The Most Efficient Rig

[gc40]

I am looking for a rig that will run well and will not destroy my electricity bill. I am not looking to be cheap, I want this to be a powerful rig, but I don't want to go for the "new, top of the line" stuff. I am looking for the current consumer market stuff that will be easily traded off when looking at what it can compute vs. its electrical consumption.

What would this efficient rig consist of?

Thanks Bitcoiners

[CentroniX]

600 watts is 600 watts, for example, it doesn't matter what year the hardware was made. If it were even possible, it wouldn't matter if you had a 486DX2/66, as soon as you drop video cards in the power consumption goes up, and so does your electric bill. you can kill a few watts by not having a CD drive, booting from a USB stick, etc. But your biggest power consumption is your video cards.

Only you can make this decision. Figure out your desired hash rate and find the most power efficient card combination to achieve that rate, and determin if its cost effective given your electrical rate and budget. The power consumption of the CPU and ram is nothing compared to what the gpus will consume.

[Fakeman]

I chose an e-350 motherboard for my current computer mostly because of its low power consumption. It only has one PCIe slot but it works perfect for my purposes. I tend to think a fast CPU is kind of overrated these days, GPUs are much more efficient at a lot of the grunt work anyway as we see with mining. I could lower consumption further by booting from a 2.5" laptop drive instead of a 3.5" "green" like I have, but it wouldn't make a huge difference in the total. You can get single core Atom boards that are lower powered and cheaper as well but I haven't tried them for mining. Of course none of those ITX boards will work with more than one GPU.

[Jack of Diamonds]

Stuff as many radeon 5830's in a motherboard as possible with optimal cooling.

Nothing comes close in terms of price/efficiency, unless you can get a really good deal on 5850s or a dual gpu.

They're mostly sold out though.

[Basiley]

depend price/scale.
if new AMD A38x0 APU is cheap enough, maybe using them on small mITX boards w/o discrete GPU, become interesting ?
just stack'em all !! like confetti, hundreds, thousands

[bcpokey]

depend price/scale.
if new AMD A38x0 APU is cheap enough, maybe using them on small mITX boards w/o discrete GPU, become interesting ?
just stack'em all !! like confetti, hundreds, thousands

They'd have to be crazy cheap. The llano chips have 400SP max I believe? That's like 1/4th of a 5870, so it'd have to be about $50-$75 for mobo+cpu+ram+HDD+psu

[Basiley]

depend price/scale.
if new AMD A38x0 APU is cheap enough, maybe using them on small mITX boards w/o discrete GPU, become interesting ?
just stack'em all !! like confetti, hundreds, thousands

They'd have to be crazy cheap. The llano chips have 400SP max I believe? That's like 1/4th of a 5870, so it'd have to be about $50-$75 for mobo+cpu+ram+HDD+psu

yep, thats what im mean. it become, then, way more price/cost/power-efficient solution, than both GPU and FPGA - based-solution, both
plug 2Gb stick for $25(?), $35 CPU, $15 USB-flash, $20 PSU and here you go !!
but again, thats meaningful, only if APU's/boards is reasonably-priced, which is not case, yet.

[jsidhu]

Stuff as many radeon 5830's in a motherboard as possible with optimal cooling.

Nothing comes close in terms of price/efficiency, unless you can get a really good deal on 5850s or a dual gpu.

They're mostly sold out though.

The 5830's are anything but efficient. It is however the best price/performance (not efficiency) if you can find it cheap enough... I picked up a few at $109/card.

By using the data from the mining harware comparison, here's a power effecincy comparison:

5830: 300 MH / 200 W = 1.5 MH/W
5850: 335 MH / 180 W = 1.8 MH/W
5870: 420 MH / 210 W = 2.0 MH/W
5970: 780 MH / 350 W = 2.2 MH/W
6870: 300 MH / 175 W = 1.7 MH/W
6950: 350 MH / 200 W = 1.7 MH/W (2GB Unlocked)
6970: 350 MH / 200 W = 1.7 MH/W
6990: 770 MH / 410 W = 1.8 MH/W

As you can see, the 5830's actually as bad as it gets for price/power efficiency, but if I found them cheap enough i'd surely pickup a few.

As far as building the most power efficient system I dont think its worth worrying too much about CPU/Board/Memory. For a dedicated miner, the CPU usage is minimal, and thus the power draw by CPU is negligible (unless you're also going to CPU Mine, but why?).

So, in the end, you're left with the GPU as the deciding factor in how efficient your rig is. Refer to the table above and just pick whatever's available..

[jagger27]

I cannot emphasize enough how inefficient the 5830s are. 6950s are easy to find right now and offer good performance.

[Fakeman]

Going by that chart, the 6950 is only 13% more efficient than the 5830. For that money the 5870 looks a lot better.

[TheSocialHermit]

Going by that chart, the 6950 is only 13% more efficient than the 5830. For that money the 5870 looks a lot better.

But try to even FIND a 5870.

[Jack of Diamonds]

I cannot emphasize enough how inefficient the 5830s are. 6950s are easy to find right now and offer good performance.

I picked 11 pieces of 5830 up for $85 each (common price would be around $99 as far as I've seen on TigerDirect and others that had clearance sales in the last 2 weeks), and they yield roughly 290-310mhash/s per card. I don't see anything in the price range offering even remotely similar performance at the price.

A 6950 2GB goes for $150 used, at least $199 new, & only gives about 390mhash/s. That's double the price for ~80mhash/s more.

[Fakeman]

Going by that chart, the 6950 is only 13% more efficient than the 5830. For that money the 5870 looks a lot better.

But try to even FIND a 5870.

