Nvidia Unveils Pascal Gpu: 16gb Of Memory, 1tb/s Bandwidth

[Pugs501]

At the Japanese edition of NVIDIA GTC (GPU Technology Conference), NVIDIA finally revealed details behind its 2016 graphics architecture, codenamed Pascal. The architecture was launched at the main GTC event, which took place in San Jose on March 17th, 2015 (watch Jen-Hsun Huang’s GTC keynote here). GTC Japan was hosted by Marc Hamilton.

 

As always, the Pascal GPU will be manufactured in Taiwan Semiconductor Manufacturing Company (TSMC), using the brand new 16nm FinFET process. This process is much more than a simple number, since it marks the shift from planar, 2D transistors to the FinFET i.e. 3D transistors. This shift required that the engineers make lot of changes in the thought process, and should result in significant power savings.

 

NVIDIA Pascal proof-of-concept Engineering Board with 4GB HBM memory. Retail products will carry 16GB of HBM2 memory.

 

But that is just the beginning, as Pascal will bring support for up to 32GB of HBM2 memory. However, the actual products based on Pascal will launch with 16GB HBM2 memory, and more memory will depend solely on memory vendors such as SK.Hynix and Samsung. What is changing the most is bandwidth. Both the Kepler-based Tesla (K40) and Maxwell-based M4/M40 featured 12GB of GDDR5 and achieved up to 288GB/s of memory bandwidth. Those 16GB HBM SDRAM (packed in four 4GB HBM2 chips) will bring 1TB/s in bandwidth, while internally the GPU surpasses the 2TB/s barrier.

 

NVIDIA’s Marc Hamilton said: “Using 3D memory, not only the memory capacity will go up, the memory bandwidth will go up significantly. With a much faster GPU, and higher memory bandwidth, the existing interconnects in the server are just plain outdated. So, we had to develop our own interconnect called NVLink, five times faster than existing technology.”

 

Pascal will also be available in multi-GPU packaging, replacing the Tesla K80 (NVIDIA skipped Maxwell-gen dual-GPU Tesla). Combined figures are very interesting to compare – 24GB GDDR5 and 480GB/s bandwidth should be replaced with 32GB HBM2 and 2TB/s bandwidth, mutually connected through NVLink rather than PCIe. The NVLink will enable up to 80GB/s, which should replace PLX PCIe Gen3 bridge chips that can only support 16GB/s (8GB/s per GPU). This part should be ‘warm up’ for 2018 and the Volta architecture.

 

Pascal GPU architecture highlights: Higher performance through 16nm process and GPU Boost, stacked 2.5D / 3D memory, NVLink and mixed-mode calculation.

 

Unfortunately, the company did not disclose how much would ECC (Error Correcting Code) reduce the memory performance and overhead, but that is something all HBM-powered products will have to deal with. In any case, the company is gearing for a battle with Intel Xeon Phi, which in its recent incarnation is becoming quite the competitor. Still, Pascal is expected to deliver double-digit single-precision TFLOPS performance, and a lot of focus will be placed on so-called mixed-mode precision (INT8, FP16 and FP32).

 

Pascal is expected to hit the market during the first half of 2016.

 

Article: http://vrworld.com/2015/11/16/nvidia-unveils-pascal-gpu-16gb-of-memory-1tbs-bandwidth/

Author: http://vrworld.com/author/theo_valich/

[TechGod]

WHAT!

[Altecice]

 

 

mutually connected through NVLink rather than PCIe. The NVLink will enable up to 80GB/s, which should replace PLX PCIe Gen3 bridge chips that can only support 16GB/s (8GB/s per GPU).

 

So is this an SLI bridge "upgrade" or am I reading it wrong.... So no more PCI-E ?

 

EDIT: Right so Nvlink is only for the supercomputer side of things, for Desktop it will still use PCI-E.

 

[eLucid]

Aww yiss! Just take my money...

[jollander]

So 4k is def a thing next year?

[patrickjp93]

The NVLink figure is incorrect. It's 200GB/s

https://m.youtube.com/watch?v=bNn8nltUf5o

[babadoctor]

[mikat]

cool

[AresKrieger]

[SurvivorNVL]

I'll take two.

[patrickjp93]

So is this an SLI bridge "upgrade" or am I reading it wrong.

I was reading it wrong... So no more PCI-E ?

The author is confused. The only CPU-GPU NVLink interconnect is used solely on IBM's systems. The card will still attach via the PCIe bus. The GPU-GPU interconnect is 200GB/s, dwarfing SLI completely.

[Enderman]

Im upgrading my GPU when pascal comes out.

[Swndlr]

Im upgrading my GPU when pascal comes out.

Same bb <3

[MattLantis]

c-c-c-can I have one???

[Pugs501]

So 4k is def a thing next year?

Imagine using 3 of these in the future :3

[Altecice]

The author is confused. The only CPU-GPU NVLink interconnect is used solely on IBM's systems. The card will still attach via the PCIe bus. The GPU-GPU interconnect is 200GB/s, dwarfing SLI completely.

 

After a bit of read that's the same conclusion I have come to. 

[Pugs501]

The NVLink figure is incorrect. It's 200GB/s

What's their Sources?

[SurvivorNVL]

-Snip-

I just want a 32" 4k, G-Sync OLED monitor.  That.. -that- would be AMAZING.

[The_Stig]

*Insert post about nvdia using one of AMD's best features and AMD using CUDA*

[Jerakl]

Would still not buy because of horrible drivers.

[Weak1ings]

Next GPU:  GTX 1070/80.

Faster than my 660 SLI

[jollander]

Imagine using 3 of these in the future :3

 

My wallet doesn't want me imagining that. Brb, need to wipe off all the drool from my face.

[Weak1ings]

Would still not buy because of horrible drivers.

Who has better drivers?

AMD still does not.

Intel...  If you are able to use this then a GPU doesn't matter...

I'm curious.

[Albatross]

It sounds like a good thing.

 

 

I'll be trying to get Pascal for sure.

[JebKerman]