Support NeoGAF

[Objectively Bad Opinion]

Pascal 10x Maxwell? Where you going with this, Jen?

It's going to have 2 TFLOPs DP compared to the .2 on the Titan X

lol

 

[viveks86]

It's going to have 2 TFLOPs DP compared to the .2 on the Titan X

lol

lol

 

[Smokey]

Pascal tho o_o

 

[cyberheater]

I was hoping to see more gpus announced. Maybe it's the wrong event for that. Certainly nothing to excite me in this presentation.

 

[Buburibon]

smokey's google cache link

Thank you very much. I'm happy to hear this Titan is the "real deal" this time around. There might be custom-cooled and overclocked versions of it in the future, but the GM200 will remain essentially the same.

 

[XiaNaphryz]

I'm hoping my current card can hold out until Volta comes along.

 

[PnCIa]

I want a 970ti with 6GB Ram :/

 

[decoy11]

That robot car was founded by DARPA? Do you guys think they got a secret version of the car with a gun on top working yet?

 

[-SD-]

Put together some Pascal details:

- first NVIDIA GPU with 3D memory and NVLink
- 32GB of memory
- 3x the memory bandwidth of Maxwell
- 2x the performance per watt of Maxwell
- new feature: mixed-precision
- 4x the mixed-precision FP16 performance of Maxwell

Anything to add or change?

 

[cleveridea]

wow, so the pascal.. does that compete with the 390 series coming soon from AMD or do they have their own similar card down the line next year?

Sounds like pascal is gonna be a monster

 

[HariKari]

That robot car was founded by DARPA? Do you guys think they got a secret version of the car with a gun on top working yet?

Close

http://en.wikipedia.org/wiki/TerraMax_(vehicle)

 

[logikk]

Hmm, Nvidia really getting into the self-driving car market. Could be a really smart move for the company as a whole.

 

[-SD-]

Elon Musk in the house.

 

[Smokey]

Put together some Pascal details:

- first NVIDIA GPU with 3D memory and NVLink
- 32GB of memory
- 3x the memory bandwidth of Maxwell
- 2x the performance per watt of Maxwell
- new feature: mixed-precision
- 4x the mixed-precision FP16 performance of Maxwell

Anything to add or change?

Or maybe I'll hold on to my Titan Black until Pascal o_o

 

[Kiant]

God bless Dave the Buggy.

 

[SpyGuy239]

this self-driving car and deep neural network thing is nuts.

and now ELON MUSK?

!!!!!

 

[Zel3]

Put together some Pascal details:

- first NVIDIA GPU with 3D memory and NVLink
- 32GB of memory
- 3x the memory bandwidth of Maxwell
- 2x the performance per watt of Maxwell
- new feature: mixed-precision
- 4x the mixed-precision FP16 performance of Maxwell

Anything to add or change?

If true :O

Finally a real 4k card @ 120 fps

 

[LordOfChaos]

There's Elon. Soo, Tesla model with the Nvidia brain thingy?

 

[HariKari]

Is it really worth upgrading from my 980 for gaming purposes? Finding it difficult to justify the price for the performance the Titan gives over the 980, it seems a small increase based on leaked benchmarks.

Not worth it if you're still at 1080p

 

[Backfoggen]

Put together some Pascal details:

- first NVIDIA GPU with 3D memory and NVLink
- 32GB of memory
- 3x the memory bandwidth of Maxwell
- 2x the performance per watt of Maxwell
- new feature: mixed-precision
- 4x the mixed-precision FP16 performance of Maxwell

Anything to add or change?

And that's 2016? Yeah I'm waiting.

 

[JaggedSac]

They are going balls deep into neural networks huh.

 

[wachie]

Put together some Pascal details:

- first NVIDIA GPU with 3D memory and NVLink
- 32GB of memory
- 3x the memory bandwidth of Maxwell
- 2x the performance per watt of Maxwell
- new feature: mixed-precision
- 4x the mixed-precision FP16 performance of Maxwell

Anything to add or change?

"2016"

 

[RCSI]

Put together some Pascal details:

- first NVIDIA GPU with 3D memory and NVLink
- 32GB of memory
- 3x the memory bandwidth of Maxwell
- 2x the performance per watt of Maxwell
- new feature: mixed-precision
- 4x the mixed-precision FP16 performance of Maxwell

Anything to add or change?

