55nm GT200 (GT200-400) on the way?

[zerazax]

Nvidias own PR slides list a 236W TDP but that doesn't mean it uses that much.

http://www.hardwarecanucks.com/forum...review-20.html

At least they tell us their methodolgy

[eleeter]

GTX 280 TDP is 236Watts but this seems to be inflated, as many reviews show a lot less.

I think it's because most games don't stress the GPU completely. The few GTX's that had overheating problems would skyrocket to 105 degrees before shutting down, but only in certain games and in Vantage I think did it as well. So that seems like the GPU is often not 100% stressed, maybe due to CPU bottlenecks.

[zerazax]

Well look at the hardwarecanucks review because they run the tests by stressing the GPU's on the same benchmark over and over (the only way to eliminate multiple variables).

I'm against reviews that test power usage in cases where varying situations can occur, such as in-game. Different cards will react differently to situations in a game and so power loads will vary because GPU's aren't being equally loaded. The only way to test is if you run the same benchmark over and over that stress each GPU the same amount of load and compare the relative power usage.

[gojirasan]

maybe you should stick to your own advice?

Whoosh!

[Macadamia]

Rossie O'Donell in a sauna sweting kind of hot....?

You did not just say that

[xlink]

I'd heard rumors that nVDA developed the chip for borh 65nm AND 55nm

it won't be months, it will be weeks, 9800GTX+ is 55nm supposedly...

[Anemone]

Could also check 280 power consumption via SLI vs single card consumption, which if done right subtracted should leave the consumption of a single card remaining. However, that won't give TDP of course.

[Barys]

I'd heard rumors that nVDA developed the chip for borh 65nm AND 55nm

it won't be months, it will be weeks, 9800GTX+ is 55nm supposedly...

Source? And are you talking about GT200@55nm? Because that`s what NVIDIA needs the most.

[perkam]

Whoosh!

Don't you mean...

Skadoosh!

:p

Pekram

[xlink]

Source? And are you talking about GT200@55nm? Because that`s what NVIDIA needs the most.

somewhere on XS or OCN... too lazy to dig it up.

[Macadamia]

Won't the 55nm version be GT200b-400?

This is essentially, an Ultra chip for stable running higher clocked cards.

[gojirasan]

This is essentially, an Ultra chip for stable running higher clocked cards.

Do you work for Nvidia or are you psychic? Just curious. I am so tired of people making predictions without citing any kind of source.

[Macadamia]

Do you work for Nvidia or are you psychic? Just curious. I am so tired of people making predictions without citing any kind of source.

9600GSO/8800GS - G92-150
8800GT- G92-200
8800GTS- G92-400
9800GTX- G92-420

Guess what? I'm sick of hypocritical raving people that call out others without even looking ANYTHING up. Happy succeding Sr7 and his fate though.

[gojirasan]

Oh. The G92. I thought we were talking about the GTX 280 ultra or GX2 or some other unannounced product that none of us without any inside info have any focking clue about. Obviously, since you haven't even claimed any inside info and haven't revealed your source for what the GTX code names mean I would have to conclude that you are full of it. I wish you idiots would at least mention that what you are saying is nothing but idle speculation.

[AliG]

the gtx280 is about the same as the 2900xt i think, and thats around 160W, right? i wonder if the ultra might be the first to break the 200W barrier... and then larrabee will break the 250W barrier... insane... just when cpus start to reach reasonable TDPs gpus double the TDP of the top end cpus

hard to tell, i bet even nvidia doesnt know... even if they can make some cards, it doesnt mean they can actually make enough to make money with them. same as with the current gpus, they can make them, a few per month, sure... but do they actually make money selling gtx cards? i dont think so...

Lol intel will probably end up losing money with larabee because they'll have to give everyone a pci_e 3.0 mobo and 5 kw psu to go with it just get it to post, and man once you start gaming....

[annihilat0r]

Nvidia has yield problems in the current version.

Source? Or are you just talking out of your arse?

How can you manage to quote the wrong person for the wrong message? You just click QUOTE. I wasn't the one to say that. It was Anemone who said that.

And yeah, it is correct. NVidia's GT200-related yield problems were in the rumors even before the launch, we even had figures (a single GT200 gpu costs around $150 to manufacture for NVidia).

[gojirasan]

How can you manage to quote the wrong person for the wrong message? You just click QUOTE. I wasn't the one to say that. It was Anemone who said that.

