NVIDIA GTC November 2021 wrapped up last week, with a focus on AI software and solutions and a number of hardware announcements. NVIDIA again demonstrated at GTC the range of innovation they bring to the AI ecosystem.
As NVIDIA’s only Elite Partner across all NVIDIA technologies (GPU compute, data centre solutions and networking), XENON is looking forward to brining these new solutions to our Australian and New Zealand customers. Here is a short summary of the key announcements from GTC.
NVIDIA A2 Tensor Core GPU
This low profile PCIe form factor GPU delivers computational power with lower wattage (40-60W) suitable for any server. The A2 is an entry level GPU in the Ampere Architecture family, delivering solid inference acceleration for modern machine learning and deep learning models. Compared to CPU only inference, the A2 is up to 20x faster across a range of common AI applications including Computer Vision, Natural Language Processing, and Text-to-Speech. The smaller form factor and low power draw makes the A2 ideal for AI at the Edge applications such as processing camera data or IoT sensor data. More details at NVIDIA website here.
The A2 Tensor Core GPU will be available from late November onwards. Contact XENON to express interest and pre-order yours now.
Jetson AGX Orin
This new Jetson module packs up to 200 TOPs in a small form factor resulting in 6x performance gains over the previous generation Xavier. Ideal for advanced robotics and AI edge applications in manufacturing, retail, logistics, agriculture, smart cities, health care. There is no limit to the possible solutions with the Jetson family of AI edge devices which come with pre-trained models applicable to a range of these applications. Previous generation of Jetson developer kits have been extremely popular across industry and academia and XENON expects these new AGX Orin units will be very much the go to development kit for 2022.
The new Jetson AGX Orin Developer Kit will be available in the first quarter of 2022. Contact XENON to express interest and pre-order yours now to secure the first shipment allocation.
NVIDIA Quantum-2 – InfiniBand, 400Gb/s
NVIDIA announced the new generation InfiniBand with 400Gb/s across the new NVIDIA Quantum-2 switch, the ConnectX-7 network adaptor, and the new BlueField-3 data processing unit (DPU). Combined with updated software for the new Quantum-2 architecture, this is a step-change improvement for data centres, HPC workloads and shared tenancy private and public cloud environments.
The new Quantum-2 platform advances the in-network computing capability of all components with preconfigured and programmable compute engines across the Quantum-2 Switch and the BlueField 3 DPUs. Updated Transceivers and Cables round out the end-to-end 400Gb/s InfiniBand solution.
Read the full datasheet summary of the architecture here.
AI Software and the Omniverse Replicator
NVIDIA software and support for the overall vision of AI and the ecosystem is why NVIDIA solutions are the go-to for AI solutions. GTC 21 last week included some major announcements in this area.
Omniverse Replicator
NVIDIA announced the Omniverse Replicator with some impressive demonstration video of autonomous vehicles and robotics in a logistics environment. The Omniverse Replicator is a simulation framework that produces physically accurate synthetic data in the Omniverse to help train AI models in autonomous vehicles and robotics. These two models – DRIVE Sim for autonomous vehicles and Isaac Sim for training robots will be available soon – likely Q1 2022.
The Omniverse platform continues to evolve at a rapid rate with new connectors and SDKs. Full details at the NVIDIA Developer site here.
New and Up-dated CUDA Libraries
NVIDIA announced 65 new software development kits for accelerating quantum computing and quantum simulations, as well as updates to over 150 CUDA libraries. Many of these new libraries are available for specific industry applications, such as logistics and warehousing. Updates to RAPIDS for data science provides new functions and accelerations to existing algorithms.
There are also new training courses available at the NVIDIA Deep Learning Institute for developers to learn more about the tools and SKDs available.
NVIDIA Solutions from XENON
XENON is NVIDIA’s only partner in Australia with Elite Partner status across all NVIDIA technologies. This represents deep knowledge and expertise across GPU compute, data centre solutions, and networking. Contact XENON today to discuss your requirements.
Talk to a Solutions Architect
