Real World Trifire 7970 Power Consumption Recordings.

[RussianSensation]

If anything I think the OP should go higher, but that's just me!

With that train of thought I still can't fathom why you don't go Tri-fire HD7970 OC for BTC. At $26 a pop, you are missing out. 3x 7970s OCed would clear $245 a month. What are you waiting for? $300 GTX680 for PhysX heheh? In 5 months, you'd have 3 paid off 7970s, picked up 8 free games (6+2). :biggrin:

 

[Stuka87]

That power usage seems high to me?

The rig in my sig uses ~420W peak during benchmarking, and around 375W while gaming (BF3). Now obviously a 7970 will use more than a 7950, but I am sure my CPU is consuming more than his. And my card is OC'ed, although not quite as high as his, and I know power usage does spike pretty quickly past a certain point.

 

[RussianSensation]

This is because the Vcore is 1.29V. If you overclock HD7970 beyond 1200mhz and crank Vcore to nearly 1.3V, the power consumption goes up exponentially.

 

[Elfear]

Seems really high to me too. With my cards at 1200/1500 1.25V I get ~940W from the wall running through 3DMark11 Extreme. Granted the new 3DMark probably pushes the GPUs a little harder but that is still a pretty big difference. My rig is very similar too except I'm running a SB rather than an IB.

For mining I run my cards at 1050/340 1.08V which only draws ~640W from the wall.

 

[3DVagabond]

I am not looking for efficiency mate, just performance. This topic was purely to generate awareness and discussion as I am sure many people have absolutely no idea how much power a rig can draw from the wall.

Efficiency is part of performance, and I see nothing "real world" about your power usage. This is "How much power a system can use if you don't care."

 

[lavaheadache]

Been following this thread via my phone and just got power back late last night/this morning after 5 days cause of that stupid north east snow storm.

Anyways wanted to contribute some numbers to you folks.

These are total system numbers for the following rig playing Far cry 3 at a stressful spot alt-tabbing and changing clock speeds without moving in the game.

3770k @4.5
2x8 gig 1600mhz ram
2 ssd's
1 hdd
4 case fans
xfi
7970 lightning at various clock speeds and volts

stock voltage #'s
925/1375- 66 fps - 361w
1070/1400- 74 fps - 386w
1125/1575- 78 fps - 404w
1175/1575- 80 fps - 407w
1175/1600- 80 fps - 412w

vcore 1.21

1200/1600- 82 fps - 432w

vcore 1.25

1250/1600- 86fps - 477w



**edit**

I want to make special note that it is really hard to get actual power measurement numbers because I noticed that as gpu temperature increases so does wattage

If I fire up the game in the exact same spot and a given voltage the wattage goes up significantly as the gpu temp increases.

1250/1600 starts out @ about 422w but gets up to just about 480w as the Gpu settles at 71ish degrees on stock fan profile.

The differences between 1200/1600 1.21vcore vs 1250/1600 1.25v core are pretty dramatic while using the card. The fans are noticeable louder and the temps are noticably hotter(72c vs 67c) but there is definately an increase in frame rate. Not that while using a 60 hz monitor it makes a difference.

This is where using a frame limiter comes in handy.... I'll be right back with another set of numbers with the card locked at 60 fps

ok, big difference

1250/1600 @1.25v core at same spot the card is very quiet and cool (61c) while using 383 watts.

Frame limiter offers great advantages. Minimal power increase when you reach desired frame rate even with massive overclocks yet allows the card to use maximum performance when needed, all while keeping the card cool and quiet. Obviously this scenario changes when the card can't maintain the desired framrate but that is why we are doing these tests, to figure out what is best for the individual.

 

[BallaTheFeared]

I want to make special note that it is really hard to get actual power measurement numbers because I noticed that as gpu temperature increases so does wattage

Of course it does :thumbsup:

The cooler a chip runs, the less power it will use as well it works both ways.

His same cards on air might not even be stable at those clocks, the increased heat would most likely require more voltage as well as draw more power.

