nvidia GTC 2016:Nvidia Pascal

[Notional]

Just now, i_build_nanosuits said:

no...the server with 8 GPU,s had a capacity of 170 Tflops...the teslta card is 21.2 tflops...single GPU.

Yes in half precision. That's still just 10.5 single precision. Do you honestly believe it's 21.2 double? That would make it a minimum of 42.4 TF single precision. That's impossible with 16Billion transistors.

[MEC-777]

I have to go into a meting in 10 mins. So if he's going to announce anything, he better do it soon... lol

[i_build_nanosuits]

7 minutes ago, Notional said:

Yes in half precision. That's still just 10.5 single precision. Do you honestly believe it's 21.2 double? That would make it a minimum of 42.4 TF single precision. That's impossible with 16Billion transistors.

it's a 150billion transistor GPU...the testla P100 is a 150b. transistor part...i'm 100% sure about that.

it's a beast...but don't sweat just yet this is not a consumer grade product it's meant for deep learning and neuronetworks (in cased you missed that part, it's not the next gaming GPU).

 

[DocSwag]

14 minutes ago, i_build_nanosuits said:

it's still the fastest computing device on the face of planet earth by a factor of at least x3?

So not true.

Let's look at half precision first:

P100 has 21 tflops half precision

Radeon Pro Duo has 16 tflops single and supports half precision do I'm assuming it has 16 tflops half precision.

That's 1.33 times

 

At single:

P100 is 10 tflops

Radeon pro duo is 16 tflops

Radeon pro duo wins by 1.66 times

 

At double: 

P100 is 5 tflops

Firepro w9100 (or whatever the firepro 290x is called) is 2.5 tflops

P100 wins by 2 times

 

Don't see where you got that three times.

And you keep saying p100 is 21 tflops double precision, which is wrong. It has 21 tflops half precision and only 5 tflops double.

[Thony]

What a waste of time this presentation has been.....

[NumLock21]

$129,000 for DGX-1

I was expect more front USB 3 ports, that thing only got 2 and lacks cup holders.

 

[i_build_nanosuits]

2 minutes ago, DocSwag said:

So not true.

Let's look at half precision first:

P100 has 21 tflops half precision

Radeon Pro Duo has 16 tflops single and supports half precision do I'm assuming it has 16 tflops half precision.

That's 1.33 times

 

At single:

P100 is 10 tflops

Radeon pro duo is 16 tflops

Radeon pro duo wins by 1.66 times

 

At double: 

P100 is 5 tflops

Firepro w9100 (or whatever the firepro 290x is called) is 2.5 tflops

P100 wins by 2 times

 

Don't see where you got that three times.

And you keep saying p100 is 21 tflops double precision, which is wrong. It has 21 tflops half precision and only 5 tflops double.

how about perf per watt though? and how about the fact that it's a single piece of silicon instead of two monstruous 300W GPU's?

[elago]

"you have heard today about deep learning" 

 

Oh, that's for sure

[Thony]

Whoever wrote the script for the presenter needs to be fired. I heard words "deep learning" in every other sentence for 1 hour straight......

[AlexGoesHigh]

You can buy the DXG-1 rack here, if you 130k as spare cash... http://www.nvidia.com/object/deep-learning-system.html

[Van Weapon]

Just now, elago said:

"you have heard today about deep learning" 

 

Oh, that's for sure

He says it so often its like a Tourettes tic 

[i_build_nanosuits]

1 minute ago, Thony said:

Whoever wrote the script for the presenter needs to be fired. I heard words "deep learning" in every other sentence for 1 hour straight......

deep learning is the future though...it's very important...if you want cars that drive autonomously and shit like that we need this...it's the future of humanity.

[MrTiC]

All i can make out of this hocus pocus talk is "DEEP"  And not the kind their talking about 

[Notional]

4 minutes ago, i_build_nanosuits said:

it's a 150billion transistor GPU...the testla P100 is a 150b. transistor part...i'm 100% sure about that.

