Radeon 7970 Tahiti GCN Benchmarked

[goxed]

http://hothardware.com/Reviews/AMD-Radeon-HD-7970-28nm-Tahiti-GPU-Review/?page=14

Not that much faster than Cayman 

[jjiimm_64]

well we dont know that..  the cayman has 1600?   7970 has 2048 Stream Processors


also to note: 
bitcoing miners ARE on the radar.  we have the 5th bullet on the overview of the new card

http://www.amd.com/us/products/desktop/graphics/7000/7970/Pages/radeon-7970.aspx#1

[plastic.elastic]

http://hothardware.com/Reviews/AMD-Radeon-HD-7970-28nm-Tahiti-GPU-Review/?page=14

Not that much faster than Cayman 

What are you expecting double performance?

Seriously? can you read some of the benchmark there? All the games that put bottlenecks on your GPU are really shine with 7970. Thats enough to show the next coming games will be more demanding.

And ofcourse, more driver tweaks from ATi.

[Epoch]

Tom's Hardware already reviewed the 7970 for bitcoin mining. Out of the box it is 10% faster than a stock 5870. But it uses 30% more power (i.e. 30% more expensive to run in terms of electricity), generates 30% more heat, and costs $500:

http://www.tomshardware.co.uk/radeon-hd-7970-benchmark-tahiti-gcn,review-32344-14.html

Bottom line: 7970 is not the card to get for bitcoin mining. Wait for the 78xx series ... much more interesting.  

[jake262144]

I'd like to make one point clear:
All current GPU miners are heavily optimized for VLIW4/5 hardware.

Isn't it reasonable to expect that GCN cards will do much better with miner optimizations?
Just look at the performance boosts we've had with newer versions of the current kernels.

10% performance increase over VLIW5? Rest assured, the 7970 will kick ass yet.

As to the 78xx series being THE series for bitcoin mining, they are much more familiar to us thus seem better.
Wait for the software to catch up.

[TurboK]

It would stand to reason that if per-shader-performance can be maintained on the same levels as with the VLIW hardware, then the 7970 should be 25% faster than a 6970.

With all the improvements GCN makes, I wouldn't be surprised if mining performance could be improved beyond. But I wouldn't be disappointed if it cannot be improved either, as I'm not familiar with the practices required for writing a mining application.

And 7970s can be overclocked a LOT. Though I imagine that could shoot up the wattage as well. Regardless, I'd wait for optimized kernels before believing any benchmark scores.

[RandyFolds]

And 7970s can be overclocked a LOT.

Says who? How long did they test it for?

I bet a 6970 can run at like, 1200mhz for a day...

[TurboK]

And 7970s can be overclocked a LOT.

Says who? How long did they test it for?

I bet a 6970 can run at like, 1200mhz for a day...

6970 is an ultra-stuffed card that was pushing the limit of the 40nm node. 7970 is brand new design, and it's a simpler design due to the 28nm nodes being so new and relatively immature. It's like the 5870 before, which was also a new entry at the then-current 40nm node, and could overclock to then-unheard-of levels.

Just look at it here:
http://www.guru3d.com/article/radeon-hd-7970-overclock-guide/1
The default settings in CCC allow for a 200mhz speed bump (from 925 to 1125). That's default voltage, default everything.

Add to that more efficient coolers that'll happen with 3rd party vendors, better software support in overclocking apps like afterburner, maybe even factory overclocked models, and we got ourselves a monster.

[grue]

first:
inb4 DiabloD3 posts a link to his benchmarking thread

also to note: 
bitcoing miners ARE on the radar.  we have the 5th bullet on the overview of the new card

http://www.amd.com/us/products/desktop/graphics/7000/7970/Pages/radeon-7970.aspx#1

or they're running out of things to benchmark? the only thing you can really do with GPU, besides mining is encoding and gaming.

[DeathAndTaxes]

It would stand to reason that if per-shader-performance can be maintained on the same levels as with the VLIW hardware, then the 7970 should be 25% faster than a 6970.