That's one of the reasons people go water, with chilled water he could probably take those cards to at least 1300MHz at the same wattage/voltage.


You also need to account for sample variance, one card might do 1250 @ 422w another might do it at 400w, and another still might need 450w yet another might not even be able to hit those clocks. Even stock cards with stock clocks have wattage variance.

 

[lavaheadache]

My message should have been clearer. It wasn't that I was bringing notice to the fact that power consumption went up with temperature. It was that it went up what appeared to be 50+ watts with a 10-15c increase. That just seems like a lot to me.


Balla, just wanna clarify that those wattage figures an entire rig #'s cause your statement reads that the card itself is 400+ watts.

 

[notty22]

The OP proved the 7970 is very much like the gtx 480 when over-volted and over-clocked even running water-cool. Evidently some posters mind more than him.

I'd say it's not a big deal as long as you have a big enough PSU and power at the wall/ fuse box. It's (incoming car analogy) like checking gas mileage when you hit the nitrous switch on a 200+ shot. You are after performance and all can be well ,if you know what your doing, in for.

Good information, thanks.

 

[BallaTheFeared]

My message should have been clearer. It wasn't that I was bringing notice to the fact that power consumption went up with temperature. It was that it went up what appeared to be 50+ watts with a 10-15c increase. That just seems like a lot to me.


Balla, just wanna clarify that those wattage figures an entire rig #'s cause your statement reads that the card itself is 400+ watts.

I'm well aware, you shouldn't be surprised though, I thought you had a Fermi?

Same thing with Fermi, more heat, more power, more volts, compounded. Same thing with Cayman... Pattern here I swear!

It's true with any card/cpu, the only saving grace for GK104 is that nvidia didn't allow voltage control past an incredibly low amount. If they had it would have just started spinning it's wheels in a voltage/heat bandwidth bottleneck of mud for a car analogy!

 

[lavaheadache]

i did have Fermi.. a 580 and currently a GTX 480 superclocked

 

[RussianSensation]

The OP proved the 7970 is very much like the gtx 480 when over-volted and over-clocked even running water-cool. Evidently some posters mind

Based on my feedback/ desire for more information, lavaheadache posted an excellent analysis of overclocking / power consumption relationship. It was a very informative and useful post to prospective GPU overclockers at all experience levels. One key forum function is to learn new information. This is facilitated through information sharing. Where pertinent information can serve a useful function in extending our understanding regarding real world usage of products, it should be encouraged. In this instance lava's post took OP's findings to the next level

No poster in this thread even attempted to mind the respective GPUs' power consumption levels or has objected to the findings being presented. Instead, specific observations were made with respect to the effect of applying higher levels of GPU voltage to an ASIC, the act of which induces higher levels of transistor leakage and consequently leads to an exponential rise in ASIC's power consumption. This was confirmed by detailed findings in lava's post, which outlined the impact of incremental Vcore increases.

Thanks lava!

 

[lavaheadache]

Based on my feedback/ desire for more information, lavaheadache posted an excellent analysis of overclocking / power consumption relationship. It was a very informative and useful post to prospective GPU overclockers at all experience levels. One key forum function is to learn new information. This is facilitated through information sharing. Where pertinent information can serve a useful function in extending our understanding regarding real world usage of products, it should be encouraged. In this instance lava's post took OP's findings to the next level

No poster in this thread even attempted to mind the respective GPUs' power consumption levels or has objected to the findings being presented. Instead, specific observations were made with respect to the effect of applying higher levels of GPU voltage to an ASIC, the act of which induces higher levels of transistor leakage and consequently leads to an exponential rise in ASIC's power consumption. This was confirmed by detailed findings in lava's post, which outlined the impact of incremental Vcore increases.

Thanks lava!

Glad you enjoyed it. Gave me incentive to a little more indepth testing. I'm combining the old results with new results but I reran the old numbers and they fell within margin of error so I kept the originals.

