Miners might target Nvidia Maxwell next

[Erenhardt]

Hmm. That's interesting when the 270X reviewed here:

http://www.anandtech.com/show/7400/the-radeon-r9-280x-review-feat-asus-xfx/20

Is getting 240W during gaming as the total system draw, and 284W in Furmark. So you're getting 130W total system draw. Alrighty... Apparently your Pitcairn card based system is drawing half the power of the Pitcairn system tested by AT, that's quite the accomplishment considering your 7870 is also overclocked.

You are mixing total system power consumption with card power consumption.
Those will most likely differ by a margin of 50-150Watts, based on system configuration.

 

[Zanovar]

Honestly, given the insane efficiency improvements with Maxwell - the real interesting aspect of this story is what happens when GM200 and GM204 are released and the Maxwell uarch is scaled with tons of CUDA cores. Kepler from what I remember scaled very linearly in this respect and there is no reason to expect otherwise from the large die Maxwell parts - given GM107s relative hashing performance, big Maxwell should outright beat AMD across the board in terms of hashrate. Ignoring the implications of the hash per watt of GM107 in terms of how it will directly affect the hashrate of big Maxwell, one would have to essentially wear blinders. GM200 and GM204 will be hashing monsters. It should be mentioned that the directly comparable AMD counterpart to the 750ti in terms of TDP is the R7-250 (65W TDP). The hashrate of that part is far lower than the 750ti.

Now with that stated - I am not a fan of this whatsoever. Personally I don't really care for mining and never have, as funny as this entire situation is i'd rather people just used these GPUs for gaming. I don't have any ill will toward those who do mine, it just isn't my thing. Anyone who's into it, cool. But as for me.....I want a Maxwell based GPU for PC gaming, which is what consumer level Radeons and Geforce cards are designed and marketed for in the first place.

Agreed:thumbsup:

 

[blackened23]

You are mixing total system power consumption with card power consumption.
Those will most likely differ by a margin of 50-150Watts, based on system configuration.

So please tell us how you are measuring your GPU only power consumption with a Kill-a watt. A kill a watt measures total system power draw at the wall. That's what YOU said. You measured with a kill-a-watt @ 130W with your pitcairn based PC total system draw, while AT tested total system draw at 240-280W with their pitcairn based system. It should also be mentioned that this system used by the review is a bare bones test bench system without extraneous fans, HDDs, and all that sort of thing which in the typical system would add to the total system power consumption. It has only what it needs to be reviewed and benchmarked, that's it.

So...the killa-watt - You plug it into the wall and it measures power draw at the wall. You just told us that your pitcairn system uses 130W measured by a kill-a-watt. That means you implied total system power draw - a kill-a-watt cannot measure GPU only power consumption.

Explanation?

 

[Erenhardt]

I think this post confused you a bit:

If it makes you a bit happier, it takes 130Watts at stock voltage and 1200MHz OC,

130 Watts is what Aida said the card uses. System power consumption on a kill-a-watt : 240Watt.
The second part:

Also, I measured it with killawatt

was to answer the question if I'm basing these numbers on aida only.

 

[blackened23]

Yeah. I am confused. First you said that your pitcairn is using 70W while hashing and overclocked at that. Obviously we now know that isn't what was happening, as it's an outright impossibility, case closed I guess.

 

[Erenhardt]

Yeah. I am confused. First you said that your pitcairn is using 70W while hashing and overclocked at that. Obviously we now know that isn't what was happening, case closed I guess.

Well, it does.
Im hashing 400kh/s with single 7870. System power consumption: 170watts while hashing and 100Watts at idle.
Aida says my underclocked and undervolted card takes 60 watts while hashing. System power consumption delta is confirming.
Anyway, that is offtopic now

 

[f1sherman]

Lulz, by that fail logic my 1195 core 1475 GDDR5 290 is only drawing 250-260w.

