GTC 2014 | Nvidia announces Pascal, Titan Z, IRAY VCA, Erista, Jetson TK1 in the keynote

Our slideshow highlights the opening GTC 2014 keynote from Nvidia from the gargantuan $3000 Titan Z to the small Tegra K1 successor Erista and everything in between.

Jumping straight from GDC 2014 in San Francisco just last week, we land today further south of the state in San Jose just in time for GTC 2014 to see the opening keynote presentation from none other than NVIDIA CEO Jen-Hsun Huang.

Among the highlights of the speech are bandwidth bottlenecks, 3D stacked RAM, Pascal, Machine Learning, IRAY VCA for photorealistic and near real-time rendering, GRID and the next generation of Tegra codenamed Erista.

The next generation of GPUs to follow Maxwell is now Pascal, which will include NVLink to offer 5x to 12x PCIe speeds between CPU and GPU to combat bandwidth bottlenecks. The stacked memory will also expand the memory bus of the GPU by orders of magnitude. Pascal chips will be designed to fit in modules one third the size of PCIe cards for gaming, cloud and server use. Multichip configurations of Pascal should also exceed previous generations in performance thanks to NVLink.

Meanwhile, the new $3000 Titan Z will succeed the Titan Black as king of the desktop GPU. The new GPU utilizes two Titan cores for nearly twice the performance with 5760 CUDA cores, 12 GB VRAM and 8 teraflops of compute. NVIDIA uses the extremely demanding Machine Learning concept and IRAY VCA rendering application as perfect examples for showcasing the raw power of the Titan Z. 

Finally, the Tegra K1 flexes its mobile muscle and ups its class by making an appearance with the Audi A7. The NVIDIA chip is able to replace a much larger computer on the car for autonomous driving. The K1 successor, codenamed Erista, is touched upon as well and is expected to launch by 2015 with an architecture heavily based on Maxwell.

With so many surprises, it seemed a bit odd to end the flood of new hardware by revealing Portal for their handheld Shield gaming platform. Proudly announcing a 6-year old PC title for the aging Shield device should have probably been left for the beginning of the keynote.

Nonetheless, check out our highlights of the full keynote below. If the hardware and software promises hold true, then this will certainly be a fruitful year for NVIDIA.

You can start the image gallery by clicking on the first image below.