1.174v (stock)

925/1375 - 361w
1070/1400 - 386w
1125/1575 - 404w
1175/1575 - 407w
1175/1600 - 412w ---couldn't do 1200 so I increased mem speed a little


1.21v

925/1375 - 376w
1070/1400 - 401w
1125/1575 - 422w
1175/1575 - 427w
1200/1600 - 432w

1.25v

925/1375 - 401w
1070/1400 - 427w
1125/1575 - 442w
1175/1575 - 452w
1200/1600 - 462w
1250/1600 - 477w

 

[philipma1957]

....

3770k @4.5
2x8 gig 1600mhz ram
2 ssd's
1 hdd
4 case fans
xfi
7970 lightning at various clock speeds and volts

stock voltage #'s
925/1375- 66 fps - 361w
1070/1400- 74 fps - 386w so this with no frame limit
1125/1575- 78 fps - 404w
1175/1575- 80 fps - 407w
1175/1600- 80 fps - 412w

vcore 1.21

1200/1600- 82 fps - 432w

vcore 1.25

1250/1600- 86fps - 477w



**edit**


ok, big difference

1250/1600 @1.25v core at same spot the card is very quiet and cool (61c) while using 383 watts.

vs the setting above with frame limit

Frame limiter offers great advantages. Minimal power increase when you reach desired frame rate even with massive overclocks yet allows the card to use maximum performance when needed, all while keeping the card cool and quiet. Obviously this scenario changes when the card can't maintain the desired framrate but that is why we are doing these tests, to figure out what is best for the individual

both pull 383 or 386 watts. one setting allows more frames the locks the frames but allows better over clock. for the sake of argument call the power equal.

what is the better setting pros cons? why over clock? why not the first setting? it gives more frames. yeah memory speed is 1400 vs 1600 but frames are 74 vs 60

 

[lavaheadache]

both pull 383 or 386 watts. one setting allows more frames the locks the frames but allows better over clock. for the sake of argument call the power equal.

what is the better setting pros cons? why over clock? why not the first setting? it gives more frames.

the difference is 383w vs. 477w, nearly 100w difference and the frame rate is 86 frames. I think you read the graph wrong.


The whole point to the frame limiter is to not needlessly burn power. If you have a 60hz monitor there is ZERO advantage to running higher fps than 60. The spot where I was monitoring wattage was just an easily repeatable spot and not exactly a spot where the sh!t hit's the fan. In those spots the frame rate is way lower and where the overclock comes in very handy and came make a world of difference. This is why I think frame limiters are great, they let the card work only as hard as it needs to, seemlessly.

 

[MustangSVT]

that is almost 6x the power usage of my itx build

 

[UNhooked]

So let me get this right. The OP posts numbers based on his experience. Are folks upset because the numbers are high?

 

[96Firebird]

So let me get this right. The OP posts numbers based on his experience. Are folks upset because the numbers are high?

Yes. Welcome to the wonderful world of VC&G.

 

[UNhooked]

Why are folks upset? I have to run my cards at stock else my wattage increases significantly. Probably in the same neighborhood as the OPs numbers.

I can easily surpass my 750W PSU if I were to flash to the GHZ bios and OC my cards.

However OC in my case is pointless since I only game at 1080P and at stock settings all the games play with no issues.

 

[notty22]

So let me get this right. The OP posts numbers based on his experience. Are folks upset because the numbers are high?

It seems in part.
Also you want to play games or bench at those o/c's, and not chase crashes and bsod's. You are going to have more complications or crashes from low balling voltages across 3 cards. The complexity of the o/c is complicated by extra gpu's. That's why AMD went with 1.25v on the ghz cards, even though the armchair critics, tell us AMD does not know what they are doing.

 

[RussianSensation]

Why are folks upset? s.

Please name 1 person who is upset about OP's numbers? Read the entire thread and see what's actually being discussed and expanded upon. What's the key learning point from the results? No one is disputing OP's own results. If someone decides to put 1.5v into IVB and then tries to pass that on as 'normal' / real world consumption of Ivy in over locked state would you agree with that statement? At that point what is being reported is how IVB acts under extreme levels of voltage. Most people who Actually care about power consumption would never accept 100mhz more on IVB for a 100w rise in its power consumption.