Thats pretty good sample you got there.
But mine is not bad either

R9 290 hashing at ~ GTX 780 TDP (189W)

Erenhardt... really

 

[Hitman928]

Why are people jumping all over Erenhardt for posting his measurements. . . seriously.

http://www.tomshardware.com/reviews/radeon-r9-270-review-benchmarks,3669-9.html

Shows the 7870 drawing 142 W for GPGPU related tasks. Now, Erenhardt said he undervolted to 1V which is 0.82 of stock value. Now, for CMOS designs, power is proportional to voltage squared so that undervolting would give him ~67 % of stock power or 95 W. He also underclocked to 0.95 of stock giving 90 W because frequency is approximately directly proportional to power. Now, there is also a temperature dependence on power, especially in smaller nodes and when IDC did his testing of CPU's, he found that this temperature dependence could essentially be modeled into the power-voltage dependence as a proportionality to voltage cubed. If we use that model, it would give us 142 * 0.95 * 0.82^3 or 74 W, pretty much what his delta from idle to load was showing. Also take into account variability between card designs and wafers and you have a pretty decent window to play in. Now, these aren't exact figures and most likely the real value will be between the two calculated as IDC's temperatures varied greatly whereas I don't think Erenhardt's would have, but an estimated Power usage of about 75 W for his card is not bad. Yes, it's more than the 60 he claimed initially, but his own kila-watt readings showed that anyway.

If you don't believe something someone is posting, have a nice discussion as to why you disagree. Posting only sarcasm benefits no one.

 

[blackened23]

So those tests done at Tom's are GPU only measurements, not killawatt measurements. As others have mentioned out, eren does not have a method to even test VGA only power draw. Kill a watt does not do this. It only measures total system draw.

So, per his claim, his card uses less power than the 7770 according to those GPU only power draw tests done by Tom's. I also thought he mentioned an overclock and +20% powertune, although I could be wrong. I swear I could have read that somewhere. It's just...fascinating. His card is using so so much less power than every other Pitcairn on the planet. I'm not saying I don't believe him or believe him. I'm just saying it's very interesting that his card is using half the power that everyone elses does. With an overclock and 20% powertune. The golden sample of all golden samples? I suppose.

 

[Hitman928]

Blackened, I suggest you refresh your memory of this thread and read posts more carefully. You have basically completely ignored my post by your last comments and missed many of Erenhardt's.

As far as the kill-a-watt meter goes, it is just used as an approximation as you take the difference between idle and load. Erenhardt used it to try and confirm what Aida was telling him, that's all. For a more detailed visitation on the subject, see my last post.

p.s. All power measurements, even by review sites, are not going to be 100% accurate to your own card/system, they will all differ based on many variables.

 

[Erenhardt]

Thanks for backing me up Hitman, but I don't think he sees a difference between overvolting and undervolting/overclocking and underclocking.

My Kill-a-watt is inaccurate - better use TDP values, like pcper did <facepalm>

Anyways, good to see nvidia catching AMD in kh/s/W. AMD prices were getting silly lately.

 

[ocre]

So those tests done at Tom's are GPU only measurements, not killawatt measurements. As others have mentioned out, eren does not have a method to even test VGA only power draw. Kill a watt does not do this. It only measures total system draw.

So, per his claim, his card uses less power than the 7770 according to those GPU only power draw tests done by Tom's. I also thought he mentioned an overclock and +20% powertune, although I could be wrong. I swear I could have read that somewhere. It's just...fascinating. His card is using so so much less power than every other Pitcairn on the planet. I'm not saying I don't believe him or believe him. I'm just saying it's very interesting that his card is using half the power that everyone elses does. With an overclock and 20% powertune. The golden sample of all golden samples? I suppose.

I will say that i cannot believe such claims. And doesnt matter, there is enough data out there by proper reviewers and reliable sources to see were these cards land. If someone has a completely way off result, i just dont buy it. I mean, even if there is a card out there like that, its not the norm. Its not what anyone should expect.

 

[blackened23]

Indeed. It's one of those things that everyone is thinking but doesn't want to say. I don't disagree with you. I don't buy 70W, not even close, that was my polite way of saying such.

That is of course on top of the fact of not having a method to even test VGA only power consumption. There's simply no way his card is using half the power that every other pitcairn card is using. At 70W, his claim means it is using less than a 7750 would. Not happening, obviously. Clearly his test methods are erroneous.

 

[Erenhardt]

Indeed. It's one of those things that everyone is thinking but doesn't want to say. I don't disagree with you. I don't buy 70W, not even close, that was my polite way of saying such.

Here is a real shocker for you. My 7870 takes as low as 12 Watts!
Clocks and voltage can make wonders...
here is 3rd party chart backing my claims:

 

[blackened23]

Thank you for linking an idle power consumption chart. How enlightening. Truly. Clearly that is relevant for a 100% GPU load power power consumption discussion, of which hashing definitely does push a GPU to the limits.

