NVIDIA Details "Pascal" Some More at GTC Japan

[btarunr]

NVIDIA revealed more details of its upcoming "Pascal" GPU architecture at the Japanese edition of the Graphics Technology Conference. The architecture will be designed to nearly double performance/Watt over the current "Maxwell" architecture, by implementing the latest tech. This begins with stacked HBM2 (high-bandwidth memory 2). The top "Pascal" based product will feature four 4-gigabyte HBM2 stacks, totaling 16 GB of memory. The combined memory bandwidth for the chip will be 1 TB/s. Internally, bandwidths can touch as high as 2 TB/s. The chip itself will support up to 32 GB of memory, and so enterprise variants (Quadro, Tesla), could max out the capacity. The consumer GeForce variant is expected to serve up 16 GB.

It's also becoming clear that NVIDIA will build its "Pascal" chips on the 16 nanometer FinFET process (AMD will build its next-gen chips on more advanced 14 nm process). NVIDIA is innovating a new interconnect called NVLink, which will change the way the company has been building dual-GPU graphics cards. Currently, dual-GPU cards are essentially two graphics cards on a common PCB, with PCIe bandwidth from the slot shared by a bridge-chip, and an internal SLI bridge connecting the two GPUs. With NVLink, the two GPUs will be interconnected with an 80 GB/s bi-directional data path, letting each GPU directly address memory controlled by the other. This should greatly improve memory management in games that take advantage of newer APIs such as DirectX 12 and Vulkan; and prime the graphics card for higher display resolutions. NVIDIA is expected to launch its first "Pascal" based products in the first half of 2016.



View at TechPowerUp Main Site

 

[TheGuruStud]

Another $1k card, folks. Hell, with that much ram, maybe more.

 

[Naito]

Another $1k card, folks. Hell, with that much ram, maybe more.

Would buy.

 

[Steevo]

With 16GB of RAM available to each GPU core what the hell are they planning on putting in the rest of the memory that it would need access to more?

We are at the point of needing 3GB of vmem with good memory management and 4GB for textures that may not be optomized, not 16 with piss poor management. 8GB would be enough with better timings to reduce latency.

 

[RejZoR]

Keeping developers lazy since, I don't know, too long. The only good thing with consoles is that devs actually have to code games wel in order to make them look epic and run the same. But on PC, who cares, just smack extra 8GB of RAM and a graphic card with twice the memory and you can have a sloppy coded game that will run kinda fine because of that...

 

[Naito]

Keeping developers lazy since, I don't know, too long. The only good thing with consoles is that devs actually have to code games wel in order to make them look epic and run the same. But on PC, who cares, just smack extra 8GB of RAM and a graphic card with twice the memory and you can have a sloppy coded game that will run kinda fine because of that...

Do you have a reference to such an occurrence? Maybe it's just for the next wave of insane 4K optimized games? Or perhaps the 16GB will be reserved for that whole Titan X-style scenario and the more reasonably priced SKUs will have 8GB?

 

[HumanSmoke]

Do you have a reference to such an occurrence? Maybe it's just for the next wave of insane 4K optimized games? Or perhaps the 16GB will be reserved for that whole Titan X-style scenario and the more reasonably priced SKUs will have 8GB?

That's probably closer to the truth I think. Two HBM2 stacks for a performance/mainstream product would greatly reduce overall cost (final assembly). The whole Pascal presentation seems to be an adjunct to the Supercomputing Conference (SC15) just held in Texas, so I'm figuring the thrust of the Pascal presentations are more toward Tesla than GeForce.

NVIDIA will build its "Pascal" chips on the 16 nanometer FinFET process (AMD will build its next-gen chips on more advanced 14 nm process).

Oh?
1. Can't say I've seen anything that actually backs up your assertion that 14LPP is more advanced than 16FF+/16FFC
2. I also haven't seen it confirmed anywhere that AMD will tap GloFo exclusively for GPUs. Somewhere in the last 7-8 weeks "could" has turned into "will". Some sources seem to think that GloFo will be tapped for lower end GPUs and Zen, with TSMC tasked with producing the larger GPUs.

 

[ZoneDymo]

Its that NVlink crap that has me worried.
Instead of, oh idk, having a single damn gpu capable of running games at 4k @ 120hz, they seem to focus and plan on us getting 2...