What's actually being reported here is what happens to a chip when the optimal level of Vcore is breached and massive amount of leakage sets in. It is clear these specific 7970's are not top binned ones which is why so much voltage is required to keep them at 1225mhz. Exact same scenario Xbitlabs explored in detail with HD5850. What's being stated is under such conditions each 7970 would use up 60-70 watts of extra power for what often amounts to just a 25-30 MHz GPU clock increase. How many people would accept a 200-250w penalty on 3 7970s for a 25-35mhz GPU clock bump on an 1185mhz chip? Lava clearly outlined this in great detail. If someone wants to squeeze maximum performance out of a chip at all costs, then they obviously do not care about power consumption. That's obvious.

It is completely incorrect to claim that an average 7970 OC draws as much power as a 480 just because someone is willing to accept a 3-4% increase in GPU clock at all costs. There is a big difference between smart overclocking and an all balls out approach. Which is why this thread has so much learning material/implications.

As an example an 1180mhz 7970 draws 239W. A stock 480 draws 270W. You could push over 300W easy if you go 1.29V and marginally higher on the GPU clock if you have a very leaky chip.
http://www.techpowerup.com/mobile/reviews/HIS/HD_7970_X_Turbo/26.html

And the key takeaway from this thread: if you want to OC to the absolute max, expect power consumption of any ASIC to reach stupid levels of power usage due to leakage galore. When overclocking always be mindful of the Vcore barrier breach as eventually the ASIC stops scaling well and exponential increases of voltage are required for minimal frequency increases. That is the learning point.

This is why I specifically asked the OP to carry out a scaling clock speed vs. Vcore analysis so we could all see how little benefit is gained from an ASIC once it reaches a certain level of frequency and Vcore. It would also be interesting to see what clocks those specific GPUs could hit at stock voltage of 1.174v and then at stock Ghz voltage of 1.256v. This extra information would have been great.

 

[UNhooked]

Please name 1 person who is upset about OP's numbers? Read the entire thread and see what's actually being discussed and expanded upon. No one is disputing about OP's own results. If someone decides to put 1.5v into IVB and then tries to pass that on as 'normal' / real world consumption of Ivy in over locked state would you agree with that statement? At that point what is being reported is how IVB acts under extreme levels of voltage. Most people who Actually care about power consumption would never accept 100mhz more on IVB for a 100w rise in its power consumption.

What's actually being reported here is what happens to a chip when the optimal level of Vcore is breached and massive amount of leakage sets in. It is clear these specific 7970's are not top binned ones which is why so much voltage is required to keep them at 1225mhz. Exact same scenario Xbitlabs explored in detail with HD5850. What's being stated is under such conditions each 7970 would use up 60-70 watts of extra power for what often amounts to just a 25-30 MHz GPU clock increase. How many people would accept a 200-250w penalty on 3 7970s for a 25-35mhz GPU clock bump on an 1185mhz chip? Lava clearly outlined this in great detail. If someone wants to squeeze maximum performance out of a chip at all costs, then they obviously do not care about power consumption.

It is completely incorrect to claim that an average 7970 OC draws as much power as a 480 just because someone is willing to accept a 3-4% increase in GPU clock at all costs.

An 1180mhz 7970 draws 239W. A stock 480 draws 270W.
http://www.techpowerup.com/mobile/reviews/HIS/HD_7970_X_Turbo/26.html

And the key takeaway from this thread: if you want to OC to the absolute max, expect power consumption of any ASIC to reach stupid levels of power usage due to leakage galore.

Sorry assumed everyone knew about the 7970's need for power past stock voltages

 

[thilanliyan]

Yes. Welcome to the wonderful world of VC&G.

Who is upset?

All I see is people sharing data.

 

[RussianSensation]

Sorry assumed everyone knew about the 7970's need for power past stock voltages

No worries. From my own experience the sweet spot for Tahiti seems to be 1.175-1.225v on air. Above 1.256v things get sour.

 

[UNhooked]

So finally got a beefier PSU and flashed my GPUs to Ghz Bios with 1.256V.