Not even sure why i'm responding. Pretty sure I was baited into the ultimate dis-ingenuousness and trolldom nonsense claim.

 

[Hitman928]

blackened and ocre,

Let me just ask you one question and see if you can find an answer. Out of all of these samples of cards that clearly show the 7870 using so much more power compared to Erenhardt's, how many of them were measured at 1V and 950 MHz? If you can find that, I believe you will have a point of argument, otherwise, you're just choosing to ignore any posts that counter what you believe, no matter how much evidence is presented.

 

[ocre]

Well he did post that link. There you have it, proof that cannot be contested. The 7870 only uses 12 watts. Case closed
You heard it here first.

But seriously, are the people reading this really gonna let this fly. Complete and total nonsense. For real? Anybody else see the insane level here? What is the world coming to

 

[f1sherman]

I will say that i cannot believe such claims. And doesnt matter, there is enough data out there by proper reviewers and reliable sources to see were these cards land. If someone has a completely way off result, i just dont buy it. I mean, even if there is a card out there like that, its not the norm. Its not what anyone should expect.

Yeah, I'm sorry but I don't think there is anything to discuss really.
Not when someone starts with:

Well, they are almost there. But not quite. (nvidia)

and then goes on to claim that 62W card (750 Ti) is really 100+W and takes up to 140W while gaming ,

but card with typical gaming load of 120W (7870) is mining @70W...because AIDA says so.
Also he forgets that CPU is not idling when mining so according to his AIDA measurements that 7870 should be closer to 50W WHILE MINING
Right, so his 7870 is undervolted. But 750 Ti can't be undervolted?

So if someone is skeptical and even LOL-ing about other ppl power measurements, he better come here with something better than AIDA, his power bill.

And things like:

Also, I measured it with killawatt

 

[el etro]

[/B]

Not really truth. Here you can see that the card takes up to 140W while gaming.

This limit don't represent real max power consumption of the card. But is clearly that 750Ti overrides its TDP in some conditions.

Those figures are so wrong... Can easily hit 950-1000 khash on the 290X while undervolting with custom bios... Also, you won't be anywhere near 294 khash with the GTX 750 Ti with 60W power consumption...

Test don't show power consumption while hashing. There's another test on the web that shows the card overriding the TDP while mining:

The GTX 650 Ti Boost hits 65 FPS on 161W of power. The R9 70 pushes 72 FPS on 172W. The GTX 750 Ti&#8217;s still-impressive 64 FPS come courtesy of just 121W of power consumption. That&#8217;s 98% of the GTX 650 Ti Boost&#8217;s performance for 72% of its power consumption. When we measured power consumption while mining Litecoins, the difference was even more dramatic. Again, the Radeon R9 270 and the GTX 650 Ti Boost both drew ~160W to hit 180KHash/s and 395KHash/s respectively. The GTX 750 Ti, in contrast, mined at a steady 246KHash/s on just 105W.

Diff in power draw of R9 270 and 750Ti at mining is 60W. The discrepancy between gaming and mining power consumption already happen with GCN cards(once mining relies so much on shader power). Compare on image below the discrepancy between peak(peak is the highest power gaming power reading on a set of many power readings) and maximum(that stress the maximum the GPU components can go).

Diference on power consumption for each chip(Only the most efficient SKU per chip):

HD 7790(100w TDP): 103w vs 77w (33%)
HD 7870(175w TDP): 144w vs 115w (25%)
HD 7950(200w TDP): 179w vs 144w (24%)
(note: Bonaire chip has higher compute efficiency against another GCN/Fermi cards)
GTX 660(140w TDP): 138w vs 118w (16%)
GTX 670(170w TDP): 162w vs 152w (6%)

Nvidia knows that their cards will not increase power consumption so high in these conditions because Kelper cards(unlike Fermi cards) don't have extra Shader Power to be well utilized in this kind of tasks. That's why AMD rates their cards with a so high TDP(like Nvidia did with GTX 400/500 cards) and this gives the impression that Nvidia cards are so more power efficient(in the case of Kelper vs GCN).

Back on mining performance, Extremetech data of mining efficiency is:



Remembering R9 270 is a downclocked HD 7870 and haves lower TDP and power consumption than the older card.

