Nvidia GF100 pulls 280W

[heky]

Has anyone of you seen this article:

http://www.semiaccurate.com/2010/01/17/nvidia-gf100-takes-280w-and-unmanufacturable/

What do you guys think about it?

 

[Easo]

If that is correct, someone we know is far from being "green"... xD

 

[Flyordie]

lolz. 550mm2... poor nvidia. You would think they had learned their lesson with the GF8 series.

 

[Error 404]

I don't really trust much from SemiAccurate, especially considering all the rumors that the author is in AMD's pocket. However, looking at the nVidia PDF, with the 448 SPU core looking likely, nVidia might be in trouble with the GF100.
With that power draw, it will be nigh on impossible to put two on a card, and even if they manage it the price will be insane (high quality power circuitry and lots of RAM chips = expensive card). It looks to be a nice performer, but if nVidia somehow managed to beat ATI on price/performance, ATI is probably able to release something to counter it soon after its released due to the head start they have on nVidia.

I'm hoping that a DX11 game as demanding as Crysis was comes out soon, then we'll get a much more accurate view of how powerful those tesselation SPUs on GF100 are, and how much of a performance hit it takes running full tesselation.

 

[bobzilla2009]

280W doesn't seem that unreasonable to me considering the apparent heat issues. Still, i'd rather see some nice reviews from TPU or hardware canucks first.

 

[Thrackan]

I don't think 280w is a weird number, considering everything else involved.
About dual GPU cards: they might fabricate another dual-PCB card, as they did with the 295's. That way they can fit both on.

Though I hope that IF this number is true, the cooling will be adequate... It'll be damn hot.

 

[air_ii]

Charlie may or may not be in AMD's pocket, but he surely hates nVidia all the way.

About the dual card. If the 280W is correct, they would have a hard time getting PCI-E approval (300W limit) with it.

 

[bobzilla2009]

probably end up liquid cooled if it were that hot. There's always a way. Eventually nvidia and ati will move on from air cooling in their quest for performance anyway.

Maybe they'll both start pushing for a new type of pci slot too, that allows for a higher power cap

 

[Thrackan]

probably end up liquid cooled if it were that hot. There's always a way. Eventually nvidia and ati will move on from air cooling in their quest for performance anyway.

Maybe they'll both start pushing for a new type of pci slot too, that allows for a higher power cap

Well the 58XX's heat output proves that air cooling on performance cards is still very much possible... And the wattage proves it can be done with "normal" power consumption too.

I might be bustin' green a little here, but the only company making energy- and heat-efficient cards right now is ATI.

 

[OnBoard]

Well my GTX 280 is already 249W, what's 31W more for almost double the power?

http://www.techpowerup.com/reviews/Powercolor/HD_5850_PCS_Plus/27.html

If it was that A2 silicon and not production A3, then it might go down again and they might tweak the voltage. I'll say it's just fine untill W1zzard puts it in his bench and proves me right/wrong

 

[FordGT90Concept]

As a point of reference, the 5870 is 188w but it is equiped to draw 250w.

 

[HalfAHertz]

Charlie usually spreads a lot of BS, but this time he may be on to something. Now from my understanding the 55nm GTX 285 is ~470mm^2. If energy consumption scaled linearly, which I higly doubt that it does, that means GT100 would consume ~17% more than the GTX285 (which consumes ~215W according to TPU) and that would put it in the ballpark of ~250W. Considering the numerous problems TSMC has had with it's 40nm production process, we can conduct that current leakage on the GT100 will be worst or equal to that of the GTX285.

Plus the CES models were not final consumer products and neither were the drivers, so I can only imagine that the power management was broken. On top of that Charlie has no first-hand experience with those cards whatsoever, so it's quite clear that this is just a rumour.

Still I'd say that 280W is the absolute worst case scenario, and a more realistic one would be 225-250W, which isn't that far away from a HD5870, if the performance improvement is indeed as great as in the recently released slides...

 

[air_ii]

Well my GTX 280 is already 249W, what's 31W more for almost double the power?

http://www.techpowerup.com/reviews/Powercolor/HD_5850_PCS_Plus/27.html

If it was that A2 silicon and not production A3, then it might go down again and they might tweak the voltage. I'll say it's just fine untill W1zzard puts it in his bench and proves me right/wrong

The 249W in W1zzard/s review is maximum power draw, which is higher than the TDP (it's common for GPU manufacturers). Charlie is referring to Fermi's TDP.

Also - as explained by knowledgable people - metal respins (A2 -> A3) are not to reduce power consumptions (not significantly, that is).

As a point of reference, the 5870 is 188w but it is equiped to draw 250w.

225W . And again, 188 is TDP, it draws more under FurMark (which is what W1zzard uses for maximum power consumption).

