Topic: [H]ard|OCP bitcoin mining GPU performance comparison (Read 2495 times)

let's assume a stock 6990 consumes 350W (and does 750mh/s, not 700) and a stock 5770 consumes 105W. at a base consumption of 50W, your 4x 5770 system draws 470W and does 800mh/s. a 3x6990 system draws 1100W and does 2250mh/s.

1.7mh/s per W vs. 2.05mh/s per W. that's 20% in favor of the 6990s at this point.

let's say you want to build a 14.3kW (random, good numbers) farm. your PSUs are at 85% efficiency. that's 550W for each 5770 rig and 1300W for each 6990 rig. so you can supply 11 6990-rigs or 26 5770-rigs with electricity. let's assume space and cooling is free. alright. using your numbers 11 6990-rigs cost ~$31k, 26 5770-rigs cost ~$24k. since you're buying so many, let's say you get a 20% discount. $24.5k vs. $19.3k.

24.75gh/s@6990 vs. 20.8gh/s@5770. you have to accumulate $5.2k with your 6990s to break even with the 5770s. you generate 4gh/s delta. these 4gh/s cost absolutely zero electricity (that's because the 6990s are just more efficient than the 5770s). now the usual part: if btc value stays at $13 and difficulty won't increase more than 7% per cycle(*), it will take you about 300 days to break even. after these 300 days you have free 4gh/s.

i totally regret doing these calculations because the last part always sucks. you might be right - going for 5770s might be better. it takes too much time before mh/s per W come into play. go for mh/s per $! (keep in mind you can do the same calculations for less than 14.3kW. the 300 day-thing won't change, though, because cost, consumption etc are constant values).


(*) i highly doubt it's gonna stay at this rate. this is "my" worst-case-scenario because i'm almost certain it won't climb any higher. i won't assume it's going down, either, though, to satisfy some people's belief that bitcoins have a future etc.

Also, their GTX570 lists 98 MH/s; I get 115 MH/s with stock clocks, and up to 130+ MH/s overclocked.

They must be using the wrong miner or wrong settings.

These numbers are from june when I was spec'ing out system builds. I suppose the cost comparison to mhash depends on what you can get parts for, but this is from newegg. "base" is the cost of the MB+CPU+MEM+CASE+PSU.


   •   Base $411: 4x 5779 GPU (200mh/e) @ $130/ea. PSU: 800 watts. capacity: 800 Mhash/s. Total cost per box: $931. Hash performance per box/$: 0.859mh/$1
   •   Base $411: 3x 6870 GPU (280mh/e) @ $170/ea, PSU: 800 watts. capacity: 840 Mhash/s . Total cost per box: $921. Hash performance per box/$: 0.912mh/$1
   •   Base $366: 2x 6870 GPU (280mh/e) @ $170/ea, PSU: 800 watts. capacity: 560 Mhash/s. Total cost per box: $706. Hash performance per box/$: 0.797mh/$1
   •   Base $467: 2x 6990 GPU (700mh/e) @ $700/ea, PSU: 1200 watts. capacity: 1400 Mhash/s. Total cost per box: $1867. Hash performance per box/$: 0.749mh/$1
   •   Base $687: 3x 6990 GPU (700mh/e) @ $700/ea, PSU: 1200 watts. capacity: 2100 Mhash/s. Total cost per box: $2787. Hash performance per box/$: 0.753mh/$1
   •   Base $411: 3x 6950 GPU (350mh/e) @ $260/ea, PSU: 800 watts. capacity: 1050 Mhash/s . Total cost per box: $1188. Hash performance per box/$: 0.883mh/$1

dual gpu cards have the least overhead consumption/cost. you need less mainboards, less ram, less cpus, etc. so if you could get a dual-gpu card for just a little bit more than 2 single-gpu cards, go for the dual-gpu ones. i'm not too sure how well 6 dual-gpu cards fare in a single setup, though.

I have a rig that uses about 600w measured with a kill-a-watt from 2x5970 for 1.36 Ghash (undervolted to 1.0V @ 751/300), 2.27 Hash/W. I also bought these when they were selling on forums for $350-400 back in January.

Um, no it isn't. Unless you happen to have some magic numbers?

The cheap 5850's from earlier this year were the best ($129)

I just wish I had bought more of them.

A 5830 uses 175 watts so I don't know what you are talking about. and I get 320 MH.

So 6 5830s would use less wattage and have more MH than their top card.


Like I said the data is useless. Not the first time for HardOCP.

[H] tested popular high end video cards likely found in their forum member rigs already.  5830s are great bang per buck but they are absolutely AWFUL gaming cards in $/performance, $/watt, $/heat... Any metric you want to use.  Manufacturers had a real hard time selling them until the $99 pricepoint and bitcoins came along.

If it wasn't for bitcoins you'd see them 10-15% more expensive than 5770/6770s, which is roughly how they perform in gaming (the primary use for a high end consumer graphics card).

Which is why [H] didn't bother including them.  They didn't test 8800GTs and 4890s either, and I'd expect those to be far more popular among their readership than 5830s.  As far as 5850s and 5870s -- you can't buy one today, not for any reasonable amount of cash.  Seems perfectly sensible not to include unavailable cards.

I consider power consumption reasonably negligble. a 6990 uses roughly 20 watts more than 2x6950s in crossfire (a 6990 is 2 6950s on a single PCB), and 50 watts less than 2 6970s in crossfire. 50Watts over a year is roughly $48 in power bill. To me that is an amount that is unimportant to consider given the much more real variability of present concerns with bitcoin mining creation.

No 5830's? BAH ! Looks like they couldn't find any to test.

Data is useless as presented for obvious reasons unrelated to the above sentence.

But you didn't even address my statement at all. I see absolutely nothing about electricity prices, or even theoretical electricity prices, or power consumption!

Say wha? @ both of you. Both are not great price / performance. 5770s are ok Price/Performance, but poor Price/Rig (Low output for all the other costs that you have to pay). 6990s are like the opposite, decent Price/Rig, but bad Price/Performance and very hard to even find. Not sure why the article only shows CFX 6990s, but still $1600 for 1579 MHash is pretty poor, compared to say 327Mhash for $210 for a 69501gb ($1.01/MHash vs. $0.64/Mhash).

The kings are still the 5xxx series, but it does seem the 6950s are the best bang right now, especially if what other people are saying is true, about being able to unlock extra shaders even on 1gb models.

http://www.hardocp.com/article/2011/07/13/bitcoin_mining_gpu_performance_comparison/3

"Sorted most expensive to least expensive, longer bars = better value"

Now I guess if they go by the prices you'd pay off ebay or amazon I could see this being true. If you go by reasonable prices then the story is far different.

The 6990 is the best price/performance if you pay for electricity (like most people). That's what they're measuring.

If anyone wants to get their feet "wet" go buy a used 5770. Hands down, nothing comes close for the mh/s per dollar. It's easily 2-2.5mh/s per dollar. (Used ones go from $60 to 80, 90 for new and you can get 200mh/s out of them super easily.)

I'm not certain where they got their prices from, I hardly doubt the 6990x2 is the best performance for your dollar. They also didn't mention 5830/50's.

http://www.hardocp.com/article/2011/07/13/bitcoin_mining_gpu_performance_comparison


Nice to see a major hardware enthusiast site touch upon it. Don't be harsh on them for not using all the tweaks we hardcore miners know as it is still a good introductory read for those thinking about getting their feet wet.