I already find it massively bs with the current gen of cards, 600 - 1000 euro for a card that feels outdated right away....

AC Unity on a mere 1080p with a freaking 1000 dollar Titan X only manages 48 fps...
BF 4 , 4k, same 1000 dollar card...a mere 41 fps....

I could go on, if I am laying down that kind of cash for the latest and greatest gpu it better damn well run atleast current ffing games at the highest settings and then some.
Its ridiculous the ask you to put down that kind of money twice.
I mean Im not even talking about the more demanding more advanced games in the future, im talking about the here and now and those cards cannot do that?

I feel we should demand better as consumers tbh but yeah that focus on better SLI using that NVlink....not a good thing to focus on imo.

 

[medi01]

Do you have a reference to such an occurrence?

The latest "The Longest Journey". (built on Unity engine, methinks)

1. Can't say I've seen anything that actually backs up your assertion that 14LPP is more advanced than 16FF+/16FFC

Actually, it is on the opposite as far as iThings go, some CPUs are manufactured on Samsung's 14nm, some on TSMC 16nm, the former consume more power.

 

[TheDeeGee]

I'm ready!

 

[HumanSmoke]

Its that NVlink crap that has me worried.

Why? Are you planning on building a supercomputing cluster?

I feel we should demand better as consumers tbh but yeah that focus on better SLI using that NVlink....not a good thing to focus on imo.

Sorry to bust up your indignation tirade, but since it isn't aimed at gaming SLI but workloads that are more intensive on system bus bandwidth - notably the Exascale computing initiative with IBM, Cray, and Mellanox, that really shouldn't be a problem.

 

[ZoneDymo]

Why? Are you planning on building a supercomputing cluster?

Sorry to bust up your indignation tirade, but since it isn't aimed at gaming SLI but workloads that are more intensive on system bus bandwidth - notably the Exascale computing initiative with IBM, Cray, and Mellanox, that really shouldn't be a problem.

If you think this will not find its way asap in gaming...
Its sorta how the internet originated in Military development, everything seems to start with Military or Space development and then finds its way to the consumer.
Sure this is first for industry but yeah will make its way to the gamers soon enough.

And even if it does not, statement still stands, hate this focus and "need" for dual gpu setups to get anything decent going.

also on a side note....dear gawd what a presentation...

 

[deemon]

Did they fix async compute with the Pascal? (or was Pascal already designed and "taped out" when the scandal started?)

Did they fix the VR preemption problem? Nvidia VR preemption "possibly catastrophic"

Did they add freesync adaptive sync compatibility?


And hopefully we get 980Ti performance in 970mini/Nano/or_even_smaller form factor.
And does anybody know yet Arctic Islands actual product availability (not paper) launch date/time?

 

[deemon]

With 16GB of RAM available to each GPU core what the hell are they planning on putting in the rest of the memory that it would need access to more?

We are at the point of needing 3GB of vmem with good memory management and 4GB for textures that may not be optomized, not 16 with piss poor management. 8GB would be enough with better timings to reduce latency.

"No one will need more than 637 kB of memory for a personal computer"

 

[Easo]

1TB/sec?
Well, damn...

 

[RejZoR]

Did they fix async compute with the Pascal? (or was Pascal already designed and "taped out" when the scandal started?)

Did they fix the VR preemption problem? Nvidia VR preemption "possibly catastrophic"

Did they add freesync compatibility?


And hopefully we get 980Ti performance in 970mini/Nano/or_even_smaller form factor.
And does anybody know yet Arctic Islands actual product availability (not paper) launch date/time?

It's not a scandal. Maxwell 2 is capable of async compute with limited queue size with minimal performance drop. Once you exceede that queue size, performance starts dropping. Radeon cards have a much larger async queue and they usually don't hit the ceiling. Not much known if games really need queues beyond what Maxwell 2 can do...

 

[arbiter]

Did they fix async compute with the Pascal? (or was Pascal already designed and "taped out" when the scandal started?)

Did they add freesync compatibility?

Kinda hard to support a tech that was AMD locked for so many years that didn't get added to DX12 til last minute which was after maxwell 2 was final. Freesync is AMD locked software solution, (go read AMD's own FAQ's before you try to call me a fanboy) Adaptive sync is the standard, freesync uses that standard in a proprietary way. Old 7000 radeon cards can do adaptive sync but not freesync. Pascal being taped out doesn't mean its final chip, it is a prototype chip that very well can change and isn't final design, But nvidia does have 5 month gap. Hope that means not 5-6 month gap between amd and nvidia's chips but could less AMD cuts corners which they will have to and likely not gonna be a good idea.

