Nvidia Turing Announced

[S w a t s o n]

LIVESTREAM OVER!

 

 

We were shown a new Quadro RTX card that ran the star wars demo on ONE card. (Previously 4x Tesla V100) Also, a new smaller NVLINK bridge running at 100GB (not gb) per second.

 

Spoiler

 

 

 

 

NVLINk Connector (Fits in his palm but it's in his jacket pocket this pic)

 

Render Farm Comparison

Spoiler

 

(A small one according to jensen)

 

Equivalent render farm on turing

 

RTX Comparison

Spoiler

RTX OFF

RTX ON

 

Press Release

Spoiler

SIGGRAPH — NVIDIA today reinvented computer graphics with the launch of the NVIDIA Turing™ GPU architecture.

 

The greatest leap since the invention of the CUDA GPU in 2006, Turing features new RT Cores to accelerate ray tracing and new Tensor Cores for AI inferencing which, together for the first time, make real-time ray tracing possible.

 

These two engines — along with more powerful compute for simulation and enhanced rasterization — usher in a new generation of hybrid rendering to address the $250 billion visual effects industry. Hybrid rendering enables cinematic-quality interactive experiences, amazing new effects powered by neural networks and fluid interactivity on highly complex models.

 

The company also unveiled its initial Turing-based products — the NVIDIA® Quadro® RTX™ 8000, Quadro RTX 6000 and Quadro RTX 5000 GPUs — which will revolutionize the work of some 50 million designers and artists across multiple industries.

 

“Turing is NVIDIA’s most important innovation in computer graphics in more than a decade,” said Jensen Huang, founder and CEO of NVIDIA, speaking at the start of the annual SIGGRAPH conference. “Hybrid rendering will change the industry, opening up amazing possibilities that enhance our lives with more beautiful designs, richer entertainment and more interactive experiences. The arrival of real-time ray tracing is the Holy Grail of our industry.”

 

NVIDIA’s eighth-generation GPU architecture, Turing enables the world’s first ray-tracing GPU and is the result of more than 10,000 engineering-years of effort. By using Turing’s hybrid rendering capabilities, applications can simulate the physical world at 6x the speed of the previous Pascal™ generation.

 

To help developers take full advantage of these capabilities, NVIDIA has enhanced its RTX development platform with new AI, ray-tracing and simulation SDKs. It also announced that key graphics applications addressing millions of designers, artists and scientists are planning to take advantage of Turing features through the RTX development platform.

 

“This is a significant moment in the history of computer graphics,” said Jon Peddie, CEO of analyst firm JPR. “NVIDIA is delivering real-time ray tracing five years before we had thought possible.”

 

Real-Time Ray Tracing Accelerated by RT Cores
The Turing architecture is armed with dedicated ray-tracing processors called RT Cores, which accelerate the computation of how light and sound travel in 3D environments at up to 10 GigaRays a second. Turing accelerates real-time ray tracing operations by up to 25x that of the previous Pascal generation, and GPU nodes can be used for final-frame rendering for film effects at more than 30x the speed of CPU nodes.

“Cinesite is proud to partner with Autodesk and NVIDIA to bring Arnold to the GPU, but we never expected to see results this dramatic,” said Michele Sciolette, CTO of Cinesite. “This means we can iterate faster, more frequently and with higher quality settings. This will completely change how our artists work.”

 

AI Accelerated by Powerful Tensor Cores 
The Turing architecture also features Tensor Cores, processors that accelerate deep learning training and inferencing, providing up to 500 trillion tensor operations a second.

This level of performance powers AI-enhanced features for creating applications with powerful new capabilities. These include DLAA — deep learning anti-aliasing, which is a breakthrough in high-quality motion image generation — denoising, resolution scaling and video re-timing.

These features are part of the NVIDIA NGX™ software development kit, a new deep learning-powered technology stack that enables developers to easily integrate accelerated, enhanced graphics, photo imaging and video processing into applications with pre-trained networks.

