GeForce GTX460 Final Specs

[Picao84]

It seems the first iteration of GF104 will really have 336 CUDA Cores, with clocks a little higher than GTX465, if the following image is not shoped. Performance is said to be around GTX465 levels, with good overclocking potential. Power consumption is also said to be about 150W, which is IMO good (not excellent, though) considering that GTX465 consumers around 215W.

[Zonker]

Nice looking card. Not sure about 768mb though. Given ati has owned this generation, developers might fairly assume anyone running mid to high settings on future games has 1gb of vram min.

[Picao84]

Nice looking card. Not sure about 768mb though. Given ati has owned this generation, developers might fairly assume anyone running mid to high settings on future games has 1gb of vram min.

I dont think it would not be too much of a hassle improve it to 1,5GB... Im sure some brands would do it.
Keep on the memory issues, would that memory clock mean that whatever problems nVIDIA is having with the memory controller in Fermi were not resolved yet?

[-Boris-]

If the amount of cores and the power requirement is true, a dual GPU version is possible.

How many times must nVidia do the same mistake with a single humongous core and correct it with two cooler ones ATi style?

[Picao84]

If the amount of cores and the power requirement is true, a dual GPU version is possible.

How many times must nVidia do the same mistake with a single humongous core and correct it with two cooler ones ATi style?

Well, I have a feeling they will keep going on with the Monolythical die strategy on the next node. Maybe they will go with smaller dies, two generations from now...

Also there is supposed to be a 384 core version, with performance closer to GTX470, but with TDP of 180W. A Dual GPU version would be formed by something in the middle of these two versions IMO. Because probably Dual GTX460 would not be enough to beat a Radeon 5970.. or maybe yes, seeing how good scaling Fermi series have in SLI..

Some people are trying to compare GTX460 to Radeon 5770, and saying TDP is too high. But GF104 is said to compete against 5830 and 5850. GF106 will be the competition for 5770.

[jasonelmore]

yeah i wish it had 1gb or ram at least. looks good otherwise, if true

[prznar1]

yeah i wish it had 1gb or ram at least. looks good otherwise, if true

some words are saying that nvidia is planning something like gtx468 with 1gb ram 256bit bus line and 384 CUDA cores gf104 based arch card.

[SKYMTL]

NVIDIA's issue is that their memory controllers are tied to the cache and ROP arrays. If they want to differentiate lower-end cards from one-another they need to scale back the number of ROPs which will also result in odd memory sizes.

[XSAlliN]

Well, GTX 4xx is mostly "brute force" - which could mean they're waiting for ATi 6xxx, if they have some ace up their sleeve... Or, maybe that's the best they could do and they finally reached the limit. If that is the case, this could be real interesting for the years to come... Finally there's some hope for better games on PC.

[grimREEFER]

they can do dual gpu with it, but it wouldnt outperform a 5970 i dont think

[Florinmocanu]

if atm a 465 cannot outperform a 5850, how can a dual GF 104, who seems to be somewhere in the neighborhood of a dual 465, defeat a 5970, who is practically a 5870 with 5850 freq, so each gpu in 5970 has higher performance than a 5850.

Not to even consider the gargantuan 5970 4GB OC edition, those VGAs are in their own realm of performance.

[Picao84]

if atm a 465 cannot outperform a 5850, how can a dual GF 104, who seems to be somewhere in the neighborhood of a dual 465, defeat a 5970, who is practically a 5870 with 5850 freq, so each gpu in 5970 has higher performance than a 5850.

Not to even consider the gargantuan 5970 4GB OC edition, those VGAs are in their own realm of performance.

The FULL GF104 is expected to have 384 "cuda cores" with a 256 bit bus, with performance around GTX470 level. With that in mind, it could come close to it yes.

[BenchZowner]

If they overclock well... if SLI scales better than CrossFire... if Jessica Alba was my gf... if.... if... if...
I don't think that it's impossible to make a dual GF-1xx based card with equal or just a tad better performance than a 5970.

For sure ? Not, we'll find out sooner or later.

[Caparroz]

The FULL GF104 is expected to have 384 "cuda cores" with a 256 bit bus, with performance around GTX470 level. With that in mind, it could come close to it yes.

You expect it, amirite?

Nvidia would have to implement MAJOR optimizations in gf104 for a theoretical card w/ 384sp/less ROPs and texture units/256bit perform at the same level with a 448sp/320bit with only a tiny clock advantage.

And why Nvidia would wait to launch a full-blown gf104 when they're already very late to the game? It just doesn't make any sense, even for Nvidia.

[SKYMTL]

[B]
And why Nvidia would wait to launch a full-blown gf104 when they're already very late to the game? It just doesn't make any sense, even for Nvidia.

Because KitGuru said so.

/sarcasm

[Picao84]

You expect it, amirite?

