NVIDIA Kepler GPU Speculation thread

[nsavop]

Fudzilla is reporting Kepler will launch in 1st quarter, I bet we will have more leaks out of nvidia as we get closer to AMD's launch of tahiti in the coming days.

 

[tviceman]

Somehow I'm a little skeptical about Maxwell being 16x... Leaked slides etc show Kepler to be a max of 2x faster and for Maxwell to be 16x that would require an 8x jump in just one more generation. I think 4x performance for Maxwell is more realistic...

When JHH and Nvidia talk about that, they are specifically talking about Compute performance, not graphical performance.

 

[tviceman]

Fudzilla is reporting Kepler will launch in 1st quarter, I bet we will have more leaks out of nvidia as we get closer to AMD's launch of tahiti in the coming days.

Yep, good news for consumers.

 

[BallaTheFeared]

Fudzilla is reporting Kepler will launch in 1st quarter, I bet we will have more leaks out of nvidia as we get closer to AMD's launch of tahiti in the coming days.

Without any doubt, AMD is the first with a successful 28nm launch and it will enjoy some time at the top of the market, at least a couple of months.

The article is from the second... The 7970 isn't even out yet though...

I hope Nvidia brings out some decent parts, the 7970 wasn't all that impressive and my 9800GT isn't cutting it for PhysX I'd like to use a water cooled 470 for PhysX with a Kepler or two pushing rendering.

 

[Arkadrel]

Q1 2012 = between Jan - April right?

And there are rumors that nvidia are starting from the buttom and working its way up the ladder.
So the first cards released for the 6xx series are supposed to be the low level and mid level cards.

Suppose the first card is something like a 640 card, and thats around April,... it ll be a few months longer until the 680 is launched. That would give the 7970 like 4-5 months or so on the top.

 

[96Firebird]

Q1 2012 = between Jan - April right?

And there are rumors that nvidia are starting from the buttom and working its way up the ladder.
So the first cards released for the 6xx series are supposed to be the low level and mid level cards.

Suppose the first card is something like a 640 card, and thats around April,... it ll be a few months longer until the 680 is launched. That would give the 7970 like 4-5 months or so on the top.

Don't sound so optimistic...

Q1 is January - March.

 

[nsavop]

Q1 2012 = between Jan - April right?

And there are rumors that nvidia are starting from the buttom and working its way up the ladder.
So the first cards released for the 6xx series are supposed to be the low level and mid level cards.

Suppose the first card is something like a 640 card, and thats around April,... it ll be a few months longer until the 680 is launched. That would give the 7970 like 4-5 months or so on the top.

There were some early rumors of nvidia launching Kepler from the bottem up but latest rumors have been pointing towards nvidia launching there performance part (GK104) first. Most likely around the time AMD launches the HD7950. Of course this is all speculation.

 

[Grooveriding]

Don't sound so optimistic...

Q1 is January - March.

nvidia's quarter ends in April.

This is bad news in my opinion, it's a clone of what we saw with the 5870 launch. AMD launches and nvidia has nothing ready, we start hearing rumours of a card with zero card shots or non-bogus leaks.

With the prior situation we had five months of nothing but the same rumors, with a month before April actual card shots and specs becoming available.

April to June until we see the top-end part get released I'll bet.

 

[96Firebird]

nvidia's quarter ends in April.

 

[bryanW1995]

really? even a 768sp GK104 would be around 25% faster than a 7970. a 7970 is only 20-25% faster than a gtx580 and a 768sp GK100 would probably be 50% faster than a gtx580.

The specs are only 50% greater than gtx 580, so I'd think that something more like 30-40% faster is likely. That would still be 10-15% faster than 7970, but not as much of a speed bump as gtx 480 was over 5870. Unfortunately for NV, it's obvious that AMD is holding back on 7970, they should easily be able to put out a 7975 or 7980 to compete with the gk104. AMD's only answer to gk100 would be a sandwich, but when is that even coming out? November? 2013?

I am being conservative because I don't want people to point out to me later that you can buy a factory pre-overclocked HD7970 with 1.1ghz clocks. I also have to account for HD7970 gaining about a 10% increase in performance through drivers. :biggrin: Not that NV won't add more performance through driver increases (so it might be a moot point). They just added 25% more performance for SKYRIM (indoors).