Ouch. I'm sorry annihilat0r. I better go edit that. I don't usually click quote to quote because I am usually responding to just a small part of a post.

Those alleged "yield problems" are not based on anything factual. They're just rumors. Is there any actual hard data on Nvidia's costs? You know like actual evidence or at least a first hand account of it? I mean it's not like there has never been a widespread rumor that was later shown to be false. Although in this case I just don't see how we are ever going to get any hard data. Why would Nvidia publish their yields whether they are worse than expected or better than expected. I see speculation about Nvidia's yields to be fairly pointless. Either they will lower prices or they won't. Still it bothers me when people claim to know something that they couldn't possibly know.

[Seraphiel]

Does NVIDIA really pay the full cost of bad dies? I presume, that when a company orders a chip from a fab they don't own, the cost of bad dies is not their fault and thus loss of money completely. TMSC are the ones that buys the waffers and produces the dies, not NVIDIA. If TMSC accepts the job and whatever deal has been made, they have more responsibility for bad dies and errored products than NVIDIA.

This is based upon usual business practice as I see it, but anything computer related is seldom normal business and other rules usually apply.

I think the figures mentioned here about cost is totally off for various reasons, but the some of the yield numbers may or may not... This is just confusing to me.

Just wondering.

[zerazax]

Well if that were true, then say a company orders a ridiculously sized chips and only pays for the good ones... how would TSMC make any money if yields were terrible?

I'm sure they have some contract down for both sides. Because after all, both sides are responsible for yields. TSMC for making sure the process is low in defect / area and also Nvidia for making sure the design is feasible for reasonable yields. Likeliest what happens is that TSMC sets a price for a wafer and the process (often a bulk discount as well given that Nvidia is buying many of these) probably with contract terms to ensure that certain defect/cm^2 is below a certain level and so on, but Nvidia is responsible for making sure their design will produce acceptable yields.

After all, if you only had to pay for what was good, then a company could create a ridiculous chip where a 300mm wafer produces one good chip that the company would pay forwhile TSMC eats the rest of the cost of the wafer. I doubt that's what happens or else TSMC would be in big financial trouble.

[Seraphiel]

Well if that were true, then say a company orders a ridiculously sized chips and only pays for the good ones... how would TSMC make any money if yields were terrible?

I'm sure they have some contract down for both sides. Because after all, both sides are responsible for yields. TSMC for making sure the process is low in defect / area and also Nvidia for making sure the design is feasible for reasonable yields. Likeliest what happens is that TSMC sets a price for a wafer and the process (often a bulk discount as well given that Nvidia is buying many of these) probably with contract terms to ensure that certain defect/cm^2 is below a certain level and so on, but Nvidia is responsible for making sure their design will produce acceptable yields.

After all, if you only had to pay for what was good, then a company could create a ridiculous chip where a 300mm wafer produces one good chip that the company would pay forwhile TSMC eats the rest of the cost of the wafer. I doubt that's what happens or else TSMC would be in big financial trouble.

That makes much sense, and thanks for replying.

Still, TSMC can refuse the job and aren't forced to anything, and I have always thought there are initial trials before any contract is sealed? Tests are being made for both, and a contract is made based upon the results of those.

I have no clue about all of this, of course

[Luka_Aveiro]

Don't you mean...

Skadoosh!

:p

Pekram

GOTCHA!

[mascaras]

somewhere on XS or OCN... too lazy to dig it up.

maybe this :

Seems like GT200 got two versions. One is 65nm and the other is 55nm.

Probably 65nm is for first wave of GT200 cards and 55nm is for later batches.

http://we.pcinlife.com/thread-929091-1-1.html



http://forums.vr-zone.com/showpost.p...3&postcount=33

regards

[jam2k]

Wow if they keep overcloking that heat monster, it'll explode ...

[DeathReborn]

Well look at the hardwarecanucks review because they run the tests by stressing the GPU's on the same benchmark over and over (the only way to eliminate multiple variables).

I'm against reviews that test power usage in cases where varying situations can occur, such as in-game. Different cards will react differently to situations in a game and so power loads will vary because GPU's aren't being equally loaded. The only way to test is if you run the same benchmark over and over that stress each GPU the same amount of load and compare the relative power usage.

I prefer seeing what sort of power the cards will draw while doing things I atually do & not in a lab environment (line conditioners etc). Saying that it could consume 500W and if the performance was there I wouldn't care, i'd just slap some Phase on it.

[dengyong]

Wow if they keep overcloking that heat monster, it'll explode ...

Do you own one?