Nvidia would have to implement MAJOR optimizations in gf104 for a theoretical card w/ 384sp/less ROPs and texture units/256bit perform at the same level with a 448sp/320bit with only a tiny clock advantage.

And why Nvidia would wait to launch a full blown gf104 when they're already very late to the game? It just doesn't make any sense, even for Nvidia.

GF104 is supposed to have more Texture Units than GTX470, with full 64 (4 more than even GTX480) TMUs, thats why. GF100 looks to be indered really badly by lack of texture units. I dont think it would need MAJOR optimizations. More TMU's and higher clocked speed would be enough. It was already demonstrated that Fermi architecture reacts good to higher clocks.

About not lauching.. How about not canibalizing sales of current GF100 products? Sales seem already bad enough.. now imagine with a better and cheaper product..

[Picao84]

Because KitGuru said so.

/sarcasm

Where did i mention Kit Guru anywhere, excuse me?

[bamtan2]

if atm a 465 cannot outperform a 5850, how can a dual GF 104

your problem is you have a narrow understanding of the situation.

just because gf104 is in gtx460 doesnt mean gtx460 is as fast as gf104 can go. 480>470>465 and they are all the same chip, hello???

pretend they save the fastest gf104 chips. and pretend those chips are as fast as gtx 470. guess what happens then. gtx490?

[SKYMTL]

GF104 is supposed to have more Texture Units than GTX470, with full 64 (4 more than even GTX480) TMUs, thats why. GF100 looks to be indered really badly by lack of texture units.

That's impossible without a fundamental and total redesign of the architecture itself. In the current architecture the number of texture units is directly tied to the number of CUDA cores so you can rule that out right away.

Where did i mention Kit Guru anywhere, excuse me?

I was attempting to be facetious. Guess I failed.

[zanzabar]

that thing should beat the 465 and be like half the wattage, i would have expected something atleast at 800mhz though but that might be saved for a 475 or something

[Caparroz]

GF104 is supposed to have more Texture Units than GTX470, with full 64 (4 more than even GTX480) TMUs (...)

According to whom? Would love to see the source of that (and the rest of) information.

FWIW, all we (kinda of) know is that Nvidia will launch a chip named gf104 with 300'ish cuda cores and ~192/256-bit bus. The rest is just speculation. Although we can be pretty sure that the ratio between cuda cores/tmu's/rop's won't change for a very loooong time like SKYMTL said.

[Aerwidh]

just because gf104 is in gtx460 doesnt mean gtx460 is as fast as gf104 can go. 480>470>465 and they are all the same chip, hello???

pretend they save the fastest gf104 chips. and pretend those chips are as fast as gtx 470. guess what happens then. gtx490?

No offense, but if the GF104 chip was so capable, why would they not just release a more high end card based on that chip directly, instead of releasing something around the 465?
With its smaller size, the GF104 should be cheaper to manufacture, so it would have made far more sense to EOL any GF100 part with similar performance and have the GF104 part replace it, thereby making more money.
They didn't release the 465 before the 470 and 480, did they? Why would they suddenly start making products the other way around now?

We can pretend anything, but that doesn't make it true, or even likely.

[zanzabar]

i dont think that NV was working on the gf104 right away and i dont think that they expected to have to deal with the TSMC 40nm process requiring extra power trenches. if u look back at the original gf100 info even the tesla (i dont think that they would intentionally BS for a server part) they were saying 512shaders, 1GHZ, under 225W. so if they got that the gf104 would be crap in comparison. what they got was a power hungry wattage limited card that did not clock well. if we assume that the gf104 will clock to 900mhz like the gf100 will under liquid then i would alse expect the gf104 to beat the 470 and maybe come near the 480 and in sli should make the 490 (if its 2x470) look like a bad joke.

[Caparroz]

No offense, but if the GF104 chip was so capable, why would they not just release a more high end card based on that chip directly, instead of releasing something around the 465?

I know this is a rethorical question but the anwser is simple: such a miracle GPU doesn't exist, period. All these wild rumors and opinions are just wishful thinking, disregarding even the most basic logical reasoning.

Hopefully, gf104 will fill the huge gap between 5770 and 5850 for a nice price (much closer to the 5850, I hope) and that's bloody nice because that's what the market needs and consumers are craving for.

[Florinmocanu]

your problem is you have a narrow understanding of the situation.

just because gf104 is in gtx460 doesnt mean gtx460 is as fast as gf104 can go. 480>470>465 and they are all the same chip, hello???

pretend they save the fastest gf104 chips. and pretend those chips are as fast as gtx 470. guess what happens then. gtx490?

your problem is you don't bring any source/proofs when assuming GF104 will be some uber/mega chip.

For all we know from rumours, GF 104 will not be a high-end chip, but mainstream.

And it would not make sense for nvidia to design yet another uber high end chip, this needs to be cheap and small, decent power etc..., so they can start making money from the biggest sales sector, mid-range and low-end GPUs.