The GT300/Fermi Thread

[Zed_X]

GF100 has more ROPs then G200 and less TMUs then G200, but that is not a problem, has new units - raster a geometry units

[Olivon]

One thing is sure. GF100/Fermi is late

[Blacky]

they are not? what makes you think that?
do you remember how jensen told tsmc to hurry up with 28nm?
and after tsmc faced problems with 28nm, jensen mentioned how they might work with GF instead, who are agressively ramping up 28nm...

nvidia knows they need 28nm... they could in theory completely screw up 40nm and not sell any 40nm fermi parts... it would hurt them... a lot... but they could survive that without too serious problems... 28nm is where it gets serious... if they cant get 28nm working right, THEN nvidia might bleed so bad it will take a while for them to recover, if ever...

I could be wrong and even if they are working on it, prolly nvidia is at full-force right now, but I'm sure of something, we won't see any high-end parts based in 28nm anytime soon after all this delays etc. they have been experiencing trying to adapt a such arch into 40nm, to meet the specs as they were original designed/presented, again I could be wrong but nvidia is known by pulling tricks out of their sleeves, not quite with fermi though, also as we all know nvidia lately is slower making die shrink transitions

[trinibwoy]

Nvidia is stretched pretty thin right now with Fermi and Tegra. I don't think they can maintain this pace for much longer, eventually they'll settle down and derivatives will come sooner since they won't be trying to revolutionize the industry with every new architecture.

[neliz]

GF100 has more ROPs then G200 and less TMUs then G200, but that is not a problem, has new units - raster a geometry units

Yeah, that's the official spin "they're much more efficient" bla bla bla.. it's still BS because the units are THERE, why disable them?

Another hint, when Huddy says he expects AMD to have the fastest card for most of 2010, he's not talking about AMD refreshes.

[Aerwidh]

Another hint, when Huddy says he expects AMD to have the fastest card for most of 2010, he's not talking about AMD refreshes.

Sounds like 5970 is going to have plenty of time to be king in the spotlight, then. Boring

[JohnJohn]

Another hint, when Huddy says he expects AMD to have the fastest card for most of 2010, he's not talking about AMD refreshes.

gf100 launch on march, availability on may/june, dual fermi june/july? & fully loaded gf100 on september

Or not so apocalyptic?

[Newblar]

Sounds like 5970 is going to have plenty of time to be king in the spotlight, then. Boring

while the rest of ati's 5xxx destroys nvidia price/perf
ive seen hardcore nvidia fanboys switch to ati recently... bad sign for nvidia

[Manicdan]

while the rest of ati's 5xxx destroys nvidia price/perf
ive seen hardcore nvidia fanboys switch to ati recently... bad sign for nvidia

i thought the 5xxx was worse than any performance comparable 4xxx in price?

[LordEC911]

LordEC911, on a related note:Random Hungarian Forum
These intertubes things are amazing. I can sleep and attend secret conferences at the same time! Unless Google thinks that it translates to me actually being on the good side of the FUDwar.

Wow... that is ridiculous.
You give all the info you can that is deemed credible, for both sides.
Just because most of the previous news on GF100/Fermi was bad doesn't mean you are trying to discredit Nvidia, the blame lies soley on Nvidia for that.

[neliz]

Wow... that is ridiculous.
You give all the info you can that is deemed credible, for both sides.
Just because most of the previous news on GF100/Fermi was bad doesn't mean you are trying to discredit Nvidia, the blame lies soley on Nvidia for that.

If I had (all) the specs for GF100, I'd post them straight up, just like I did with any info before it. Thing is, AMD seems to "not mind" and they actually ask random people to post parts of their specs. NV really seems to be uptight about it's info. That's why I only get the codenames like "Optimus" from their presentations a few days before their embargoed outing.

But that's just the difference in companies. A marketing driven closed society versus a business driven open one.

[trinibwoy]

Is post #905 an example of posting stuff straight up?

[neliz]

Is post #905 an example of posting stuff straight up?

y-y-yes?

[trinibwoy]

Ok, then. Just checking

[ulticool]

So basically it won't be out until summer, then it won't be in stock until fall then it will be time to shrink it already to keep up with ATI/AMD.

Mmmhmm

[H2O]

So basically it won't be out until summer, then it won't be in stock until fall then it will be time to shrink it already to keep up with ATI/AMD.

Mmmhmm

It dosen't have to mean that GF100 itself is delayed. Nvidia may have delayed that dual GPU version, and from the estimates that we have seen, ATI should retain the performance crown with the HD5970 even after the "GTX380/GTX360" release.

