nvidia GTC 2016:Nvidia Pascal

[NumLock21]

 

Quote

The NVIDIA

^® Tesla^® P100 is purpose-built as the most advanced data center accelerator ever. It taps into an innovative new GPU architecture to deliver the world's fastest compute node with higher performance than hundreds of slower commodity compute nodes. Lightning-fast nodes powered by Tesla P100 accelerate time-to-solution for the world's most important challenges that have infinite compute needs in HPC and deep learning.

 

 

Nvidia has announced their Tesla P100 graphic accelerator based on the newest Pascal architecture. The chip houses 15.3 billion transistors, built on the 16nm FinFET process, 16GB of CoWoS HBM2 stacked memory, giving a total memory bandwidth of 720 GB/s. With Nvidia's GPU Boost Technology, the Tesla P100 can achieve

* 5.3 TeraFLOPS double-precision performance with NVIDIA GPU Boost

* 10.6 TeraFLOPS single-precision performance with NVIDIA GPU Boost

* 21.2 TeraFLOPS half-precision performance with NVIDIA GPU Boost

And its NVLink provides up to 160 GB/s interconnect between the gpus. Other features includes ECC and server optimization.

Nvidia's Pascal architecture it's said to support their upcoming cuDNN v5  (CUDA Deep Neural Network or cuDNN) library. This new update includes the following:

• High Performance Deep Neural Network Training
• Accelerates Deep Learning: Caffe, CNTK, Tensorflow, Theano, Torch
• Performance continues to improve over time

The cuDNN v5 library, available sometime in April, with the Pascal architecture should improve Deep Learning and AI performance, compare to its predecessor by 70%.

Sadly, there is no Pascal announcement on the consumer side.  Maybe we'll get that, probably 2 months later, during Computex?

Source

http://wccftech.com/nvidia-pascal-gpu-deep-neural-ai-performance/

http://wccftech.com/nvidia-pascal-tesla-p100-gp100-gpu/

Tesla P100

http://www.nvidia.com/object/tesla-p100.html

Deep Learning

https://developer.nvidia.com/deep-learning

Pic credit

http://anandtech.com/show/10225/the-nvidia-gtc-2016-live-blog

 

[Mr.Meerkat]

2 hours and 15 mins till it starts

[goaway]

3 minutes ago, NumLock21 said:

Nvidia is going to reveal their upcoming pascal gpus today at gtc 2016. The card has a 70% improvement in deep neural network and ai. 

Will update when more info comes out later today.

Source?

[MrTiC]

Is there gonna be a live presentation online ? if so link pls so i can watch it 

[Guest]

1 minute ago, larsi239 said:

Source?

Nvidia email sent out a day or two ago

[Thoaee1993]

anywhere this will be live streamed by chance?

 

[VagabondWraith]

Can't wait!!! ???

[Mirdon]

Damn! anyone has a source? screw x2 exp and loot, I'm gonna watch this.

[Guest]

I LOVE how incredibly easy and safe it is to sell a Titan X for a good price online in a reasonable amount of time... wait

[Guest]

1 minute ago, Mirdon said:

Damn! anyone has a source? screw x2 exp and loot, I'm gonna watch this.

It was in an email sent out by Nvidia

[Mirdon]

Just now, TheCovertCamper said:

It was in an email sent out by Nvidia

But I'm not subscribed to Nvidia .

[TrigrH]

Just now, Mirdon said:

But I'm not subscribed to Nvidia .

 

4 minutes ago, Thoaee1993 said:

anywhere this will be live streamed by chance?

 

 

4 minutes ago, MrTiC said:

Is there gonna be a live presentation online ? if so link pls so i can watch it 

here i think: http://www.ustream.tv/channel/fWbQyaEMfbh

[goaway]

3 minutes ago, TheCovertCamper said:

It was in an email sent out by Nvidia

Do you have a link? Now I'm curous  

[MrTiC]

1 minute ago, TrigrH said:

 

 

here i think: http://www.ustream.tv/channel/fWbQyaEMfbh

Cheers man, much apriciated

[Mirdon]

2 minutes ago, TrigrH said:

 

 

here i think: http://www.ustream.tv/channel/fWbQyaEMfbh

Oh wow, thank you very much.

[Humbug]

I guess the focus will be on compute performance?

[TrigrH]

1 minute ago, gFrenken97 said:

Are we talking actual cards? Or just the architecture

Quote

Attention graphics card fanatics: at approximately 2AM AEST, Nvidia will be kicking off its keynote address for GTC 2016; otherwise known as the premiere event for GPU developers. This year, the focus is on artificial intelligence, virtual reality and self driving cars along with some big hardware announcements that we’re not allowed to talk about (yet). Once again, Gizmodo will be blogging from the event live. 

http://www.gizmodo.com.au/2016/04/nvidia-gtc-2016-keynote-live-blog/

[Notional]

That 70% increase is almost #CEO math. It's FP16, which is half precision, not full precision (FP32) and only a quarter of double precision (FP64). I hope Jen Sun doesn't do his bs #CEOmath with 10x performance increase on some obscure compute scenario, telling more of the weakness of Maxwell than the strength of Pascal.

 

Either way it's going to be interesting to see what Pascal has to offer, and more importantly, when. Especially Pascal's usage of HBM is going to be interesting. Are they going to use it? On what models, which generation, capacity, etc. This is very interesting considering Polaris won't have it, and that we will have to wait for Vega for HBM2.

[Aytex]

3 minutes ago, Notional said:

That 70% increase is almost #CEO math. It's FP16, which is half precision, not full precision (FP32) and only a quarter of double precision (FP64). I hope Jen Sun doesn't do his bs #CEOmath with 10x performance increase on some obscure compute scenario, telling more of the weakness of Maxwell than the strength of Pascal.

 

Either way it's going to be interesting to see what Pascal has to offer, and more importantly, when. Especially Pascal's usage of HBM is going to be interesting. Are they going to use it? On what models, which generation, capacity, etc. This is very interesting considering Polaris won't have it, and that we will have to wait for Vega for HBM2.

"10 times"

  lol

[TrigrH]

4 minutes ago, Notional said:

 I hope Jen Sun doesn't do his bs #CEOmath with 10x performance increase on some obscure compute scenario, telling more of the weakness of Maxwell than the strength of Pascal.

don't ask for the impossible, he ALWAYS does it.

[AHaskin14]

The hype is real! show me what I'm buying this summer!

[Prysin]

PREPÅAAAAAAAARE YOUR ANUS

and prepare to be dissappointed.

[Ashaira]

22 minutes ago, NumLock21 said:

The card has a 70% improvement in deep neural network and ai. 

but but...... 10x maxwell............

[NumLock21]

Official live stream

http://www.gputechconf.com/

[Notional]

3 minutes ago, Aytex said:

"10 times"

  lol

There's no sad face option.. If I give you a funny, can we just agree it represents tragicomic?