 

Faster Simulation and Rasterization with New Turing Streaming Multiprocessor
Turing-based GPUs feature a new streaming multiprocessor (SM) architecture that adds an integer execution unit executing in parallel with the floating point datapath, and a new unified cache architecture with double the bandwidth of the previous generation.

Combined with new graphics technologies such as variable rate shading, the Turing SM achieves unprecedented levels of performance per core. With up to 4,608 CUDA cores, Turing supports up to 16 trillion floating point operations in parallel with 16 trillion integer operations per second.

Developers can take advantage of NVIDIA’s CUDA 10, FleX and PhysX SDKs to create complex simulations, such as particles or fluid dynamics for scientific visualization, virtual environments and special effects.

 

Availability
Quadro GPUs based on Turing will be initially available in the fourth quarter.

Some notes from the conference:

• RTX makes use of both Tensor Cores and the new RT Cores meaning we're probably gonna see them on consumer post-volta GPUs
• Highest RTX Quadro card will support a 96GB frame buffer with NVLink
• Turing Memory is running at 14GHz (Jensen stood in the way of my screenshot sorry)
• Star wars demo ran at real-time with 5 rays per pixel
• No word on PATH TRACING which is the actual holy grail but let's not get greedy
• 4AM Bonus: i was wondering what frame rate the real time star wars demo was running at since it was a single card. Nvidia gave us the compute time per frame for that at 45ms which is approx 22FPS, not bad at all

Stay tuned to the Gamescom nvidia event for likely RTX Titan or RTX 2080?

[PlayStation 2]

Should update the title to Turing instead of Turning.

[Cedimedi]

Prices:

RTX8000 - 10k

RTX6000 - 6.3k

RTX5000 - 2.3k

 https://www.cnet.com/news/nvidia-turing-architecture-arrives-in-high-end-quadro-rtx-gpus/

 

I wonder if the new GeForce cards will have such a huge RT Core too. That most likely is alot of the die.. when it's not in use by most games thats alot of wasted space if it can't process anything else except raytracing

 

[Trixanity]

Nvidia is really destroying everyone right now and it's interesting that they've created these more specialized units to massively increase performance in specific scenarios. 

 

I'm pretty sure they'll tear out those units for consumer cards and that this is the year they'll completely split the consumer lineup from the rest. AMD has said it'll be necessary so it'll happen if you're in the high performance GPU industry.

 

I'm imagining they'll remove the tensor and rt cores as well as the nvlink while keeping the other blocks similar or identical (SM and memory/cache,  encode/decode and display controller). Of course putting PCI-E on there instead. Obviously from there it's a matter of scaling down in number of units and corresponding features for a balanced chip.

[S w a t s o n]

2 minutes ago, Trixanity said:

I'm pretty sure they'll tear out those units for consumer cards and that this is the year they'll completely split the consumer lineup from the rest. AMD has said it'll be necessary so it'll happen if you're in the high performance GPU industry.

 

The RT cores are needed for RTX so the consumer cards will have them just maybe not this gen

[S w a t s o n]

As this goes on it seems like AI based factors (like denoising probably) are part of RTX so we might even see tensor cores AND RTX cores on consumer cards eventually.

[DildorTheDecent]

oof

[Sakkura]

In any case, 754 mm² is huge. Even if that goes consumer, it'll be super expensive.

[Trixanity]

17 minutes ago, Swatson said:

The RT cores are needed for RTX so the consumer cards will have them just maybe not this gen

Perhaps. I don't imagine it'll be the same amount. Takes too much die space from what I can tell.

 

Nvidia claims the Pascal and Volta cards can do RTX in some fashion so perhaps SM changes are good enough to avoid dedicated hardware.

[S w a t s o n]

Just now, Trixanity said:

Perhaps. I don't imagine it'll be the same amount. Takes too much die space from what I can tell.

