GeForce Kepler 104 and 100 GPU Specifications Compiled

[Bjorn_Of_Iceland]

Once we compare any Fermi card against anoother Fermi card, scaling is near perfect though and this is most likely going to be inherited to Kepler.

Yes.. precisely what they were aiming for in the first place. It was built with a mindset on what would be more conventient to them in terms of releasing a card in a rapid manner for the target demograph

 

[Benetanegia]

Yes.. precisely what they were aiming for in the first place. It was built with a mindset on what would be more conventient to them in terms of releasing a card in a rapid manner for the target demograph

Yup exactly.

Something worth mentioning is that Kepler seems to be made of 64 SP wide SMs*, as oposed to GF100/110's 32SP or GF104's 48 SPs.

This means better area performance, and better perf/watt, following the improvement seen in this regard between GF100/110 and GF104/GF114.

* Assumption I make based on the fact that the SKUs are 768 SP and 640 SPs, with the most probable config being 2x64 SP clusters disabled. 128 SP wide clusters are not likely, and 4 clusters disabled are definitive no.

Also 768/64 = 12 clusters -> 12 x 8 TMU == 96 TMU -> 2 disbled == 80 TMU.

 

[AsRock]

I call bullshit AMD will flop once again just like there cpus did.

I say your talking BS as you know BS how they are yet lol. And ATI GPU's have been improving not getting worse.

 

[Over_Lord]

I said this would happen in the 79xx thread.

No way will this impact HD7900.

HD7900 launches Jan 9th.

GK104, somewhere in Q2 2012. And it's already been reported GK100 is towards end of 2012.

So how will this even scratch HD7900?

 

[the54thvoid]

No way will this impact HD7900.

HD7900 launches Jan 9th.

GK104, somewhere in Q2 2012. And it's already been reported GK100 is towards end of 2012.

So how will this even scratch HD7900?

My post in the other thread was simply NV would release PR. It's what they do when the other team come close to a release. I made no other statements, simply, to expect NV PR. And i'll happily admit, no way can NV impact on sales of 79xx.

 

[nt300]

No way will this impact HD7900.

HD7900 launches Jan 9th.

GK104, somewhere in Q2 2012. And it's already been reported GK100 is towards end of 2012.

So how will this even scratch HD7900?

This is only to prevent AMD from getting all the spot light.

 

[Deleted member 24505]

Trouble is the Nvidia equivalent to the 79xx will cost shit loads more.

 

[Casecutter]

the54thvoid / thunderising - Exactly, this is just placed by Nvidia PR to keep their base from considering jumping ship. But it's a lot about nothing as we know specification are subject to change, something the article eluded when saying "depending on what the CUDA core count ends up being" so this means nothing.

The thing that caught my eye was "Products based on this will launch in the first quarter of 2012" so less than three months? Interesting that they’ve all of a sudden found an extra two months, at least moved it up from the last reports. Hum, wonder if Nvidia is seeing that if there late to the party by more like 5 months they see more of their fan base stumbling out the door long before they find a place to park.

Tells me Nvidia will come to market with whatever they can squeeze from the yields TSMC process can give, them in the next two months not waiting for good yield of top shelf chips.

 

[bear jesus]

For me the most important spec is still missing, unless the new nvidia cards support 3 or more monitors from a single card i can't even consider them an option.

 

[Lionheart]

I call bullshit AMD will flop once again just like there cpus did.

Your such a fanboy bro --_--

 

[qubit]

GK100

• 1024 CUDA cores
• 128 TMUs
• 512-bit GDDR5 memory interface, 64 ROPs

Hell yeah, you can say that's gonna have an impact on me all right. Just look at that perfect power of 2 design! That's the ideal for a digital design and will max out its performance. I bet performance is gonna be awesome. Saving my pennies now.

 

[erocker]

I will spend no less than $700 bucks on the Kepler 100 to play all of my favorite console ports.

 

[Crap Daddy]

With some pHysX too!

