nvidia GTC 2016:Nvidia Pascal

[Colonel_Gerdauf]

That is one hideous race car.

[i_build_nanosuits]

nvidia racing now...roborace...and now what...DX12 support?

[goaway]

And its done....

[MrTiC]

1 minute ago, i_build_nanosuits said:

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!

Model T-100

Model (T)esla P-100

....

Illuminati ?

[Thony]

6 minutes ago, i_build_nanosuits said:

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

Yea 0 gaming stuff  i only watched it to see Pascal news but all we got is server side tech.

Edit: hoild on, its 3 day event ??? So there will be Gaming stuff ! Fuck yea ! Shame I wasted my time with todays stream though lol

[panzerK]

28 minutes ago, Notional said:

That was the U2 rack server with several chips in it. The actual Pascal chip used has 16 billion transistors with single precision of 10,6TF.

RIP GTX news...

[Mr_Troll]

its over ... no new pascal graphic cards anouncements... rip

[BonSie]

No Pascal Demo.That doesn't sound very promising with such an event.

[ivhyn]

Now that was a giant waste of my time.

[i_build_nanosuits]

3 minutes ago, Thony said:

Yea 0 gaming stuff  i only watched it to see Pascal news but all we got is server side tech.

Well...not really...you now know that if need be (if AMD put their shit togheter for once) nvidia could potentially release a GPU with 12GB of HBM2 memory, 15 billion transistors and 21.2Tflops of computing horsepower on a single chip that would be nearly twice as energy efficient as a big maxwell chip pretty cool stuff...now let's see how bad they nerf it so it's just a tiny bit faster than the fastest next gen card from team red, as it's always the case!

[Thony]

1 minute ago, i_build_nanosuits said:

Well...not really...you now know that if need be (if AMD put their shit togheter for once) nvidia could potentially release a GPU with 12GB of HDM2 memory, 15 billion transistors and 21.2Tflops of computing horsepower on a single chip that would be nearly twice as energy efficient as a big maxwell chip pretty cool stuff...now let's see how bad they nerf it so it's just a tiny bit faster than the fastest next gen card from team red, as it's always the case!

They are behind with production of HBM2 (not sure if there isnt enough of it since they both get it from same source ?! Or something happened lately that slowed them down (an accident in one of manufacturers ?!). Last month they said there will be GDDR5 and GDDR5X GPUs coming out and then later this year or beginning of next year they will release HBM2.

[Notional]

8 minutes ago, i_build_nanosuits said:

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

Probably, but that would have been their P100 based deep learning U2 rack computer:

 

[i_build_nanosuits]

WCCFtech posted:

 

NVIDIA has officially unveiled the Pascal based Tesla P100 GPU which is their fastest GPU to date. The Pascal GP100 chip is NVIDIA’s first GPU to be based on the latest 16nm FinFET process node which delivers 65 percent higher speed, around 2 times the transistor density increase and 70 percent less power than its 28HPM tech. The new FinFET process allows NVIDIA to gain up to 2 times the performance per watt improvement on Pascal compared to the Maxwell GPUs.

 

15 Billion Transistors on a 600mm2 16nm Die – Full HBM2 Memory Design With Insane Compute Fastest Compute Performance in History on a Single Graphics Board

[Paragon_X]

"It’s the first GPU built on  NVIDIA’s 11th generation Pascal architecture. It’s built to blaze new frontiers for deep learning applications in hyperscale data centers. It’s got muscle—150 billion transistors, with 20 teraflops of half-precision performance."

I don't really care about performance of this card as it is for pro applications but stacking 150B transistors we essentially have the first 3D chip and its fucking amazing

[i_build_nanosuits]

2 minutes ago, Paragon_X said:

"It’s the first GPU built on  NVIDIA’s 11th generation Pascal architecture. It’s built to blaze new frontiers for deep learning applications in hyperscale data centers. It’s got muscle—150 billion transistors, with 20 teraflops of half-precision performance."

I don't really care about performance of this card as it is for pro applications but stacking 150B transistors we essentially have the first 3D chip and its fucking amazing

GP100 is composed of an array of Graphics Processing Clusters (GPCs), Streaming Multiprocessors (SMs), and memory controllers. GP100 achieves its colossal throughput by providing six GPCs, up to 60 SMs, and eight 512-bit memory controllers (4096 bits total). The Pascal architecture’s computational prowess is more than just brute force: it increases performance not only by adding more SMs than previous GPUs, but by making each SM more efficient. Each SM has 64 CUDA cores and four texture units, for a total of 3840 CUDA cores and 240 texture units.

 

On compute side, Pascal is going to take the next incremental step with double precision performance rated over 5.3 TFLOPs, which is more than double of what’s offered on the last generation FP64 enabled GPUs. As for single precision performance, we will see the Pascal GPUs breaking past the 10 TFLOPs barrier with ease. The chip comes with 4 MB of L2 cache. The GPU is in volume production and will be arriving to HPC markets very soon.

[dalekphalm]

1 hour ago, El Diablo said:

wow u realy need to upgrade that gpu

Why is that? I have a 1680x1050 monitor and I can still play pretty much every game maxed out (or nearly) and still get 60 fps.

 

With the way GPU prices are in Canada right now, I'd have to spend easily $400-$500 CAD to get a better performing card. Even a R9 380 is around $300, which would BARELY be an upgrade (We're talking a few percent better).

 

I'm waiting for Polaris at best, since I don't need to upgrade now, and when I do upgrade, I want a significant one.

