Nvidia's Next Gen Speculation

[kaktus1907]

Exactly what they need if they want to be successful again... looks like the point finally went home with Fermi.

yeah its good to see nVidia has learned something about perf/w.. like JHH said transistors are free but not power..


and i hope they release one of their tessellation demo called blabla city^!? from nvidia.com.. wonder to see how performs with my GTX460..

[Mechanical Man]

agreed
we can expect about 1.5-2x perf per watt just going to 28nm, which means its going to get 2x better than that from their improvements. which should mean that a 28nm midrange chip should offer some incredible bonuses. imagine a physics card for <100$ that is the same as a gt480, but uses only 60-80W

Remember its about DP performance, so fatter cudacores with more fat that does nothing for games. I really hope they will release two different dies, one for gaming one for gpgpu.

[Manicdan]

Remember its about DP performance, so fatter cudacores with more fat that does nothing for games. I really hope they will release two different dies, one for gaming one for gpgpu.

nothing for games is a little harsh
what about physics or tessellation, DP does some pretty useful stuff right? even if its not applying textures and lighting

[Mechanical Man]

nothing for games is a little harsh
what about physics or tessellation, DP does some pretty useful stuff right? even if its not applying textures and lighting

Yes I mean in context of graphics that is mandatory for game card. Physics is extra that is not mandatory. Also note that it is about perfomance per watt, so it might be that real power difference is just 2x but power need is lot lower (not that is bad, just so that dont jump your horses yet about 3-4x "real" performance).

About tesselation i dont know really nothing, but im in impression that Fermi does it with dedicated units so nothin to do with DP performance, correct if im wrong.

[cal_guy]

nothing for games is a little harsh
what about physics or tessellation, DP does some pretty useful stuff right? even if its not applying textures and lighting

Still useless for both.

[Jowy Atreides]

ATI will definitely release HD6000 before Nvidia can get out a refresh (and that's all HD6000 really is)

No it's not.

Architecture rework.

[ElSel10]

No it's not.

Architecture rework.

Yes, there have been some changes. But nowhere near a full reworking. A lot of stuff had to be taken out (kept the same) as it has to remain on 40nm.

[Dimitriman]

Yes, there have been some changes. But nowhere near a full reworking. A lot of stuff had to be taken out (kept the same) as it has to remain on 40nm.

The question is: need there be more at this point?

[Iconyu]

nothing for games is a little harsh
what about physics or tessellation, DP does some pretty useful stuff right? even if its not applying textures and lighting

In games anything that needs more than 32bit precision is done on the CPU, Double Precision is basically GPGPU stuff.

This smells like the same trap nvidia fell into on the FX chips, DX 9 only needed 24bit precision for it's shaders on release but nvidia builds a 16/32bit part that's fast on 16bit but slow on 32bit. This meant every game needed to be optimised with 16bit shaders to come even close to ATI's 9k series.

Everything mentioned since the Keynote started has been either proprietary (which means it'll only be supported if nvidia's card succeeds) or aimed at making a super computer. And while I loved DEC Alpha technology, I've not tried to find one second hand to own.

[ToxAvenger]

28nm Kepler

Focus is apparently on performance/watt with Kepler aimed to be 3-4 times more efficient..

How can Nvidia even begin to call Fermi a 2009 product? The way things are going, NV will take until the end of 2010 to release a full DX11 lineup. Unless Jensen counts the woodscrew wonder as a product release.

As for 28nm, how is this going to save Nvidia? AMD will also move to that node (and might have something from GlobalFoundries as well). Nvidia needs to rethink their strategy, otherwise I fear they are going to get further and further behind.

[ElSel10]

OMG!! I just noticed, Day 2 of GTC is going to be with U of I at Urbana-Chanpaign!!! I got accepted there but didn't go. Everyone at my school goes there for parties though!

[Jowy Atreides]

Kepler looks far off with the 2 year gap from fermi (2009, wtf?) but the gap until maxwell looks concerning.

A few years of renames

[Chumbucket843]

No it's not.

Architecture rework.

Kepler looks far off with the 2 year gap from fermi (2009, wtf?) but the gap until maxwell looks concerning.

A few years of renames

your bias entertains me and others. keep it up.

[Sampsa]

Lets see if GF110 will be a "mid-life kicker" which Huang said would be a product which launches in-between each new NVIDIA chip.

[saaya]

First rumors from Muropaketti (me) didn't make much sense and now they are obvious?

the first 2 infos were obvious, the cuda core count doesnt make sense imo...

He is one of those that is confused on this forum and distort a lot.

Lol I'm not biased at ALL. I buy whatever is best for me. But Nvidia has sickened me over the past couple of years. They release the same over and over with a different name, and worst of all, they can't even release top of the line but expensive cards anymore.

They are talking about everything EXCEPT gaming now because even they are smart enough to realize how terrible their recent GPU's have been.

If you really think all the mumbo jumbo they're trying to feed you now actually matters, you must not know how businesses work.

Nvidia's business plan: If you're failing at making the best gaming GPU's, start talking about other useless technologies that look pretty on a Power Point presentation.

Nvidia's gaming market business plan: If you're failing at making the best gaming GPU's, start focusing on some specific advantage of your GPU's, and release benchmarks in games that take advantage of said technology.