 

[SPBHM]

this is probably the best reference available for "mining" power usage, it's no mining but I suppose the load is closer to real than "TDP" or gaming tests

obviously not much point in compare undervolt or OC results with a stock card, I'm sure you can also "optimize" the 750 Ti or any other card, PCper did a strange thing I think, they've used the same "power usage" value (TDP!?) for stock and OC card, is that correct?

 

[3DVagabond]

this is probably the best reference available for "mining" power usage, it's no mining but I suppose the load is closer to real than "TDP" or gaming tests

obviously not much point in compare undervolt or OC results with a stock card, I'm sure you can also "optimize" the 750 Ti or any other card, PCper did a strange thing I think, they've used the same "power usage" value (TDP!?) for stock and OC card, is that correct?

There's no way of knowing if this bench is any more or less accurate at determining mining power than any other. We need to see a bench that is while actually mining. Surely if sites are going to show kh/s they could also show power usage while mining, not simply state the TDP of the 2 cards for calculations. Hasn't any site done this?

 

[SPBHM]

There's no way of knowing if this bench is any more or less accurate at determining mining power than any other. We need to see a bench that is while actually mining. Surely if sites are going to show kh/s they could also show power usage while mining, not simply state the TDP of the 2 cards for calculations. Hasn't any site done this?

I think it's the closest we have, you are right in saying we need actual mining numbers, but this is the best we got, it's pure GPU load, and GPU load for the "Shaders" and memory more than the rest of the GPU probably, like mining!?

and yes, PCper did a poor job in using TDPs and not bothering to actually even use a "kill a watt", the only valid numbers they are showing are the hashrates, forget about the power and efficiency stuff (and we also know the real prices for each card)

 

[el etro]

At 70W, his claim means it is using less than a 7750 would. Not happening, obviously. Clearly his test methods are erroneous.

At the default configurations, Pitcairn tops out Cape Verde on the efficiency chats(And TPU charts shows it too). Only one that matches/wins Pitcairn in this metric is Bonaire.

I'm just saying it's very interesting that his card is using half the power that everyone elses does. With an overclock and 20% powertune. The golden sample of all golden samples? I suppose.

You read wrong. Is the opposite.

There's simply no way his card is using half the power that every other pitcairn card is using. At 70W, his claim means it is using less than a 7750 would. Not happening, obviously. Clearly his test methods are erroneous.

Wait a sec... you actually believe your 7870 is hashing at ~70 Watt

Temperatures influence heavily on power consumption too. If his card fan is set at 100% power consumption can be heavily reduced at this monstrous load that mining is. Combined to a 15% undervolt and 5% lower clocks, maybe he can reach the 70w power consumption at this kind of load.

and then goes on to claim that 62W card (750 Ti) is really 100+W and takes up to 140W while gaming

This 140w peaks don't reflects on anything but dimensioning the PSU for the card. The card will never draw this 140w of the PSU. Look at r9 290x results on the same test:

392 watts power consumption peaks are under the 48A recommended at the 12v rails by AMD. 392 watts will not reflect gaming power consumption of the AMD card.

But 750 Ti can't be undervolted?

Can too.

I think it's the closest we have, you are right in saying we need actual mining numbers, but this is the best we got, it's pure GPU load, and GPU load for the "Shaders" and memory more than the rest of the GPU probably, like mining!?

Is not Shaders more than everything...
A more accurate approach is to say that GPGPU power consumption and performance, compared to in gaming conditions, reflects more the given TDP(especially in AMD's case) and theoretical performance.

 

[MrK6]

Maxwell shows promise but nvidia needs to put out some bigger chips first before they're worth the slot(s) they occupy. That said, the sites that "review" mining potential haven't the first clue about it, if anything they're doing miners a favor by keeping the ignorant out of the way (maybe that's the point? who knows).

 

[Abwx]

Thanks for backing me up Hitman, but I don't think he sees a difference between overvolting and undervolting/overclocking and underclocking.

My Kill-a-watt is inaccurate - better use TDP values, like pcper did <facepalm>

Anyways, good to see nvidia catching AMD in kh/s/W. AMD prices were getting silly lately.

Your numbers are indeed realistic , for thoses who say otrherwise
they can check at hardware.fr the cards comsumption measured
in isolation.

http://www.hardware.fr/news/13568/nvidia-lance-geforce-gtx-750-ti-750-maxwell.html