[neliz]

It dosen't have to mean that GF100 itself is delayed. Nvidia may have delayed that dual GPU version, and from the estimates that we have seen, ATI should retain the performance crown with the HD5970 even after the "GTX380/GTX360" release.

It does mean GF100 availability has been delayed. I'm saying that MP doesn't start for quite some time, perhaps hinting at an A4 revision.

Now.. Fermi2 otoh.. that might be closer than everyone thinks.

[xVeinx]

Now.. Fermi2 otoh.. that might be closer than everyone thinks.

That's going to depend on when the 28nm process is available to nvidia. The design will be ready, but they are still susceptible to delays per the fabs.

[annihilat0r]

I think Fermi design is just crying for a shrink, if what I am hearing is correct.

[Jamesrt2004]

i thought the 5xxx was worse than any performance comparable 4xxx in price?

well ATI seem to have headroom in that... i mean even on their slides they had msot stuff priced 50~100$$$ less plus im sure they could drop it more then that.... so I think they will jsut drop prices to kill the price/perf like last round

That's going to depend on when the 28nm process is available to nvidia. The design will be ready, but they are still susceptible to delays per the fabs.

as they use tsmc it'll will be 32nm right 28nm is GF and I thought nvidia said they wont use them?




is everyone forgetting ATI's supposed launch in september~october of a new complete arch... isn't that kind of bad for DUAL fermi if that seems to be hinting coming out around then :/.... hell if ati get their ass in to gear they may even be able to have it as a tool to take the thunder away from them....


saying that I hope nvidia's fermi suprises us atleast a little bit cos last round they got beat pretty bad(price/perf wise).... and this time it seems worse..... :/

[Florinmocanu]

I think it makes sense why ati is ahead right now.

Smaller gpus are easier to design, easier to produce, easier to reach expected speeds since they are less hot. And yields are better, way better.

I really don't understand why Nvidia is still going on with monolithic huge designs, it makes them loose ground to the competition, it's pretty stupid.

Even if they get at some point the halo with top of the line products, if you don't have the perf/price ratio right and you cannot invade every segment of the market with products because you cannot shrink the design, then in the long term you will loose more than you gain.

[Jamesrt2004]

I think it makes sense why ati is ahead right now.

Smaller gpus are easier to design, easier to produce, easier to reach expected speeds since they are less hot. And yields are better, way better.

I really don't understand why Nvidia is still going on with monolithic huge designs, it makes them loose ground to the competition, it's pretty stupid.

Even if they get at some point the halo with top of the line products, if you don't have the perf/price ratio right and you cannot invade every segment of the market with products because you cannot shrink the design, then in the long term you will loose more than you gain.

yeah but 5870 is pretty monilithic... i thinks it's just random annoying crap thats keeping them back like solving one thign leads to another thing... :/

rest i agree with... I think they will end up renaming again.. to stay competitive in the lower segments

[Kuroimaho]

as they use tsmc it'll will be 32nm right 28nm is GF and I thought nvidia said they wont use them?

TSMC drops 32nm in favor of 28nm

[Florinmocanu]

yeah but 5870 is pretty monilithic... i thinks it's just random annoying crap thats keeping them back like solving one thign leads to another thing... :/

rest i agree with... I think they will end up renaming again.. to stay competitive in the lower segments

well, RV 870 at 330mm2 is a medium sized gpu, i would say over 400mm and more is where the real fun/problems begin, and at over 500 mm, fermi is big, one of the biggest ever made.

[Jamesrt2004]

TSMC drops 32nm in favor of 28nm

Still a rumour... plus they showd things at 32nm and saying it's on track so I personally don't believe it but as theres obv' some information out there i'll happily eat my words if wrong

well, RV 870 at 330mm2 is a medium sized gpu, i would say over 400mm and more is where the real fun/problems begin, and at over 500 mm, fermi is big, one of the biggest ever made.

ok fair point :P i'm just used to small-er chips from the last couple series I guess

but I agree that going monolithic although having advantages has a pretty heft(huge) downside, I think ati's approach has been much better :/ ok so this round their card was power hungry but as it was only a stop gap until the new arch we cant really complain too much....

I think fermi will show its actual power on the 32/28nm node's (w/e they choose to go with) I just hope it's sooner then later.... im not sure how far tsmc has got.. but we know GF seems to have some pretty complicated wafer's on the go that DON'T look like tests so they are obviously pretty easily on track.

TSMC doesn't seem as vocal.... apart from they showed that 32nm wafer(but I don't know much about it as i havent seen pics of it) but i hope it's pretty advanced/more advanced then GF's 28nm tech as nvidia could be helped quite a bit if they can get the shrink out around the ATI launch (as long as that isn't plagued as much as nvidia's one is/was)