 

Nvidia claims the Pascal and Volta cards can do RTX in some fashion so perhaps SM changes are good enough to avoid dedicated hardware.

Not pascal, volta only and that's via the tensor cores

[Shreyas1]

https://developer.nvidia.com/rtx

 

So I found this

[S w a t s o n]

HE SAID IT AGAIN

"The more you buy, the more you save"

[Trixanity]

2 minutes ago, Swatson said:

Not pascal, volta only and that's via the tensor cores

Just checked. That's true. Was misremembering the press briefing. Guess that's why one shouldn't be posting before bed.

[S w a t s o n]

If they drop RTX Titan at Gamescom I'm gonna lose my mind (not that any games use it yet, Metro Exodus HYPE)

[Shreyas1]

They're looking at new games now or something

 

 

EDIT: Srsly a new Quadro

Edited August 14, 2018 by Shreyas1

[S w a t s o n]

Livestream over I will try to make the OP look a bit nicer

[The Benjamins]

REAL TIME

10 GIGARAYS

16 TIPS

[Commodus]

Curious to see how the consumer cards fare.  And just watch as you get a GPU this powerful in a mainstream PC in a few years.

[rawrdaysgoby]

YAWN 

[S w a t s o n]

Guys remember that Nvidia is hosting a "Geforce Gaming Celebration" at Gamescom in a week. We already had rumors of a new RTX Titan, RTX 2080 and RTX 2070 before this announcement confirmed RTX anything.

 

Edit: I am going to guess those are going to be VOLTA based though, tensor cores only which does support RTX but nowhere near as fast as Turing with RT cores

 

Edit 2: Or maybe RTX Titan would be Turing with RTX2080/2070 being Volta? I think that could be very likely

 

Edit the third: Ok so I change my mind again, I think Turing will be titan, 2080 and 2070 based on very nice price of RTX5000

[jasonvp]

45 minutes ago, Swatson said:

Edit 2: Or maybe RTX Titan would be Turing with RTX2080/2070 being Volta? I think that could be very likely

Have they done a Volta spin yet that uses GDDR?  As far as I recall, all Volta chips are set up for HBM2, which NVidia certainly will not ship on consumer cards.  So unless they've redone Volta to use GDDR, then I tend to doubt it.  My gut tells me Volta's days are numbered.  But I guess we'll find out in a week.

 

[S w a t s o n]

Wait, RTX5000 is only $2300.... so if Titan RTX is that chip it's gonna be cheaper than Titan V, less than $2300 for sure. What if RTX Titan could be RTX6000, which looks to be full raytracing capability Turing?

I actually have no idea at this point but if nvidia can sell cut down Turing as RTX5000 QUADRO (expensive) for only $2300 we might see a full Turing release after all?

[S w a t s o n]

Just now, jasonvp said:

Have they done a Volta spin yet that uses GDDR?  As far as I recall, all Volta chips are set up for HBM2, which NVidia certainly will not ship on consumer cards.  So unless they've redone Volta to use GDDR, then I tend to doubt it.  My gut tells me Volta's days are numbered.  But I guess we'll find out in a week.

 

We wont know for a week, they wouldn't announce a non-HBM volta respin for geforce until it was ready for release

[jasonvp]

1 minute ago, Swatson said:

Wait, RTX5000 is only $2300.... so if Titan RTX is that chip it's gonna be cheaper than Titan V, less than $2300 for sure.

I still firmly believe that the next Titan will be a $1200-$1500 card and will be aimed squarely at the gotta-have-it gamers with the appropriate level of disposable income.

 

eg: me.  :-P

[S w a t s o n]

4 minutes ago, jasonvp said:

I still firmly believe that the next Titan will be a $1200-$1500 card and will be aimed squarely at the gotta-have-it gamers with the appropriate level of disposable income.

 

eg: me.  :-P

Sure but it's not gonna be slower than Titan V To be fair our previous discussion assumed Turing was cut down Volta and it's clearly not.