 

[medi01]

Freesync is AMD locked software solution, (go read AMD's own FAQ's before you try to call me a fanboy)

Adaptive sync is the standard, freesync uses that standard in a proprietary way.

1) GSync is as locked down as it gets (to "nope, won't license it to anyone" point)
2) adaptive sync is THE ONLY standard, there is no "freesync" standard
3) nothing stops any manufacturer out there to use adaptive sync (dp 1.2a), no need to involve AMD or any of its "freesync" stuff in there

 

[TheinsanegamerN]

Do you have a reference to such an occurrence? Maybe it's just for the next wave of insane 4K optimized games? Or perhaps the 16GB will be reserved for that whole Titan X-style scenario and the more reasonably priced SKUs will have 8GB?

Batman arkham knight and black ops III come to mind.

 

[qubit]

I'm really looking forward to that unified memory architecture and the elimination of SLI problems.

 

[dj-electric]

Arma and battlefield also love super high res textures.
Basically many non-single play , non "GPU tests" scenarious just like much more VRAM

 

[Casecutter]

(AMD will build its next-gen chips on more advanced 14 nm process).

Oh?
1. Can't say I've seen anything that actually backs up your assertion that 14LPP is more advanced than 16FF+/16FFC
2. I also haven't seen it confirmed anywhere that AMD will tap GloFo exclusively for GPUs. Somewhere in the last 7-8 weeks "could" has turned into "will". Some sources seem to think that GloFo will be tapped for lower end GPUs and Zen, with TSMC tasked with producing the larger GPUs.

Yea, btarunr came out of left-field with that snippet as soon as I read it it was WTF .
Thanks for that clean up.

NVIDIA is expected to launch its first "Pascal" based products in the first half of 2016.

Do we have this as confirmation? While sure it could see starting delivery's for HPC customer initiatives first (Exascale/IBM, Cray, etc), then professional products (Tesla/Quadro), while GeForce use of the GP100 should be out a ways.

 

[FreedomEclipse]

Batman arkham knight and black ops III come to mind.

Its amazing how people were so hyped about BLOPS III - It was all over the internet and now its like its completely faded into obscurity. Nobody talks about it no more.


But then again Fallout 4 happened so even though Activision we're the first to get their game out. its Bethesda thats gettin all the pussy

:::EDIT:::

Oh, and not to forget about battlefront of course which has already been available in many countries apart from the UK. People are either playing one of the two games

 

[HM_Actua1]

Hell yah! bring it! Excited for Pascal!

 

[deemon]

Kinda hard to support a tech that was AMD locked for so many years that didn't get added to DX12 til last minute which was after maxwell 2 was final. Freesync is AMD locked software solution, (go read AMD's own FAQ's before you try to call me a fanboy) Adaptive sync is the standard, freesync uses that standard in a proprietary way. Old 7000 radeon cards can do adaptive sync but not freesync. Pascal being taped out doesn't mean its final chip, it is a prototype chip that very well can change and isn't final design, But nvidia does have 5 month gap. Hope that means not 5-6 month gap between amd and nvidia's chips but could less AMD cuts corners which they will have to and likely not gonna be a good idea.

1) GSync is as locked down as it gets (to "nope, won't license it to anyone" point)
2) adaptive sync is THE ONLY standard, there is no "freesync" standard
3) nothing stops any manufacturer out there to use adaptive sync (dp 1.2a), no need to involve AMD or any of its "freesync" stuff in there

so much misinformation.

Adaptive sync IS FreeSync.

FreeSync is the brand name for an adaptive synchronization technology for LCD displays that support a dynamic refresh rate aimed at reducing screen tearing.[2] FreeSync was initially developed by AMD in response to NVidia's G-Sync. FreeSync is royalty-free, free to use, and has no performance penalty.[3] As of 2015, VESA has adopted FreeSync as an optional component of the DisplayPort 1.2a specification.[4] FreeSync has a dynamic refresh rate range of 9–240 Hz.[3] As of August 2015, Intel also plan to support VESA's adaptive-sync with the next generation of GPU.[5]

https://en.wikipedia.org/wiki/FreeSync