I got a GTX 970 specifically to hold me over until Pascal. I am not disappointed.

 

[Zaph]

Put together some Pascal details:

- first NVIDIA GPU with 3D memory and NVLink
- 32GB of memory
- 3x the memory bandwidth of Maxwell
- 2x the performance per watt of Maxwell
- new feature: mixed-precision
- 4x the mixed-precision FP16 performance of Maxwell

Anything to add or change?

Jesus...

 

[Man]

Elon Musk is on stage.

 

[ashecitism]

That 32 is only max potential capacity btw. I doubt they'd release a card with that memory next year already? o_o Maybe for workstations..

 

[undercoverb0ss]

Is it really worth upgrading from my 980 for gaming purposes? Finding it difficult to justify the price for the performance the Titan gives over the 980, it seems a small increase based on leaked benchmarks.

Worth it if you plan on going to 4k soon and don't want to deal with multi-gpu

 

[BeEatNU]

Or maybe I'll hold on to my Titan Black until Pascal o_o

that is what I was thinking, it's like if I buy a Titan X, maybe I can resell it?
or just wait untill next year for pascal

 

[undercoverb0ss]

The 40" Philips 4K monitor looks like it will be shipping soon in NA, so I'm looking to upgrade before I make the jump.

 

[mugurumakensei]

32GB of vram, lawd jeezus.

 

[-SD-]

"2016"

That 32 is only max potential capacity btw.

Added to the list in the OP, along with the slides.

 

[J-Rzez]

Welp, looks like I'm waiting until next year for a GPU upgrade. Then again, I've said that in the past then submitted an order on newegg a couple days later in the past.

 

[cleveridea]

so...isnt he osborn-ing the new titan?

If pascal next year is 10x it kinda makes spending a whole bunch of money on titans very undesirable.

It seems more about pride and posturing, and I would have thought such information would make sense AFTER AMD releases their top end cards. Right now, Nvidia control the high end pretty much with 970/980.

 

[Kezen]

Put together some Pascal details:

- first NVIDIA GPU with 3D memory and NVLink
- 32GB of memory
- 3x the memory bandwidth of Maxwell
- 2x the performance per watt of Maxwell
- new feature: mixed-precision
- 4x the mixed-precision FP16 performance of Maxwell

Anything to add or change?

32gb of VRAM ? And I thought 12gb was almost too much. :O

 

[HariKari]

If pascal next year is 10x it kinda makes spending a whole bunch of money on titans very undesirable.

10x is only for a special type of compute that games won't use. It'll likely be 2-3x faster, with a lot more VRAM.

 

[Zerozerozero]

I was ready to buy a Titan X, but after this Pascal thing I'm not sure anymore.

 

[BeEatNU]

I was ready to buy a Titan X, but after this Pascal thing I'm not sure anymore.

same lol

but it could be like end of 2016 so I'm like man that is a long wait lol

 

[thuway]

Any more information on Nvlink? Is it going to work on existing motherboards? What will be the vendors that make it? Etc.

 

[wachie]

Any more information on Nvlink? Is it going to work on existing motherboards? What will be the vendors that make it? Etc.

AFAIK Nvlink is the interposer for the "3D" memory.

 

[TC McQueen]

Even Elon Musk is like "Nvidia, cool it with the Skynet stuff."

 

[Rip]

same lol

but it could be like end of 2016 so I'm like man that is a long wait lol

Then it will get delayed till first quarter 2017 and you will be like VOLTA, right around the corner bro! 20x faster than Pascal!

 

[TC McQueen]

That pic of Pascal really nice looking. I can't wait to sell off my 970 to get one.

 

[ashecitism]

Any more information on Nvlink? Is it going to work on existing motherboards? What will be the vendors that make it? Etc.

No new info

and this what anadtech wrote last year

But the rabbit hole goes deeper. To pull off the kind of transfer rates NVIDIA wants to accomplish, the traditional PCI/PCIe style edge connector is no good; if nothing else the lengths that can be supported by such a fast bus are too short. So NVLink will be ditching the slot in favor of what NVIDIA is labeling a mezzanine connector, the type of connector typically used to sandwich multiple PCBs together (think GTX 295). We haven’t seen the connector yet, but it goes without saying that this requires a major change in motherboard designs for the boards that will support NVLink. The upside of this however is that with this change and the use of a true point-to-point bus, what NVIDIA is proposing is for all practical purposes a socketed GPU, just with the memory and power delivery circuitry on the GPU instead of on the motherboard.

