NVIDIA CEO Jen-Hsun Huang kicked off the company's annual GPU
Technology Conference in San Jose this morning, unveilling the
company's new Tegra roadmap and graphics chips.
Nvidia's Maxwell graphics chip is due in 2014 and will use virtual
memory shared by the GPU and host CPU. Maxwell is coming with
unified virtual memory, which makes it possible for GPU operations
to see the CPU memory and vice versa, so programability is easier.
The approach, likely implemented in Nvidia's proprietary Cuda
environment, is similar to what AMD is enabling with its HSA group
supported by ARM, Microsoft and others.
After that is Volta, which is even more energy efficient, which has
a new technology called stacked DRAM. Available in 2015, Volta
graphics processor will be the first to use stacked memory with
through-silicon vias, similar to the Hyper Memory Cube defined by
Micron.
Nvidia says Volta will solve one of the biggest challenges facing
GPUs today, which is access to memory bandwidth. With Volta, Nvidia
willl prevent from getting off its chip onto a PC board - it will
have DRAM on same silica sub-strate, which will carry a whole bunch
of DRAMs stacked atop each other. Nvidia will cut a hole through
the silicon and connect each layer. This is expected to achieve
one terabyte per second of bandwidth.
Regarding mobile processors, Nvidia will also start offering samples of the Tegra5, aka Logan,
this year. This will be the company's first mobile processor to
feature a Kepler GPU, it support its Cuda environment for
general-purpose graphics. It also supports OpenGL 4.3 and will be
in production in 2014.
In 2015, Nvidia will also pack a Maxwell core along with its first
64-bit ARM design into a sixth-generation Tegra chip called Parker.
The device is the first of Nvidia's Project Denver designs and the
first 64 bit ARM processor coupled with Nvidia's next-gen GPU
Maxwell. It will be the first to use FinFET transistors,
said Jen-Hsun Huang.
Nvidia's designers also built a powerful ARM computer based on a
super-low power GPU combined with ARM. The new Kayla platform pairs
a powerful Tegra mobile processor with an NVIDIA Kepler-based GPU
to support a suite of technologies that have never appeared on
mobile devices before - including CUDA 5 and Open GL 4.3.
Kayla offers a sneak-peak at the capabilities that will be
unleashed by Logan. "What's amazing is that Logan will be the size
of a dime, whereas Kayla is now the size of a tablet PC" Nvidia CEO
Jen-Hsun Huang said.
Support for Open GL 4.3 will unleash advanced desktop-class
graphics capabilities - such as tessellation and compute shaders -
that are the building blocks for cutting-edge PC games.
And CUDA 5, a sophisticated parallel computing framework, will
unlock the ability to tackle complex computer vision problems, such
as face and object recognition.
Also at GTC, Nvidia is rolling out its Grid Enterprise initiative.
It consists of a new layer of systems and software to provide
business users virtualized access to apps running on Nvidia
graphics processors.
Dell, HP and IBM are offering NVIDIA GRID-based servers. Citrix,
Microsoft and VMware are offering NVIDIA GRID-enabled software.
At the heart of NVIDIA GRID enterprise solutions are NVIDIA GRID
VGX software -- a complete stack of GPU virtualization, remoting
and management libraries -- and NVIDIA GRID K1 and K2 boards, which
are purpose-built for demanding server environments.
NVIDIA VGX software unlocks the virtualization and remoting
capabilities of NVIDIA GRID GPUs and is licensed by Citrix for use
in XenDesktop, XenApp and XenServer; VMware for use in vSphere and
Horizon View; and Microsoft for use in RemoteFX.
NVIDIA GRID K1 boards, which include four NVIDIA Kepler
architecture-based GPUs and 16GB of memory, are designed to host
the maximum number of concurrent users. NVIDIA GRID K2 boards,
which include two higher end Kepler GPUs and 8GB of memory, are
designed to serve the specific needs of users of graphics-intensive
applications.
Dell, HP and IBM announced servers with the NVIDIA GRID K1 and K2
boards, including the Dell PowerEdge R720; the HP ProLiant WS460c
Gen8 and HP ProLiant SL250 Gen8 and the IBM iDataPlex dx360 M4.
NVIDIA has launched a certification program and testing center
where leading OEMs and ISV partners can validate and certify
solutions. To accelerate deployment-ready validated designs, NVIDIA
and Citrix are working together to establish an NVIDIA and Citrix
Ready verification program offering fast track validation to OEMs
and ISVs on NVIDIA GRID K1 and K2 cards and Citrix XenDesktop
solutions.
NVIDIA also introduced the first visual computing
appliance -- enabling businesses to deliver ultra-fast GPU
performance to any Windows, Linux or Mac client on their network.
The NVIDIA GRID Visual Computing Appliance (VCA) is a GPU-based
system that runs complex applications such as those from Adobe
Systems Incorporated, Autodesk and Dassault Systemes, and sends
their graphics output over the network to be displayed on a client
computer.
NVIDIA GRID VCA provides flexibility to small and medium-size
businesses with limited IT infrastructures. Their employees can,
through the simple click of an icon, create a virtual machine
called a workspace. These workspaces -- which are, effectively,
dedicated, high-performance GPU-based systems -- can be added,
deleted or reallocated as needed.
NVIDIA GRID VCA is a 4U appliance. Its 16 NVIDIA GPUs and NVIDIA
GRID VGX software provide NVIDIA Quadro-class graphics performance
for up to 16 concurrent users, with low latency, high resolution
and maximum interactivity for unparalleled quality of service.
Available in the United States in May, NVIDIA GRID VCA is offered
in 8 GPU or 16 GPU configurations, with pricing starting at
$24,900, plus an annual software license of $2,400.
Jen Hsun Huang also said he "hopes" the Project Shield portable handheld gaming device will be available through retail outlets in the latter part of the second quarter.
The Shield combines an Android-based computer running on a Tegra 4 processor with a 5-inch display. Nvidia expects the handheld will attract developers to produce better games for Android, some of which will appear in Nvidia's Tegra Zone gaming app store.
Jen-Hsun Huang also commented on Microsoft's Windows RT software, saying that devices using the OS aren't selling very well.
"Windows RT is disappointing to us because we expected to have sold more than we did," Huang said. "Everybody expected to have sold more than we did."