NVIDIA’s GTC events are always inspiring and important platforms for NVIDIA to announce new products and solutions. GTC 2021 has been no different. Kicking off on Monday 12 April, the even runs over the course of the week.
NVIDIA CEO Jensen Huang’s keynote included visually stunning examples of the NVIDIA technology at work, and key announcements of relevance to XENON customers. Here are the updates so far!
NVIDIA CPUs – Grace
NVIDA is developing a data centre focused CPU, code named Grace. This CPU is designed to be tightly integrated with NVIDIA’s data centre GPU’s, and NVIDIA is claiming a Grace based system will deliver 10x faster performance than the current state of the art DGX A100 systems (which run on x86 CPUs). The Grace CPU line is expected to be widely available in 2023. Watch this space for up-dates as they progress with this interesting innovation which brings the third horse to the NVIDIA processing troika of GPU, DPU and now CPU. Jensen announced that there would be a two year release cycle, with an alternating focus on ARM and x86 platforms which should deliver gains across HPC and AI with integrated CPU/GPU development.
NVIDIA DPUs – BlueField 3
NVIDIA also announced advanced acceleration of data centre HPC and AI workloads with NVIDIA BlueField 3 DPU (Data Processing Unit). The BlueField 3 DPU delivers the equivalent of 300 CPU cores on the edge at the network interface. The BlueField 3 also includes enhanced security, delivering zero-trust processing to the data centre environment. This is the first BlueField DPU built expressly for AI and accelerated computing, it is optimised for multi-tenant, private or public cloud environments with enhanced cyber security and the ability to drive networking, storage and processing management at the edge with software. This is also the industry’s first 400GbE/NDR DPU, providing advanced acceleration that will integrate nicely with the 400Gb/s capability of the DGX A100.
DGX Station 320G
Last year’s new DGX Station A100 was released with the new A100 GPUs at it’s core. The newly announced DGX Station A100 320G delivers the up-graded 80GB A100 GPUs (with 4, resulting in 320GB available GPU memory). The additional memory sees a 3x increase in speed for complex AI workloads such as natural speech processing. With the A100 Multi-Instance GPU capability, the DGX Station A100 320G can be shared between as many as 28 data sciences simultaneously.
New A10, A30 GPU PCIe Cards
NVIDIA also announced new additions to their GPU line – the A10 and A30. All build from the new Ampere architecture, the A10 is more graphics focused while the A30 is more compute focused.
- The A10 has 2nd generation RT Cores and 3rd generation Tensor Cores with 24GB memory. Highly power efficient 1 slot full height/full length form factor it only draws 150W and can be deployed in virtual workstations or desktops. Ideal for high performance graphics, video and AI work this powerful and efficient GPU will be a popular addition to the line up and will be ideal supporting virtual desktops and video conferencing.
- The A30 is built for versatile compute and mainstream enterprise applications. Featuring Multi-instance GPU (MIG) capabilities, the individual A30 can support 4 instances @ 6GB each or 2 instances at 12GB each. The A30 is a 2 slot full height/full length GPU that draws 165W, packing an efficient level of compute and agility in a compact form. The A30 is ideal for natural language processing, conversational AI and recommender systems.
These two additions to the GPU line up give customers a wider range of options, with 2 units now MIG capable (the A100 and A30), and a full range of graphics focused GPUs. Contact XENON to design the right GPU acceleration platform for your workloads.
NVIDIA AI Software
NVIDIA continues to invest heavily in the AI software stack, and there were a raft of announcements related to their AI software. These included NVIDIA Megatron – a framework for natural language processing which can generate document summaries, complete phrases in email, grade quizzes, comment on live sport and even generate computer code.
Also there were new models for NVIDIA Clara Discovery – the acceleration libraries for computational drug discovery and life science.
NVIDIA released details of their new simulators for quantum computing, allowing researchers to simulate quantum circuits and design quantum computers.
Other software releases included NVIDIA Morpheus for data centre security; NVIDIA Jarvis for speech recognition, language understandings and translations; NVIDIA Merlin for recommender systems; NVIDIA Tao to fine-tune and adapt NVIDIA pre-trained models while protecting customer data privacy; and finally, NVIDIA Triton which is an inference server which can harvest insights from continuous streams of data.
The NVIDIA Omniverse is an interesting application of virtual, real-time 3D worlds for advanced simulations. The Omniverse can be inhabited by people, AI and robots. This unique collaboration allows for advanced simulations of environments and it will be interesting to see how this is applied in the future in a variety of contexts – from training to advanced design simulations. The keynote included a fascinating walk through of a virtual BMW car factory built with Omniverse, allowing for production engineers to test factory designs, efficiency and safety in a virtual world.
NVIDIA has been at the forefront of the development of smart car systems and autonomous vehicles which present one of the most intense machine learning and robotics challenge. NVIDIA is building an end-to-end modular platform so partners have a robust tool set to build autonomous vehicles. Volvo has been developing automotive systems using NVIDIA Drive since 2016, and at GTC they announced that they were implementing NVIDIA Drive for their up-dated XC90 SUV which will be available in 2025 which will advance their their software defined automotive vision and deepens the collaboration between NVIDIA and Volvo.
If you are interested in how these NVIDIA innovations can benefit your organisation, please feel free to reach out to XENON and we can do deep dive with you on specific areas of interest.Talk to a Solutions Architect
Watch the Keynote