NVIDIA’s Pascal test vehicle is one such example of what a card would look like. We cannot see the connector itself, but the basic idea is that it will lay down on a motherboard parallel to the board (instead of perpendicular like PCIe slots), with each Pascal card connected to the board through the NVLink mezzanine connector. Besides reducing trace lengths, this has the added benefit of allowing such GPUs to be cooled with CPU-style cooling methods (we’re talking about servers here, not desktops) in a space efficient manner. How many NVLink mezzanine connectors available would of course depend on how many the motherboard design calls for, which in turn will depend on how much space is available.

One final benefit NVIDIA is touting is that the new connector and bus will improve both energy efficiency and energy delivery. When it comes to energy efficiency NVIDIA is telling us that per byte, NVLink will be more efficient than PCIe – this being a legitimate concern when scaling up to many GPUs. At the same time the connector will be designed to provide far more than the 75W PCIe is spec’d for today, allowing the GPU to be directly powered via the connector, as opposed to requiring external PCIe power cables that clutter up designs.

With all of that said, while NVIDIA has grand plans for NVLink, it’s also clear that PCIe isn’t going to be completely replaced anytime soon on a large scale. NVIDIA will still support PCIe – in fact the blocks can talk PCIe or NVLink – and even in NVLink setups there are certain command and control communiques that must be sent through PCIe rather than NVLink. In other words, PCIe will still be supported across NVIDIA's product lines, with NVLink existing as a high performance alternative for the appropriate product lines. The best case scenario for NVLink right now is that it takes hold in servers, while workstations and consumers would continue to use PCIe as they do today.

Meanwhile, though NVLink won’t even be shipping until Pascal in 2016, NVIDIA already has some future plans in store for the technology. Along with a GPU-to-GPU link, NVIDIA’s plans include a more ambitious CPU-to-GPU link, in large part to achieve the same data transfer and synchronization goals as with inter-GPU communication. As part of the OpenPOWER consortium, NVLink is being made available to POWER CPU designs, though no specific CPU has been announced. Meanwhile the door is also left open for NVIDIA to build an ARM CPU implementing NVLink (Denver perhaps?) but again, no such product is being announced today. If it did come to fruition though, then it would be similar in concept to AMD’s abandoned “Torrenza” plans to utilize HyperTransport to connect CPUs with other processors (e.g. GPUs)

http://www.anandtech.com/show/7900/nvidia-updates-gpu-roadmap-unveils-pascal-architecture-for-2016

 

[Smokey]

Pascal got me legit shook right now

 

[finalflame]

AFAIK Nvlink is the interposer for the "3D" memory.

I thought NVLink effectively replaces PCI-E as the BUS for the GPU? In which case all new motherboards with NVLink ports would be necessary.

 

[thuway]

I want a new GPU so bad but my 280x is holding on at 1080p and I have yet to update to a 4k monitor. Once I go 4k, I need one of them Pascal.

 

[Auto-Reply]

I didn't know who Elon Musk is.... looked him up on wiki, god damn

 

[Devildoll]

In the end there, he just said, Pascal is going to be 10 times faster than Maxwell .... in deep learning.

All in all, the Titan X seems like a decent Titan card, bummer for the DP people though.

 

[Durante]

Put together some Pascal details:

- first NVIDIA GPU with 3D memory and NVLink
- 32GB of memory
- 3x the memory bandwidth of Maxwell
- 2x the performance per watt of Maxwell
- new feature: mixed-precision
- 4x the mixed-precision FP16 performance of Maxwell

Anything to add or change?

That sounds accurate.

What you should add is that some of those, in particular the 32 GB of memory, will probably be only on professional cards that only Smokey can afford for gaming

I thought NVLink effectively replaces PCI-E as the BUS for the GPU? In which case all new motherboards with NVLink ports would be necessary.

NVlink is an interconnect. It will be used in the two largest supercomputers in the upcoming years, currently being built by IBM and NV.

 

[Alo81]

Yooooo pascal.