Megadrive1988
Veteran
Thanks guys for answering my question. I guess Kepler will be a Fermi upgrade/refresh.
straaljager
Newcomer
Is Kepler going to be the first GPU bearing the fruits of Project Denver or is that honor reserved for Maxwell?
Considering Denver's timeframe it's obviously for Maxwell. There is atleast one mention of this:
But if Kepler will be nextgen console winner, then i think it could or even should be bundled with Denver in SoCThe Register said:
Last edited by a moderator:
DegustatoR
Legend
I don't think that Maxwell+Denver parts will hit GeForce/Quadro market. They'll probably use them for Tesla/console markets and GeForce/Quadro will get just Maxwell, without ARM cores integration.
Game porting is much more dependant on API compatibility than on binary h/w compatibility these days.
ARM cores should be helpful for Quadro as well, they will break last barriers on the road to completely GPUs based offline rendering solutions
Key words are "these days", but this could easily change in future. My opinion that decrease of software layer would be one of key directions for next gen XBox, command list generation and some other usually api/driver's stuff in the end could be standardized between the vendors, however looks like SM ISA apparently couldn't be due to fundamental architectural differences
Megadrive1988
Veteran
I just bought a Fermi-based GTX 460 and I love it. I cannot wait to see what Nvidia does with Kepler and Maxwell.
Blazkowicz
Legend
I just bought.. a GT240 with gddr5.
because the Fermi equivalent costs more and has slow fillrate, lower bandwith.
I'm eager to see Kepler as well, with 28nm allowing a better < 75W card to be made.
because the Fermi equivalent costs more and has slow fillrate, lower bandwith.
I'm eager to see Kepler as well, with 28nm allowing a better < 75W card to be made.
It might be already too late for Kepler, but I wonder if NVIDIA will keep their "one size fits all" approach to design. I mean, these huge humongous chips are not the best silicon designs around to make a lofty margins in the desktop arena.
If they keep on this track we'll start talking wafer per chip ratio, not the other way around.
Couldn't they find a way to modulate their designs enough to be able to "easily" remove (at some of) the general purpose from the desktop parts without unbalancing the chip and making padding overuse a must? I know the design of the pipeline stages themselves are altered and thus it's not a simple external/independent block and thus making it hard to remove. But, there might be some ways to re-factor this issue and allow them to produce smaller and leaner desktop part.
If they keep on this track we'll start talking wafer per chip ratio, not the other way around.
Couldn't they find a way to modulate their designs enough to be able to "easily" remove (at some of) the general purpose from the desktop parts without unbalancing the chip and making padding overuse a must? I know the design of the pipeline stages themselves are altered and thus it's not a simple external/independent block and thus making it hard to remove. But, there might be some ways to re-factor this issue and allow them to produce smaller and leaner desktop part.
You are kidding, right?
They did something similar with gf100 vs gf104 and with gf110 vs gf114.
A1xLLcqAgt0qc2RyMz0y
Veteran
9 pages so far on this thread and no REAL speculation on how Kepler will be different from Fermi 1.1 (GF110).
What is known is that Kepler will be on 28nm process and that high speed process at TSMC will be in production at the end of 2011.
I still expect that the high end/HPC Kepler will once again be a very large chip.
So again what Arch changes can we expect in Kepler vs Fermi 1.1 (GF110)
What is known is that Kepler will be on 28nm process and that high speed process at TSMC will be in production at the end of 2011.
I still expect that the high end/HPC Kepler will once again be a very large chip.
So again what Arch changes can we expect in Kepler vs Fermi 1.1 (GF110)
It's typical of long running speculation threads.
caveman-jim
Regular
Is it tape out time yet? Seems like it ought to be. I wonder if Kepler will have PCI-Express 2.1 or 3.0?
I wouldn't be surprised if Kepler is a bunch of Tegras stitched together. Seems like that's all they talk about nowadays. Hard to get very excited about new hardware this year though. Software and games are already lagging behind considerably and that's not gonna change this year.
Didn't know you loved G7x so much Trini!
Since when had G7x or even generations before that FP20 PS ALUs?
D
Deleted member 13524
Guest
Erm.. there was an option for either FP16 or FP32 IIRC.. but that doesn't make them less advanced than the Tegra's PSs...