Even if true that would be horrible.  5970 are available used for $300.  Paying double for negligible improvement per watt and 33% less performance sound like crap to me.  

Given the rise of FPGA if not in 2012 then certainly in 2013/2014 dumping that kind of coin on a lackluster gain in performance and hopes of better kernel optimization seems like a fools errand to me.

[TurboK]

Mining itself is a fools errand with the extreme high difficulty and $4 dollar per btc. Unless you can run clusters of videocards or fpga, AND you have free electricity. And even then, the initial investment required is huge.

[jjiimm_64]

Mining itself is a fools errand with the extreme high difficulty and $4 dollar per btc. Unless you can run clusters of videocards or fpga, AND you have free electricity. And even then, the initial investment required is huge.

Not a fools errand: current snapshot of miners:
Totals    31.93 GH/s       15150 W    ฿ 19.03 ($79.91) / day   (after electricity)

[Epoch]

Mining itself is a fools errand with the extreme high difficulty and $4 dollar per btc. Unless you can run clusters of videocards or fpga, AND you have free electricity. And even then, the initial investment required is huge.

Not a fools errand: current snapshot of miners:
Totals    31.93 GH/s       15150 W    ฿ 19.03 ($79.91) / day   (after electricity)

I think the jury is out on that one.

Hmm, if my paper napkin calculations are are anywhere near correct, looks like you are paying ~$0.10/kWh for your electricity (15000W is 360kWh/day. Your electicity costs work out to about $35/day according to your data, or ~$0.10 per kWh). That is near the average rate in the USA (average is more like $0.112 according to http://en.wikipedia.org/wiki/Electricity_pricing).

I suspect you've been at this for a while already, so you've likely already paid off your investment (the June-August 2011 period was especially profitable for high hashrate miners). But for people considering just now whether to start mining, things are not so rosey. Consider:

32GHash/s is the equivalent of 80 5870 GPUs. Assuming $200 apiece (used), that's an investment of $16000. With $80/day profit, it would take 7 months to pay that down. 7 months is a long time in the young bitcoin universe; prices, difficulty, will be quite volatile in that time period. Anyone willing to invest this much for this long in such a volatile environment is taking a big risk. A leap of faith, nothing more. Yes, you can end up ahead. You can also end up in a deep hole.

Back to your numbers: So at current prices ($4.20 per BTC) and difficulty, your electricity costs represent 30% of your gross earnings. If BTC price drops back down to $2 (as it did several weeks ago) 2/3 of your earnings would be going towards electricity. $1.4/BTC is your no-profit point.

Come December 2012 when the mining reward gets cut in half, so will profits. 

[DeathAndTaxes]

32GHash/s is the equivalent of 80 5870 GPUs. Assuming $200 apiece (used), that's an investment of $16000. With $80/day profit, it would take 7 months to pay that down. 7 months is a long time in the young bitcoin universe; prices, difficulty, will be quite volatile in that time period. Anyone willing to invest this much for this long in such a volatile environment is taking a big risk. A leap of faith, nothing more. Yes, you can end up ahead. You can also end up in a deep hole.

7 month payback is a stupid metric.  A better one is assuming 3 year depreciation of hardware it is an annual return of 113%.  

Let me know when you have an alternative high risk investment w/ 113% yield.

See the $16K (to use your numbers) is never ALL at risk. Lets assume in the short term the equipment can be sold for 50% ($8K) the reality is likely much more but lets be conservative.  Still even the $8K is only at risk on day 1.  $80 in daily revenue means the amount "at risk" falls by $500 or so a week.  Unless revenue were to fall by 80% in first few weeks very little capital is put at risk.

Come December 2012 when the mining reward gets cut in half, so will profits.  

Simplistic nonsense.  Difficulty will fall as the marginal miners get squeezed out.  Nominal returns are worthless metric.  What matters is revenue/difficulty or revenue per unit of hashing power. The only way revenue gets cut in half is if hashing power and price remain constant.  Which would mean marginal miners continue to mine at a loss into perpetuity.