I am almost shocked that HD7970 is only 25-30% faster than a GTX580 because NV added 15-18% just by going from GTX480 to 580 without even trying. :hmm:

Still I will choose to remain conservative for Kepler because those specs are just rumors, not facts. For all we know GK100 might have 768 SPs and the 1024 version is 2x GK104s? Still, AMD set the bar really low this time.

I don't think they set it much lower than they did with the 40nm refresh. Wasn't 5850 ~ gtx 285 speeds at launch? And it's reasonable to expect that 7950 will be ~ gtx 580, maybe a few % higher, right? AMD's "winning" battles like this is their ability to launch while NV is still ramping up their larger/harder to manufacture dies. Unfortunately for AMD, 2 1/2 yrs ago they launched in time for dx11/BF but were hamstrung by TSMC's epic fail, and this time they completely missed the holiday shopping season. This isn't like the david/goliath battle they have vs intel, AMD should be able to execute their strategy better than this. As much as I liked reading Carroll Killebrew's interviews, I wonder if issues like this weren't the primary reason for his ouster.

I'm just really surprised by your attitude. Yesterday you were bashing and moking 7970 for not being as good as the speculations suggested it to be. And now you're posting your own speculations and want people to believe in them. :whiste:

You're posting the unconfirmed rumours that were floating around for a few months already. Creating a new redundant thread with old rumours the day new tech launches. Err I smell biasiness. ()

I don't think he's biased at all, he's just not a blind fanboy for either camp and he remembers all the hoopla that erupted around here over the past few years when one camp or another came out with the "next best thing".

Seriously if the highest end Kepler part only has gtx590 performance, it will be a fail in my book. GTX590 was entirely constrained by it's heat AND power draw, thus it was never able to be the fully realized product it could have been.

Why? Both camps have had pretty stable updates/improvements over the past few years. gtx 480 ~ gtx 295, so if gtx 680 is ~ gtx 590 then why should we be surprised? That would still make it quite a bit faster than 7970. Remember, the big new architecture is going to be Maxwell in 2013/14. Kepler is just continuing the iterative refinement of the original Fermi architecture.

 

[Lonbjerg]

nvidia's quarter ends in April.

This is bad news in my opinion, it's a clone of what we saw with the 5870 launch. AMD launches and nvidia has nothing ready, we start hearing rumours of a card with zero card shots or non-bogus leaks.

With the prior situation we had five months of nothing but the same rumors, with a month before April actual card shots and specs becoming available.

April to June until we see the top-end part get released I'll bet.

It's just a sign of NVIDIA's new 2 year GPU plan.

GeForce GTX 480: March 26, 2010
GeForce GTX 680: Q2 2012
GeForce "next": Q2 2014

Throw in refreshes as needed.

 

[bryanW1995]

Different companies use a different monthly fiscal cycle for financial reporting purposes. Just like the US govt's year ends in sept IIRC, go figure that one.

 

[Gryz]

Different companies use a different monthly fiscal cycle for financial reporting purposes. Just like the US govt's year ends in sept IIRC, go figure that one.

And when companies talk about a quarter in their fiscal cycle, they'll use the abbreviation FQ. Like FQ01. And FY2012 when they talk about the fiscal year: Feb 2012 - Feb 2013. When you say "Q1" it means quarter in the calendar year. If not, then it is sloppy writing.

 

[Crap Daddy]

Here's something to discuss about:

http://www.chiphell.com/thread-338350-1-1.html

 

[MrK6]

Here's something to discuss about:

http://www.chiphell.com/thread-338350-1-1.html

Those core counts are odd, to say the least. From a completely macroscopic inspection, it looks like these new cores are more like AMD's - less work per core. Using logic as my only backup, I imagine that's probably a more efficient design, so it should be interesting to see what the final product will be like.

Thanks for the link Crap Daddy :thumbsup:

To reply to some of the sentiment in this thread, as I've mentioned before, NVIDIA has somewhat backed itself into a power consumption wall with where it can go in regards to Kepler's performance. The GTX580 sets the bar high in both performance and power consumption which I think leaves little wiggle room for NVIDIA's design. While it may not be as important a factor to some individual customers, corporate/scientific customers keep a cautious eye on power profiles. I'd be very surprised if NVIDIA pushed their power profile any further, and would assume they'd rather keep it the same or even take it back a bit with 28nm. If GK104 comes in at 300W, its performance will have to more than make up for it. Like AMD has shown with the 7970, creating a beefy chip but then scaling the clocks back to optimize power requirements seems to leave everyone smiling. Time will tell.