 

You must be trolling?

4 minutes ago, DocSwag said:

So not true.

Let's look at half precision first:

P100 has 21 tflops half precision

Radeon Pro Duo has 16 tflops single and supports half precision do I'm assuming it has 16 tflops half precision.

That's 1.33 times

 

At single:

P100 is 10 tflops

Radeon pro duo is 16 tflops

Radeon pro duo wins by 1.66 times

 

At double: 

P100 is 5 tflops

Firepro w9100 (or whatever the firepro 290x is called) is 2.5 tflops

P100 wins by 2 times

 

Don't see where you got that three times.

And you keep saying p100 is 21 tflops double precision, which is wrong. It has 21 tflops half precision and only 5 tflops double.

Half precision should be double og single precision. As for double, look at the Pro Duo, not the Hawaii based cards.

1 minute ago, i_build_nanosuits said:

how about perf per watt though? and how about the fact that it's a single piece of silicon instead of two monstruous 300W GPU's?

Pascal would obviously be better at 16nmFF than a 28nm part. That goes without saying. But this chip is not out yet. So it should not be better than an AMD 14nmFF FirePro part, once those show up.

[Thony]

1 minute ago, i_build_nanosuits said:

deep learning is the future though...it's very important...if you want cars that drive autonomously and shit like that we need this...it's the future of humanity.

Fair enough, but im sure if they can produce such amazing technology they can hire someone who can play with words and sentences better than "bla bla bla deep learning, deep learning, bla bla bla deep learning..."

U dont have to mention what u talk about every 2 minutes.... Imagine if u did that at school for an essay. U would probably be told to stop repeating it.

[i_build_nanosuits]

1 minute ago, Notional said:

 

 

You must be trolling?

 

yeah no you're right it's 15billion...my bad i see it on wccf tech now...wow what a slow ass POS then...all this fuzz about it.

[i_build_nanosuits]

2 minutes ago, Thony said:

Fair enough, but im sure if they can produce such amazing technology they can hire someone who can play woith words and sentences better than "bla bla bla deep learning, deep learning, bla bla bla deep learning..."

well...it's because the ''blah blah blah'' parts are the parts you did not understand...isn't it?

[Thony]

Just now, i_build_nanosuits said:

well...it's because the ''blah blah blah'' parts are the parts you did not understand...isn't it?

It was an example, i wasnt litteral. My point is he said "deep learning" more times than necessary so all i could hear was "deep learning"...

[Frank6247]

OMG two pascals in the board!!! That's it guys we can go home.

 

[Notional]

Just now, i_build_nanosuits said:

yeah no you're right it's 15billion...my bad i see it on wccf tech now...wow what a slow ass POS then...all this fuzz about it.

Well the AMD Duo Pro is a dual GPU part, where this is a single chip part, so it's definitely a large step up, no doubt. But based on what we know of Vega, that should beat the P100 significantly. But of course it's all speculation until the chips are actually on the market.

 

Also 150b transistors, would make the chip the size of an ITX motherboard or something like that

[i_build_nanosuits]

1 minute ago, Notional said:

Also 150b transistors, would make the chip the size of an ITX motherboard or something like that

i'm pretty sure he mentionned ''a hundred and fifty billion transistor'' at some point though...didnt he?

[i_build_nanosuits]

4 minutes ago, Thony said:

It was an example, i wasnt litteral. My point is he said "deep learning" more times than necessary so all i could hear was "deep learning"...

i would have liked it better if he would have had the gaming GPU and showed some of the latest games running at ultra 4k settings 60FPS locked...but it didn't happened maybe next time lol

[MrTiC]

Hello officer ... Yes my car is drunk Kappa

 

This is about the AI driving 

[i_build_nanosuits]

At this rate nvidia will be held responsible for skynet before the end of 2017...that deep learning and newtwork training...Davenet they call it? omfg...skynet would have been more appropriated since we are already familiar with the name...you know!

[Thony]

Gotta give them cudos for KnightRider theme song