I'll bet you guys Nvidia will NOT have a response to ATI's October lineup. If they have the audacity to release a dual GF104 card, I will laugh, cause it will be such an obvious and pathetic attempt at keeping up. It will prolly consume 400W, and run at 100C.

what else COULD they do?
they can either shut up, which isnt a good idea for a company, or they can try to focus on areas where they are still competitive... every company does that...

i dont like this marketing bs either, but thats really all they CAN do in situations like this, and the sad part is that for 99% of the people out there it actually works... so not doing it would mean theyd lose money...

agreed
we can expect about 1.5-2x perf per watt just going to 28nm, which means its going to get 2x better than that from their improvements. which should mean that a 28nm midrange chip should offer some incredible bonuses. imagine a physics card for <100$ that is the same as a gt480, but uses only 60-80W

28nm Kepler

Focus is apparently on performance/watt with Kepler aimed to be 3-4 times more efficient..

according to the graph its 5x, but if you check actual numbers, then youll notice that tesla is positioned too high on the graph and fermi too low... which artificially increases the jump to keppler making it look a lot more impressive than it actually is... according to the graph keppler will have 5gflops/W while fermi supposedly has around 1gflop/W... in reality tesla c2070 cards are at close to 2.5gflops/W, so keppler is merely a double of that, slightly more than that, but definitely not a 5x jump as the graph makes it look

I really hope they will release two different dies, one for gaming one for gpgpu.

me2, but i dont think so... :/ theyd have announced it, and i think the hardware would be so similar it doesnt make sense... all they really need to do is make those massive chips on a node once they AND tsmc had some experience with it, and not when the node is brand new... i think then their strategy of having one big chip that does it all can actually work out... but of course it looks like they will throw a massive chip at tsmc again once they announce their next node...

The question is: need there be more at this point?

if nvidia wants to own the highend, yes...

Lets see if GF110 will be a "mid-life kicker" which Huang said would be a product which launches in-between each new NVIDIA chip.

thats just jensen marketing speak for "refresh" :P

[bamtan2]

Lets see if GF110 will be a "mid-life kicker" which Huang said would be a product which launches in-between each new NVIDIA chip.

I think it is safe to say there will be a fermi respin, just because that commonly happens.

I also expect kepler to be an shrunk, evolved fermi, not totally new

[Sn0wm@n]

this conference is not for just gamers, so why talk forever about that stuff.
theres ALOT of good technology being discussed and the uses for GPUs are growing, gaming wont be the only thing you buy them for

so what's the current use for you besides gaming that you could do today if you buy a gpu ?????


and is it really needed to buy an nvidia gpu to do those task or can the competition do it right now ?????

That chart proves they were late releasing Fermi as it was a 2010 release not 2009. What do you think that means for Kepler...

2012 release???

Fermi was a significantly redesigned architecture on a new process (a flaky one at that) all on one step. That is why it was late. Presumably Kepler will be based on Fermi, just scaled up and tweaked. Nvidia's experience with Fermi and not very signifiant architectureal changes should mean a much more timely release.

Remember how late ATIs new architecture was (R600)? Then after that, how smooth everything went? As you can see, ATIs new Northern Islands architecture has been significantly delayed due to process roadmaps not hitting their targets, so instead we get Southern Islands which will be very similar to HD5000 and on the same process.

could it be because tsmc canceled a node in between ???? you know.. 32nm .. and now they are going with 28nm .. maybe those delays are because of that instead of architecture design flaws .....

JHH defeated the kraken with 4 bladez of fury combo !!!

only 4 blades ???? LOL i doubt that very much ...

Yes, there have been some changes. But nowhere near a full reworking. A lot of stuff had to be taken out (kept the same) as it has to remain on 40nm.

why dont we wait to see actual numbers of the big gun to see how they compare.... then you can stop making those claims that its nothing beside a rebrand .. to wich an article by an amd exec said it wasnt ...

[Xoulz]

this is just more lipservice...

[570091D]

the amd fanboi force is strong in this thread...

[Mechanical Man]

the amd fanboi force is strong in this thread...

nVidiotism is great in you, JHH is pleased

[Olivon]

the amd fanboi force is strong in this thread...

As always, spittin' on nVidia is their favourite "passe-temps". Intel is too strong for the moment so let's spitte on nVidia !

Some AMD fanbois are really really boring ... always the same song ...

[JohnZS]

the amd fanboi force is strong in this thread...

But they are not Jedi yet

Joking aside I can seriously see nVidia branching out their products purely to suit each niche market.

Raw DP performance will be reserved for Telsa based products, and then we will see some sort of "half way house" for the Quadro type products with GOOD DP performance and then we will see mediocre to moderate DP performance for the consumer products.

What this means is that consumer end users get cheaper, less power hungry and cooler running cards.

GF104 was just the beginning....

[Jowy Atreides]

your bias entertains me and others. keep it up.

My gpu is a 470.
Bias or fanboyism can be misconstrued easily.

There's a difference between beating a dead horse and blindly arguing without actually integrating new logical input.
I'll beat down any manufacturer when it's enjoyable but I'll buy whatever part is the best £/perf at the rate I'm willing to pay.

I am not ignorant, bias is for idiots.

Let's face the elephant in the room, nvidia is massively behind and in not very good shape to catch up.
Their strategy is to impose false gpu restrictions to upsale SLI to customers (fake perf restriction x) and lie through the teeth about upcoming products to stave customers.

Remember when the 5 series came out and nvidia was bragging at launch about their "soon to be released" product?
That's what this conference is about, they see 6xxx being released and are trying to divide end users in the community.

[Manicdan]

i keep wondering who people think the conference was shown for
i dont think it was for gamers to watch about on youtube. i think it was to try and find more investors to expand the capabilities for cuda. so much was about how far they have come, only a hint was about gaming or future products.

[Baron_Davis]

what else COULD they do?
they can either shut up, which isnt a good idea for a company, or they can try to focus on areas where they are still competitive... every company does that...

i dont like this marketing bs either, but thats really all they CAN do in situations like this, and the sad part is that for 99% of the people out there it actually works... so not doing it would mean theyd lose money...

i would laugh if their supercomputing/professional/mobile segment tech ends up taking them to the next level financially speaking.