NVIDIA closed their GTC Fall 2022 conference last week, and there were a number of significant announcements for HPC / AI / DL. We’ll cover the highlights here, and go into more depth on later posts.
New GPU Architecture
NVIDIA announced a new GPU architecture, to be known as, and named after Ada Lovelace, the first computer programmer. This new architecture is focused on graphics workloads, for creators, gamers, and to dive the Omniverse. Ada’s advancements include a new Streaming Multiprocessor, a new RT Core with twice the ray-triangle intersection throughput, and a new Tensor Core with the Hopper FP8 Transformer Engine and 1.4 petaflops of Tensor processor power. Key to graphic and gaming performance, the new Ada architecture implements NVIDIA DLSS, for AI enhanced frame generation – resulting in faster, more accurate frame rendering.
The new architecture will be available in new RTX GPUs and a new L40 data centre GPU.
New GeForce RTX 40 Series GPUs
There are three new GeForce RTX GPUs that will be available in the coming months. The GeForce RTX 4090 will be shipping in mid-October, with the Ada Lovelace architecture, 24GB GPU memory delivering 2x to 4x better performance on a range of graphics tasks.
There are two versions of the GeForce RTX 4080 which have a planned shipping date of November. The RTX 4080 comes with 16GB. Both new RTX cards are PCIe Gen4, and designed for gaming and creative workflows – animation, special effects and game design. All will be available from XENON in our Nitro GPU workstations.
Contact XENON to scope the right card for your requirements and pre-order yours today.
L40 Data Centre GPU
The new L40 also uses the Ada Lovelace architecture to create a balanced GPU that can be utilised for a range of requirements – from AI to data analytics and graphics. While the L40 is designed to be the “best universal GPU” it will also be the best in class for graphics workloads. With a TPD of 300W and passive cooling, these L40 cards will suit GPU servers with adequate cooling and airflow, and XENON can help you design or upgrade your GPU servers to take advantage of these new cards.
The L40 is also the basis of second generation OVX systems – which will pack eight L40’s into a GPU server purpose built for running digital twins and simulations in the NVIDIA Omniverse. These systems will be available in early 2023. Current plans have the L40 shipping in December as individual cards.
H100
The H100 GPU was announced at GTC Spring 2022 back in April 2022. These units are now in full production and will be available as H100 PCIe Cards, as well as in the DGX H100 systems and will be shipping next month!
The new architecture in the H100 is delivering up to 7x higher performance for HPC applications and advanced processes like genomic sequencing. The H100 PCIe is a dual slot, air cooled unit, with 80GB of GPU memory that streams at 2TB/s. The H100’s can be configured for TPD of between 300W and 350W. Second generation NVIDIA Multi-Instance GPU (MIG) is available with 10GB per instance, and up to 7 MIGs per GPU.
In the DGX H100 systems, the GPUs take full advantage of integrated NVLink across all GPUs, and allows all 56 MIG instances to be aggregated as a single GPU, or split out individually as required for data analytics, training and inference workloads.
Jetson Orin Nano
The Nano form factor is back in the Jetson family, with the new Jetson Orin Nano featuring an NVIDIA Ampere architecture GPU, Arm-based CPUs, next-generation deep learning and vision accelerators, high-speed interfaces, fast memory bandwidth and multimodal sensor support. Packing 25 trillion operations per second (TOPS) into the Jetson Orin Nano, empowers more customers to commercialize products that once seemed impossible, from engineers deploying edge AI applications to Robotics Operating System (ROS) developers building next-generation intelligent machines. And we expect to see this land in Australia in a sub $500 AUD price point when it arrives in early 2023.
XENON NPN Elite Partner
XENON remains NVIDIA’s only Elite partner for their entire stack, including NVIDIA Networking and GPU computing. XENON is also DGX-Ready Managed Service partner, providing managed services for DGX systems, networking and storage.
Contact XENON today to discussion your NVIDIA Solution requirements.
NOTE: Update 20-October-2022
NVIDIA had originally announced at GTC that there would be a RTX 4080 with 12GB. They have since “unlaunched” that model, and have not announced a future date, name of config for that. XENON believes this is a good choice, and having two cards named 4080, with different memory capacity, core counts and capabilities only confuses the community. We look forward to bringing you the new updated RTX 4090 and RTX 4080 16GB cards in our systems next month.