 

[(FIH) The Don]

sooooo, first $1000 consumer gaming gfx ever goes to nfabia

you wanna bet?

 

[Recus]

[yt]x7DQQ3Uki-c[/yt]

 

[GAR]

It wonderful the speculate. But this time round the HD 7900 is a brand new design something that was in design for some time now. Looking at the specs of Nvidia's Kepler, it may be a close competition for them to keep up with AMDs HD 7970 IMO.

Looks at your avatar and stops reading.

 

[Beertintedgoggles]

Looks at your avatar and stops reading.

Wow, that was almost as insightful as this post.

 

[[H]@RD5TUFF]

Wonder what pricing will be.

 

[HumanSmoke]

sooooo, first $1000 consumer gaming gfx ever goes to nfabia

you wanna bet?

Kind of loaded in your favour isn't it given that the original Asus Mars was well over $1k?

Of course if you're talking about being closer to reference then the HD 5970 4GB also tipped the scales at ~$1100. (Sapphire Toxic, XFX BE)

Taking inflation into account, the 8800 Ultra wouldn't be far off the $1k number either.

 

[Benetanegia]

Wonder what pricing will be.

Me too. It could be anything from $150 to $500 for KF104, and $250-600 for GK100 depending on how Tahiti performs in comparison.

Before people start saying I'm crazy, remember that GK104 is the replacement for the GTX460/560, and the chip will be smaller than GF104/114, most probably. 28 nm allows for nearly twice the transistors in the same space, and this GK104 is not even 2 times GF104, it has only 50% more ROP/TMU/memory bus. SPs is what takes less die area. Wild guesstimate without anything to support it, besides evolution/differences on Fermi based chips, 3.3 billion transistor, 320 mm^2.

Same with GK100, only 33% wider memory bus and ROPs. 64 SP per cluster versus 32 on GF100 == lots of savings. It will be smaller than GF100 for sure, much smaller probably. Guesstimate, 4.4 billion transistor, 420 mm^2.

Botom line, IMO, GK104 cards will be "cheap" to produce, as cheap as a GTX560 once there's enough supply. But looking at specs they do look like they might be able to compete with Tahiti based card(s), if even only with the HD7950, so at how much they decide to launch them... that's another story.

 

[buggalugs]

The 7970 will be great I'm not really concerned what nvidia will do months down the road.

It will be the same old story anyhow, Amd brings out new gen, Nvidia waits to see how good they are, then they make Nvidia cards 10% faster and charge $100-$150 more.

 

[Wile E]

Wake me when there are benchmarks.

 

[seronx]

If Kepler were Fermi

1024 Cores will mean it will have to be the same clock rate as Tahiti to get the same TFlops

Putting down this:
1. If one has the higher clock rate get the one with the higher clock rate
2. If the clock rate is the same look at the memory bandwidth to make your decision instead

512bit GDDR5 vs 384bit GDDR5
will be a big thing for GK100

Just sad no one went XDR2 when that was made for GPUs in mind

 

[John Doe]

If Kepler were Fermi

1024 Cores will mean it will have to be the same clock rate as Tahiti to get the same TFlops

Yeah, since we know how fast Tahiti exactly is, we can assume it's about as fast as a 580 with 1024 cores... no?

Putting down this:

2. If the clock rate is the same look at the memory bandwidth to make your decision instead

Memory isn't what sets performance. It's about the core and the clock speeds.

Come on man, you can do better than to spout out AMD fanboyism. I know it, you know it, so why are you still doing it?

 

[xenocide]

Considering the only information we have on Tahiti's performance is some AMD Marketting slides (same people that had dozens of slides showing the 8150 crushing the 6-core Intel CPU's in all tasks) I'll wait until both are benchmarked to see which one I go after. I have gone for AMD GPU's 3/4 of the time, and would like to swing to an nvidia since AMD drivers have as of recent been giving me problems, and I haven't used one since my 7800GS.