Still you are right.  Mining is a fool's errand.  If you haven't already you should contribute to my increasing profits by turning off your rig.

[P4man]

Let me know when you have an alternative high risk investment w/ 113% yield.

Ive heard of one. SOmething about a crypto currency thing. Bidcoin or something  .

Seriously, if I had to invest $16K in a high risk 3 year investment; Id probably choose bitcoin itself over a mining installation. Yes it is higher risk, after all, you could be wiped out almost entirely, but chances of achieving >113% annual yield seem at least as good.

[Epoch]

32GHash/s is the equivalent of 80 5870 GPUs. Assuming $200 apiece (used), that's an investment of $16000. With $80/day profit, it would take 7 months to pay that down. 7 months is a long time in the young bitcoin universe; prices, difficulty, will be quite volatile in that time period. Anyone willing to invest this much for this long in such a volatile environment is taking a big risk. A leap of faith, nothing more. Yes, you can end up ahead. You can also end up in a deep hole.

7 month payback is a stupid metric.  A better one is assuming 3 year depreciation of hardware it is an annual return of 113%.  

Let me know when you have an alternative high risk investment w/ 113% yield.

See the $16K (to use your numbers) is never ALL at risk. Lets assume in the short term the equipment can be sold for 50% ($8K) the reality is likely much more but lets be conservative.  Still even the $8K is only at risk on day 1.  $80 in daily revenue means the amount "at risk" falls by $500 or so a week.  Unless revenue were to fall by 80% in first few weeks very little capital is put at risk.

Come December 2012 when the mining reward gets cut in half, so will profits.  

Simplistic nonsense.  Difficulty will fall as the marginal miners get squeezed out.  Nominal returns are worthless metric.  What matters is revenue/difficulty or revenue per unit of hashing power. The only way revenue gets cut in half is if hashing power and price remain constant.  Which would mean marginal miners continue to mine at a loss into perpetuity.

D&T, I agree with your points. My goal was to throw an alternate perspective out there to counterbalance jjiimm's overly positive tone. There are always at least 2 sides to any argument, and jjiimm was leaning a bit too heavily on one side (no offense intended, jjiimm); I may not be the best proponent of either, though. I merely like to see things are kept in balance.

As for myself, I mine as well (though nowhere near your 10GHash/s or jjiimm's 32GHash/s) and plan to continue. So if you were to ask me whether I think mining is a fool's errand, I would be a hypocrite to say yes.

I think we both understand the situation: what matters is the bottom line (net profit). And net profit depends directly on difficulty, BTC price, hashrate, and expenses (mainly electricity and equipment depreciation). As long as people understand the whole picture, they can make up their own minds about whether mining makes sense for them personally. Neither choice is wrong.

[DeathAndTaxes]

D&T, I agree with your points. My goal was to throw an alternate perspective out there to counterbalance jjiimm's overly positive tone. There are always at least 2 sides to any argument, and jjiimm was leaning a bit too heavily on one side (no offense intended, jjiimm); I may not be the best proponent of either, though. I merely like to see things are kept in balance.

As for myself, I mine as well (though nowhere near your 10GHash/s or jjiimm's 32GHash/s) and plan to continue. So if you were to ask me whether I think mining is a fool's errand, I would be a hypocrite to say yes.

I think we both understand the situation: what matters is the bottom line (net profit). And net profit depends directly on difficulty, BTC price, hashrate, and expenses (mainly electricity and equipment depreciation). As long as people understand the whole picture, they can make up their own minds about whether mining makes sense for them personally. Neither choice is wrong.

Makes sense.  I try to keep it simple.  You don't want to be the marginal miner (miner w/ highest cost and thus lowest profit).  You make less during good times, your time to break even is longer and when price/difficulty falls (either falling price or rising difficulty) you are the one forced into negative cashflow first.

I can make plenty of money at $1 BTC or $100 BTC. I can also make plenty at 10M difficulty or 100K difficulty.  The goal is to be as efficient as possible (MH/W, $ per kWh, $ per MH/s).  Then the demise of marginal miners acts as a buffer in long term profitability.