Charlie usually spreads a lot of BS, but this time he may be on to something. Now from my understanding the 55nm GTX 285 is ~470mm^2. If energy consumption scaled linearly, which I higly doubt that it does, that means GT100 would consume ~17% more than the GTX285 (which consumes ~215W according to TPU) and that would put it in the ballpark of ~250W. Considering the numerous problems TSMC has had with it's 40nm production process, we can conduct that current leakage on the GT100 will be worst or equal to that of the GTX285.

Still I'd say that 280W is the absolute worst case scenario, and a more realistic one would be 225-250W, which isn't that far away from a HD5870, if the performance improvement is indeed as great as in the recently released slides...

Power consumption depends on many things - voltage, clocks, complexity of the chip, etc. I don't think you can do the math that easily .

 

[HalfAHertz]

The 249W in W1zzard/s review is maximum power draw, which is higher than the TDP (it's common for GPU manufacturers). Charlie is referring to Fermi's TDP.

Also - as explained by knowledgable people - metal respins (A2 -> A3) are not to reduce power consumptions (not significantly, that is).



225W . And again, 188 is TDP, it draws more under FurMark (which is what W1zzard uses for maximum power consumption).



Power consumption depends on many things - voltage, clocks, complexity of the chip, etc. I don't think you can do the math that easily .

Neither do I. I was jst saying that Charlie's numbers are highly unlikely...

 

[Sasqui]

If it's true - good news for PSU and heatsink/waterblock manufacturers

Perhaps they'll be kind and make the mounting compatible with a 1366 waterblock

 

[Super XP]

Talk about jacking up your power bill.

 

[newtekie1]

Well the 58XX's heat output proves that air cooling on performance cards is still very much possible... And the wattage proves it can be done with "normal" power consumption too.

I might be bustin' green a little here, but the only company making energy- and heat-efficient cards right now is ATI.

They may be energy and heat efficient, but they will probably fail to compete with nVidia's top end performance, and area where people really don't care about energy and heat.

I'm sure the cards that compete with ATi products will do so on all fronts.

Oh, and Charlie is almost always full of shit...

 

[Flyordie]

lol, If Project X comes through in May 2010.. Nvidia will be running in circles wondering what to do...

A 587mm2 monster from ATI @ 40nm. Expected to be produced using 28nm though.

 

[air_ii]

lol, If Project X comes through in May 2010.. Nvidia will be running in circles wondering what to do...

A 587mm2 monster from ATI @ 40nm. Expected to be produced using 28nm though.

I have no idea what you're talking about

 

[Fishymachine]

280 is surely bull,I would bet that the TDP will be over the 225W limit of dual 6 pin,still 280...maybe it's the max wattage the card can support before frying(using volt mods)

 

[3volvedcombat]

More and more im seeing this puffs of semi accurate and many other threads and everything, so on and so on........

Fermi or gt100 is looking looser and looser every day.

could 40nm really make the card that efficent. I mean, its 512 stream processors here. I couldnt imagine a process from 55nm to 40nm is going to drop wattages in HALF. That would be cool to see though.

Im just saying this as in 285 uses 150+ watts at full load.
This has 512 processors, its like haveing a pair of 285's in sli and a gt 230 all in sli, but on a single gpu.

But i cant just say its loose yet, because if they sell the 512 big boy for 450-500 dollars consider it sold

 

[Deleted member 3]

The 249W in W1zzard/s review is maximum power draw, which is higher than the TDP (it's common for GPU manufacturers).

TPU reviews measure the whole system. Not just the graphics card.

 

[phanbuey]

it possible that early engneering samples could pull that much, but it has no bearing on whether the finished product will pull that much. Especially now that TSMC is claiming that they fixed their yield issus.

better manufacturing process=less leakage.


plus NV wouldnt release a cart that pulls 280W by itself, they would sooner release an underclocked/undervolted version of the card than something so hot.

Im thinking it will have the power/thermal properties of the 2900XT which i think was a 225W draw.

 

[Super XP]

it possible that early engneering samples could pull that much, but it has no bearing on whether the finished product will pull that much. Especially now that TSMC is claiming that they fixed their yield issus.

better manufacturing process=less leakage.


plus NV wouldnt release a cart that pulls 280W by itself, they would sooner release an underclocked/undervolted version of the card than something so hot.

Im thinking it will have the power/thermal properties of the 2900XT which i think was a 225W draw.

If 280W is true, then Fermi will be delayed until they reach much bellow that with the yields they are looking for. All I know is the sooner they release those cards the sooner we can all enjoy the reduced priced HD 5800's

 

[HalfAHertz]

TPU reviews measure the whole system. Not just the graphics card.

You mean to tell me that W1z's entire system, running a hd4550 consumes the whooping 19W?!