The GPU Technology Conference (GTC) has been growing bigger every year according to NVIDIA
The GPU Technology Conference (GTC) has been growing bigger every year according to NVIDIA
Naturally, CUDA is one of the first topics of discussion
Naturally, CUDA is one of the first topics of discussion
Bandwidth between core components like CPU and GPU continue to be a problem for professional CUDA programmers
Bandwidth between core components like CPU and GPU continue to be a problem for professional CUDA programmers
NVLink is NVIDIA's solution to this problem for the next generation of GPUs
NVLink is NVIDIA's solution to this problem for the next generation of GPUs
With NVLink, GPUs and CPUs can be linked together directly for faster communications
With NVLink, GPUs and CPUs can be linked together directly for faster communications
3D stacked RAM can widen the memory bus to over 1000 bits according to Huang
3D stacked RAM can widen the memory bus to over 1000 bits according to Huang
Previously known as Volta, the new architecture has been renamed after Blaise Pascal
Previously known as Volta, the new architecture has been renamed after Blaise Pascal
Initial Pascal cards for professional use will include NVLink and be as small as two credit cards
Initial Pascal cards for professional use will include NVLink and be as small as two credit cards
Pascal will be suitable for both gaming and high-end workstations. Hopefully for high-end notebooks, too!
Pascal will be suitable for both  gaming and high-end workstations. Hopefully for high-end notebooks, too!
Another feature of Pascal is "unified memory" as shown on the roadmap
Another feature of Pascal is "unified memory" as shown on the roadmap
"Machine Learning" and AI is another vaunted capability
"Machine Learning" and AI is another vaunted capability
As an example, NVIDIA discusses how neurons can be key to Machine Learning
As an example, NVIDIA discusses how neurons can be key to Machine Learning
The "neural net" can be utilized for edge detection and feature detection for faces or other objects
The "neural net" can be utilized for edge detection and feature detection for faces or other objects
However, this method has been known to be computationally demanding. Google uses hundreds of servers to study the neural net
However, this method has been known to be computationally demanding. Google uses hundreds of servers to study the neural net
Simulating the neural net of a human brain would currently require over 40000 years of computation from hundreds of servers
Simulating the neural net of a human brain would currently require over 40000 years of computation from hundreds of servers
NVIDIA claims that by using their GPUs and CUDA approach, only 3 servers would be needed to output the same amount of work as 1000 of these Google servers.
NVIDIA claims that by using their GPUs and CUDA approach, only 3 servers would be needed to output the same amount of work as 1000 of these Google servers.
As a result, smaller companies and academic researchers can afford NVIDIA hardware
As a result, smaller companies and academic researchers can afford NVIDIA hardware
Moving on from Pascal, NVIDIA reveals the Titan Z GPU to succeed the Titan Black due to the success of the original Titan. The Titan Z carries dual Titan cores.
Moving on from Pascal, NVIDIA reveals the Titan Z GPU to succeed the Titan Black due to the success of the original Titan. The Titan Z carries dual Titan cores.
Three of these jaw-dropping GPUs can replace 1000 standard servers used for Machine Learning
Three of these jaw-dropping GPUs can replace 1000 standard servers used for Machine Learning
As usual, a complex water simulation demo is shown to boast the power of NVIDIA's latest GPU
As usual, a complex water simulation demo is shown to boast the power of NVIDIA's latest GPU
More complex physics rendered in real-time
More complex physics rendered in real-time
Rendering photo realistic images is also possible
Rendering photo realistic images is also possible
NVIDIA announces a scalable rendering server called IRAY VCA. The VCA is the world's first scalable rendering application
NVIDIA announces a scalable rendering server called IRAY VCA. The VCA is the world's first scalable rendering application
The IRAY renderer supports Catia, 3DS Max, Bunkspeed, and Maya. The car in this image was rendered with 19 VCAs
The IRAY renderer supports Catia, 3DS Max, Bunkspeed, and Maya. The car in this image was rendered with 19 VCAs
The one Petaflop of performance available in this setup allows the rendering of each image in just a few seconds
The one Petaflop of performance available in this setup allows the rendering of each image in just a few seconds
Thus, viewing angles can be changed in near real-time
Thus, viewing angles can be changed in near real-time
A single VCA block offers 60x the performance of a Quadro K5000 workstation
A single VCA block offers 60x the performance of a Quadro K5000 workstation
With a price of about $50,000, it is a "cheap" solution for this amount of performance
With a price of about $50,000, it is a "cheap" solution for this amount of performance
Next, NVIDIA unveils GRID
Next, NVIDIA unveils GRID
VMWare will utilize GRID in their 2015 Horizon DaaS platform
VMWare will utilize GRID in their 2015 Horizon DaaS platform
Tegra K1 is also mentioned briefly
Tegra K1 is also mentioned briefly
As CUDA can now run on the Tegra K1 GPU, developers can use their code on supercomputers down to mobiles
As CUDA can now run on the Tegra K1 GPU, developers can use their code on supercomputers down to mobiles
The Jetson TK1 is an embedded development kit using Tegra K1. It is available now for $192
The Jetson TK1 is an embedded development kit using Tegra K1. It is available now for $192
Thanks to Visionworks, developers can easily develop computer vision applications on Tegra K1 (Jetson TK1)
Thanks to Visionworks, developers can easily develop computer vision applications on Tegra K1 (Jetson TK1)
Whereas the Tegra 4 was codenamed Logan, the Tegra K1 successor is naturally named after Logan's son, Erista. Erista is currently planed for a 2015 launch
Whereas the Tegra 4 was codenamed Logan, the Tegra K1 successor is naturally named after Logan's son, Erista. Erista is currently planed for a 2015 launch
The high performance of Tegra K1 and Erista can potentially be used for autonomous driving
The high performance of Tegra K1 and Erista can potentially be used for autonomous driving
As a demonstration, an Audi A7 appeared onto the stage without a driver in sight
As a demonstration, an Audi A7 appeared onto the stage without a driver in sight
The Tegra K1 is able to replace a large computer that was originally in the rear of the car
The Tegra K1 is able to replace a large computer that was originally in the rear of the car
In fact, the NVIDIA module is small enough to fit in a box hidden in the back of the car
In fact, the NVIDIA module is small enough to fit in a box hidden in the back of the car
Finally, CEO Jen-Shun Huang closes the keynote and announces a port of Valve's Portal for Nvidia's Shield
Finally, CEO Jen-Shun Huang closes the keynote and announces a port of Valve's Portal for Nvidia's Shield
 
 
static version load dynamic

Comment this article:

> Notebook / Laptop Reviews and News > News > News Archive > Newsarchive 2014 03 > Nvidia announces Pascal, Titan Z, IRAY VCA, Erista, Jetson TK1 in the keynote
Author: Klaus Hinum, 2014-03-25 (Update: 2014-03-26)