 

[Arkadrel]

Here's something to discuss about:

http://www.chiphell.com/thread-338350-1-1.html

Is that for real?.... looks off.

GK104 = slower than 7970/7950/7890?....1536 Cuda cores? only 256bit?
Is nvidia handing over intire GPU lineup except the 1 card at the top? doesnt make sense.

GK100 = 512bit,64 RoPs, 300watts TPD..... only 320 GB/s BW?..
Nvidia disabled double precision performance in other cards, but not this one. Nvidia wants people to buy this version.


thread has comments like:

Entertainment paste, copy A Card Specifications

False can not be false, a moderator can lock the

Obviously false.

There is no need to discuss, too fake.

There are no source

Nonsense

YY have to go too far. Moderators can delete posts up to

Who design chips but also get to run sub-parameters? If the package of this level of fraud.

Does not exist GK100, only GK110, false map.

^ in short the Chinese guys on Chiphell are labeling that as a fake, to generate page hits.

 

[MrTeal]

GK100 = 512bit,64 RoPs, 300watts TPD..... only 320 GB/s BW?..
Nvidia disabled double precision performance in other cards, but not this one. Nvidia wants people to buy this version.

512bit * 5GHz effective * (1byte/8bits) = 320GB/s

Unless GPUs start being released with XDR2 or they start moving to a larger than 512bit data bus (unlikely any time soon), I don't think we'll be seeing much more memory bandwidth than that on stock cards.

 

[Eisenstein]

That graph has to be fake because NVIDIA said they will only be using a single clock for geometry + shader processing. So no more 800 core/1600 shader figures. But to compensate their base clocks will be a lot higher than 1000 MHz. That slide there is showing traditional Fermi/Tesla type clocks.

To top it off, it's showing some obscene number of cuda cores with very low ROP and memory bandwidth... That's just too unlikely for NVIDIA to cripple their cards thus.

 

[tviceman]

To reply to some of the sentiment in this thread, as I've mentioned before, NVIDIA has somewhat backed itself into a power consumption wall with where it can go in regards to Kepler's performance. The GTX580 sets the bar high in both performance and power consumption which I think leaves little wiggle room for NVIDIA's design. While it may not be as important a factor to some individual customers, corporate/scientific customers keep a cautious eye on power profiles. I'd be very surprised if NVIDIA pushed their power profile any further, and would assume they'd rather keep it the same or even take it back a bit with 28nm.

Same argument used with the gtx480. Turned out it wasn't a problem when moving to the gtx580. New node, new focus on energy efficiency, and updated architecture = not backed into corner. I think the flagship Kepler GPU (at reference specs) will consume less power than the gtx580 (at reference specs).

http://seekingalpha.com/article/307...2-results-earnings-call-transcript?part=qanda

We have more notebook design wins for the Ivy Bridge cyle than we ever had in notebooks. And this (current) is likely the most successful notebook cycle we've ever experienced. And so we've got a lot of engineers dedicated to getting those notebooks into production. The reason for our success, I believe, is because our Kepler generation of GPU was designed for intense energy efficiency. And with energy efficiency, we were able to translate that to simultaneously higher performance, as well as longer battery life.

We will get concrete answers regarding Kepler's release date when Nvidia has their next conference call in February.

 

[Genx87]

Same argument used with the gtx480. Turned out it wasn't a problem when moving to the gtx580. New node, new focus on energy efficiency, and updated architecture = not backed into corner. I think the flagship Kepler GPU (at reference specs) will consume less power than the gtx580 (at reference specs).

http://seekingalpha.com/article/307...2-results-earnings-call-transcript?part=qanda


We will get concrete answers regarding Kepler's release date when Nvidia has their next conference call in February.

I completely agree. It is laughable to say Nvidia has backed itself into a corner from a power perspective knowing Nvidia has said they focused on perf\watt this generation and their mobile wins based on Kepler.