realise that your power supply is not 100% efficient? If you are pulling 300x3 + 100 = 1000W it's pretty normal to see an extra 20% in power supply losses (pretty sure most run close to 80-90% efficient, depending on where in the power capacity range you run them.)

So, it would make sense for you to see 1200W of draw from 1000W of 12V.

With the new settings....

draw 950W from wall...
90% efficient from a 80 gold supply at 80% load makes for a...

950 x .9 = 855W 12V power

(855-100W)/3 ~250W per card.

Seems right to me  

[Superdaan]

Thanks for the advice

In my experience these cards draw more power the hotter they are.

Interesting, I've actually let them run hotter now (temp target set at 80 degrees celsius). Can't remember my reasoning, I think I was experimenting with getting a higher hashrate and figured that hotter meant less fan activity equals less power consumption... But I'm new to all this so just experimenting Maybe should try with temp target at 70 again and see the results.

It does run a whole lot quieter now which could be beneficial if I have to drop this thing in my bedroom! It's currently in the kitchen, my housemate isn't complaining yet but hmm....

draw 950W from wall...
90% efficient from a 80 gold supply at 80% load makes for a...

950 x .9 = 855W 12V power

(855-100W)/3 ~250W per card.

Seems right to me  

My PSU is a Corsair AX1200i which is supposed to be ~92% efficient at 100% usage, or 90% with a different voltage input (not sure which of the two im feeding it) - regardless that seems about right then in your equation.


When I got home from work just now the power strip read 1000w so I'm still working out what the actual usage is. I've now also underclocked the CPU from 2.8mhz to 2.0mhz, see if that does anything.

What kH/s are You getting by doing it? Well. maybe it needed more power to "kick" at start? And after some hours they are becoming stable?

Each card is averaging around ~732-738 kh/s now, which seems a decent result. Will keep trying to squeeze more out of it I guess!

[pontiacg5]

I think the cards running with more power under more heat has to do with transistor loss. If the card is cooler, resistance is lower so it takes less power to complete a clock cycle. If resistance is too high, the ASIC has to add voltage or power to get a "clean" signal through. There's a lot about GPU ASIC quality on the net too if you want to read more, that also plays a part in power draw/clock ratio.

https://bitcointalk.org/index.php?topic=64998.0

The blower fan on 290 cards is rated at 1.7A! That's 20W if you are running full bore on the fan. I recently took apart some dual-x sapphire 280's, and those fans were .7A each still. Wonder if that is considered as part of the TDP of the card?

[Superdaan]

Well, I just set the temp target back to 70 degrees as a test. After running for 20 minutes the reported wattage is around 980w, whereas at 80 degrees it was at 1000w so it does seem to help (in the short term anyway). Hundred times more noisy but that's OK for now Hashrates are about the same by the way, averaging around 735 per card.

Cheers!