Nvidia (NVDA) CEO Jensen Huang made quite a few hardware and software announcements during a 160-minute keynote talk. He also disclosed some new partnerships, and shared a handful of interesting stats.
Here’s a run-down of highlights from Huang’s Monday keynote address at Nvidia’s GPU Technology Conference.
New Workstations and Servers for Data Scientists
Huang unveiled a line of workstations — to be sold by OEM partners such as Dell, HP and Lenovo — that feature two GPUs from Nvidia’s Quadro RTX workstation GPU line, and support up to 96GB of memory. The systems come pre-installed with Nvidia’s data science software stack.
Also shown off: The Data Science Server, which contains four of Nvidia’s Tesla T4 GPUs and supports up to 64GB of GDDR6 graphics memory. The system is a less powerful complement to Nvidia’s DGX-2 server, which packs up to 16 of its powerful Tesla V100 GPUs and is meant for AI and high-performance computing (HPC) workloads.
New Server Options for Graphics Rendering and Cloud Gaming
Last August, Nvidia showed off the RTX Server, a reference architecture for servers using its just-unveiled Quadro RTX GPUs. The company is now updating its RTX Server lineup to include servers packing up to 40 GPUs. In addition, the servers can be combined into a “pod” featuring up to 32 server nodes. OEM partners are selling RTX servers featuring relatively small 2U and 4U form factors; Nvidia will offer an RTX server featuring a larger 8U chassis in Q3.
Nvidia also showed off some impressive demos of Quadro RTX GPUs powering real-time ray tracing (a graphics rendering technique that can enable photorealistic imagery) for 3D content creation workloads. And it rolled out Omniverse, a collaboration software platform for graphics content creators. Huang called it “the Google Docs for 3D design.”
A New Cloud Computing Offering from Amazon
Nvidia and Amazon Web Services (AWS) announced the launch of a new AWS cloud computing instance (known as the EC2 G4) for powering inference workloads — the running of trained AI models against new data and content, such as a voice command, a language-translation request or a photo taken with a smartphone camera — using Nvidia’s Tesla T4 GPUs. The EC2 G4 arrives a few months after Amazon unveiled Inferentia, an internally-designed chip for powering AWS inference workloads.
Separately, AWS exec Matt Garman shared a slide highlighting major AWS clients who are using Nvidia’s powerful Tesla V100 server GPU, which is meant for AI training and high-performance computing (HPC) workloads. The client list included Salesforce.com, Verizon, Siemens, Comcast, Lyft and Western Digital.
Autonomous Driving Moves
Huang disclosed Nvidia is expanding its autonomous driving partnership with Toyota, which was first announced in 2017. Toyota will the first customer for Nvidia’s Drive Constellation system, which simulated autonomous driving activity within data centers. The auto giant is also planning to use Nvidia’s Drive autonomous driving platform, and to deploy its server GPUs for AI workloads.
Also revealed: A slew of new software updates for Drive. These include the introduction of what Nvidia calls “high-function Level 2+ autopilot” capabilities that include real-time mapping and surround perception for automatic lane changes, as well as software that attempts to predict the paths that surrounding objects will take.
A $99 Embedded Computer
The Jetson Nano is the latest addition to Nvidia’s Jetson family of embedded computers for “edge” devices performing AI inference (drones, robots, etc.). It consumes just 5 watts of power, contains a GPU based on Nvidia’s older Maxwell architecture and isn’t much larger than a credit card.
Impressive Stats About Developer Support
Huang stated the developer base for Nvidia’s CUDA GPU programming model has grown by 50% in the last year to 1.2M, and that downloads of apps relying on Nvidia’s CUDA programming models rose by 60% last year to 13 million.
Also disclosed: Downloads for Nvidia’s TensorRT inference software library rose by a factor of 6 last year to 300,000, and Nvidia now has over 300,000 gamers using its GeForce Now cloud gaming service. Another million gamers are said to be on the waiting list for GeForce Now.
What Wasn’t Announced
Not surprisingly, given how many gaming GPU launches it has carried out over the last six months, Huang didn’t announce any new gaming GPUs during his keynote.
Huang also didn’t announce a successor to the Tesla V100; the V100 has been widely adopted for use in AI training systems and supercomputers, but is now almost two years old and is starting to see new rivals emerge. Likewise, a successor to Nvidia’s V100-powered DGX-2 server, which launched last year, wasn’t announced.