Benchmarks comparing inference performance of the NVIDIA GH200 Grace Hopper Superchip, enhanced by ZeRO-Inference, to NVIDIA H100 and A100 Tensor Core GPUs.
Persistent storage is now available in all Lambda Cloud regions and for all on-demand instance types, including our NVIDIA H100 Tensor Core GPU instances.
Benchmarks on NVIDIA’s Transformer Engine, which boosts FP8 performance by an impressive 60% on GPT3-style model testing on NVIDIA H100 Tensor Core GPUs.
GPU benchmarks on Lambda’s offering of the NVIDIA H100 SXM5 vs the NVIDIA A100 SXM4 using DeepChat’s 3-step training example.
Lambda has launched a new Hyperplane server combining the fastest GPU on the market, NVIDIA H100, with the world’s best data center CPU, AMD EPYC 9004.
How to use FlashAttention-2 on Lambda Cloud, including H100 vs A100 benchmark results for training GPT-3-style models using the new model.
On-demand HGX H100 systems with 8x NVIDIA H100 SXM instances are now available on Lambda Cloud for only $2.59/hr/GPU.
How to build the GPU infrastructure needed to pretrain LLM and Generative AI models from scratch (e.g. GPT-4, LaMDA, LLaMA, BLOOM).
Lambda Cloud has deployed a fleet of NVIDIA H100 Tensor Core GPUs, making it one of the FIRST to market with general-availability, on-demand H100 GPUs. The high-performance GPUs enable faster training times, better model accuracy, and increased productivity.
In early April, NVIDIA H100 Tensor Core GPUs, the fastest GPU type on the market, will be added to Lambda Cloud. NVIDIA H100 80GB PCIe Gen5 instances will go live first, with SXM to follow very shortly after.
Native support for FP8 data types is here with the release of the NVIDIA H100 Tensor Core GPU. These new FP8 types can speed up training and inference.
This article discusses the performance and scalability of H100 GPUs and the whys for upgrading your ML infrastructure with the H100 release from NVIDIA.
...