Yeah there are not many out there, if I wanted to travel to a store ~150 km away I could get 2 for C$220 each but I don't want to piss off my landlord with the power bill.

[jsidhu]

I did some searching and found some reviews that covered Thermal and Power consumtion for all of these cards. The summary table is below, and while its not sane to simply divide the Temp by MH, it still gives you a fair idea of the heat generated by the card and its cooling, now keep it mind, this is probably done on a test bench with adequate airflow for the card.

Another point to note is how much Power and Heat is caused by 1GB of memory. The difference in the 6950 1GB + 6950 2GB is memory. So for those with dedicated rigs.. Underclock your memory and save as much power and heat as possible!

Model	MH/sec	Heat (Deg C)	MH/Centigrade	Power (Watts)	MH/Watts
5770	200	33	6.06	213	0.94
5830	300	60	5.00	263	1.14
5850	335	38	8.82	237	1.41
5870	420	35	12.00	270	1.56
5970	780	41	19.02	462	1.69
6870	300	44	6.82	247	1.21
6950 1gb	240	46	5.22	278	0.86
6950 2gb	350	54	6.48	306	1.14
6970	350	54	6.48	306	1.14
6990	770	63	12.22	475	1.62

One thing to note above is that while the power consumption is represented accurately for the 5970 and 6970, temperature is not. Since these cards have two GPU's, the single value for temperature needs to be adjusted accordingly. And one thing to note is that the 5970 has one fan at the far end of the card, and only cools one of the gpu's effeciently, the other one always runs very hot. The 6990's do not suffer from this problem because it has two fans a fan in the middle, afaik.

Looks like the 5870 is the king of the jungle.

Sources:

• http://www.bit-tech.net/hardware/graphics/2010/10/22/ati-radeon-hd-6870-review/9
• http://www.bit-tech.net/hardware/graphics/2010/03/22/xfx-radeon-hd-5830-1gb-review/7
• http://www.bit-tech.net/hardware/graphics/2009/10/13/amd-ati-radeon-hd-5770-review/11
• http://www.bit-tech.net/hardware/graphics/2011/06/10/sapphire-hd-5850-xtreme-1gb-review/7
• http://www.bit-tech.net/hardware/graphics/2009/12/01/sapphire-radeon-hd-5870-1gb-vapor-x-review/8
• http://www.bit-tech.net/hardware/graphics/2010/05/11/sapphire/8
• http://www.bit-tech.net/hardware/2011/02/11/amd-radeon-hd-6950-1gb-review/8
• http://www.bit-tech.net/hardware/graphics/2010/12/15/ati-radeon-hd-6970-review/10
• http://www.bit-tech.net/hardware/graphics/2010/12/15/ati-radeon-hd-6970-review/10
• http://www.bit-tech.net/hardware/2011/03/08/amd-radeon-hd-6990-review/9

The 5830's are anything but efficient. It is however the best price/performance (not efficiency) if you can find it cheap enough... I picked up a few at $109/card.

By using the data from the mining harware comparison, here's a power effecincy comparison:

5830: 300 MH / 200 W = 1.5 MH/W
5850: 335 MH / 180 W = 1.8 MH/W
5870: 420 MH / 210 W = 2.0 MH/W
5970: 780 MH / 350 W = 2.2 MH/W
6870: 300 MH / 175 W = 1.7 MH/W
6950: 350 MH / 200 W = 1.7 MH/W (2GB Unlocked)
6970: 350 MH / 200 W = 1.7 MH/W
6990: 770 MH / 410 W = 1.8 MH/W

As you can see, the 5830's actually as bad as it gets for price/power efficiency, but if I found them cheap enough i'd surely pickup a few.

As far as building the most power efficient system I dont think its worth worrying too much about CPU/Board/Memory. For a dedicated miner, the CPU usage is minimal, and thus the power draw by CPU is negligible (unless you're also going to CPU Mine, but why?).

So, in the end, you're left with the GPU as the deciding factor in how efficient your rig is. Refer to the table above and just pick whatever's available..

[Sukrim]

Most efficient below 1000 USD = FPGA currently. No GPU will beat a good FPGA at the moment and most likely in the future too...

[Fakeman]

You really can't multiply and divide by values in Celsius and get meaningful numbers. You would need to convert to Kelvin (i.e. add 273.15) to compare temperatures in that way. But really the actual temperature is highly dependent on factors other than the card itself. The wattage is enough to determine how much heat is produced.

[Sukrim]

You really can't multiply and divide by values in Celsius and get meaningful numbers. You would need to convert to Kelvin (i.e. add 273.15) to compare temperatures in that way.

Why this? Kelvin and Celsius are using the exact same scale + stepping an are just 273.15° apart (as you said).

Heat dissipation however is most likely NOT linear but rather exponential (it takes far shorter to cool down from 100°C to 90°C than from 30°C to 20°C if the room temperature is 20°C) so this chart is anyways useless. Heat is not generated in degrees Celsius, as a small hint...

[Fakeman]

Maybe you're right, the most relevant value to compare temperatures to in this case is probably the ambient temperature, rather than 0 C or 0 K. But really the total wattage tells you all you need to know about how much heat is produced. Something useful would be a column showing MHash/s/$ next to the one with MHash/J.

[Fakeman]

Model   MH   $   MH/s/$
5770   200   90   2.22
5830   300   100   3.00
5850   335   175   1.91
5870   420   220   1.91
5970   780   ?   ?
6870   300   190   1.58
6950   240   ?   ?
6950   350   280   1.25
6970   350   210   1.67
6990   770   680   1.13

[steamboat]

your numbers are way off for the 6950 2gb.

i'm running 2 @ 400mh/s pulling 220watts per card.