[Kunter]

This 150B transistor thing still baffles me. It was there on the slide, but it makes no sense. 15B would be perfectly fine for the increase in computational power compared to Titan X (6B->15B, 6Tflops->10.5Tflops) but then his comment in stacking layers makes no sense at all, because you could just do one flat layer with 15B transistors...

[i_build_nanosuits]

1 minute ago, Kunter said:

This 150B transistor thing still baffles me. It was there on the slide, but it makes no sense. 15B would be perfectly fine for the increase in computational power compared to Titan X (6B->15B, 6Tflops->10.5Tflops) but then his comment in stacking layers makes no sense at all, because you could just do one flat layer with 15B transistors...

he wouldnt do such a typo on stage would he? i'm 100% sure he mentionned 150 billion transistors...but tech sites right now all seem to be reporting 15billion.

[alamox]

5 minutes ago, Kunter said:

This 150B transistor thing still baffles me. It was there on the slide, but it makes no sense. 15B would be perfectly fine for the increase in computational power compared to Titan X (6B->15B, 6Tflops->10.5Tflops) but then his comment in stacking layers makes no sense at all, because you could just do one flat layer with 15B transistors...

jen added the 32GoHBM2 transistor counts to the total, so i wouldnt really focus on that number, it's just there for show

 

the compute numbers of the tesla P100 are too low for a 16nm 600mm² chip, nvidia seem to have density issue with the process, and the problem doesnt seem to be there because of the size of the chip, otherwise they would have went in a comprimise route between perf/size, would give better yield , cheaper and more efficient chips, meaning that the problem might be there for the consumer GPU also, i dont know about you but the telsa P100 announcement, seemed like a bad news for pascal to me.

and another bad news would be the time to refine the process, if they start to stack pile the chip from now to get ready for next year, that shows that they have no reason to believe that the process will get better, so they ignore the cost difference between a 600mm² chip made today and another made next year, seriously i see bad news all around, i hope im wrong.

[i_build_nanosuits]

2 minutes ago, alamox said:

 i dont know about you but the telsa P100 announcement, seemed like a bad news for pascal to me.

say what?!

GP100 is composed of an array of Graphics Processing Clusters (GPCs), Streaming Multiprocessors (SMs), and memory controllers. GP100 achieves its colossal throughput by providing six GPCs, up to 60 SMs, and eight 512-bit memory controllers (4096 bits total). The Pascal architecture’s computational prowess is more than just brute force: it increases performance not only by adding more SMs than previous GPUs, but by making each SM more efficient. Each SM has 64 CUDA cores and four texture units, for a total of 3840 CUDA cores and 240 texture units.

 

On compute side, Pascal is going to take the next incremental step with double precision performance rated over 5.3 TFLOPs, which is more than double of what’s offered on the last generation FP64 enabled GPUs. As for single precision performance, we will see the Pascal GPUs breaking past the 10 TFLOPs barrier with ease. The chip comes with 4 MB of L2 cache. The GPU is in volume production and will be arriving to HPC markets very soon.

[Castdeath97]

So can anyone give me a TL DR for the event since the OP hasn't updated his post with anything.

[Kinda Bottlenecked]

23 minutes ago, i_build_nanosuits said:

Well...not really...you now know that if need be (if AMD put their shit togheter for once) nvidia could potentially release a GPU with 12GB of HBM2 memory, 15 billion transistors and 21.2Tflops of computing horsepower on a single chip that would be nearly twice as energy efficient as a big maxwell chip pretty cool stuff...now let's see how bad they nerf it so it's just a tiny bit faster than the fastest next gen card from team red, as it's always the case!

 

A thread about your god and you have to go ruin it. Com'on this is suppose to be a thread about Nvidia goodness. 

 

You're last person I expect to mention AMD on a big Nvidia thread. Even worse that you're bashing AMD for the for the lousiest of reasons.

 

6 minutes ago, Castdeath97 said:

So can anyone give me a TL DR for the event since the OP hasn't updated his post with anything.

 

*servers*

*fast compute cards*

*self driving cars*

*Deep learning*

*AI Formula E* 

That's all I took out of it.

 

I'm mostly interested in the formula E thing. But close to nothing that will interest anyone who is only there for the gaming stuff.  

[Castdeath97]

1 minute ago, Pohernori said:

 

A thread about your god and you have to go ruin it. Com'on this is suppose to be a thread about Nvidia goodness. 

 

You're last person I expect to mention AMD on a big Nvidia thread. Even worse that you're bashing AMD for the for the lousiest of reasons.

 

 

*servers*

*fast compute cards*

*self driving cars*

*Deep learning*

*AI Formula E* 

 

I'm mostly interested in the formula E thing. But close to nothing that will interest anyone who is only there for the gaming stuff.  

So Nvidia still haven't gotten bored from trying to fit tegra's in cars, *sigh*. All these rumours about mobile GPUs coming out turned out to be false then.

[ivhyn]

26 minutes ago, i_build_nanosuits said:

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

Went back and listened to the vod.

He says one layer of the chip has 15 billion transistors (GP100?), but by stacking them over eachother, it has a total of 150 billion transistors (10 layers, Tesla P100).

[i_build_nanosuits]

yeah, one post above it's all there...these are the specs for the testla P100 that will start shipping soon for high performance computing...and will most likely be the next big pascal titan gaming GPU for next year of something.

And all the rest is about deep learning, neuro networks...and other skynet related stuff