Get Into The ARMs Race: Future-Proof Your Workloads Now With Lambda
Get Into The ARMs Race: Future-Proof Your Workloads Now With Lambda
December 19, 2024• 3 min read
Blackwell is coming… so is ARM computing
2025 is just around the corner, and with it comes the highly anticipated launch of NVIDIA's revolutionary Blackwell platform. Blackwell promises to “usher in a new era of computing”, pushing the boundaries for building and running real-time generative AI up to trillion-parameter large language models.
One of the most exciting Blackwell elements is the NVIDIA GB200 Blackwell Superchip, which connects two NVIDIA B200 Tensor Core GPUs to the NVIDIA Grace CPU over a 900GB/s ultra-low-power NVLink chip-to-chip interconnect.
This ARM-based architecture unlocks new levels of performance and efficiency, but it also introduces a potential hurdle for developers: existing workflows, tools, and libraries designed for x86 processors require testing and potentially recompilation to function seamlessly on Grace Blackwell.
Don’t be the last to know
Waiting to find-out how your current and future workloads would fare with GB200 isn’t without risks:
“The company's CFO, Colette Kress, described Blackwell demand as 'staggering' and said Nvidia is racing to scale supply to meet customer demand.”
- New generation GPUs will deploy first to pre-orders, captured through waitlists. You can join Lambda’s waitlist for Blackwell here. Lambda will announce exciting enterprise-level deals in early 2025 :-)
- It could be months before a flagship GPU like NVIDIA GB200 is broadly available for short-term testing, as pressing deployments for production will take priority.
- Testing your existing or planned workflow on an ARM-based platform in advance could save you precious time (and costs!) once GB200 is available.
Get into the ARMs race: $1.49 per hour on GH200, and Consulting
Lambda is committed to smoothing your transition to the future. We're excited to announce that you can start testing your applications on the current generation NVIDIA GH200 Superchip, today. This chip offers the same powerful combination of ARM64 CPU and Tensor Core GPU (one H100, in GH200’s case) as the Grace Blackwell platform.
Here's why you should take advantage of this opportunity:
- Get a head start: Ensure your workflows, tools, and libraries are compatible with Grace Blackwell before its official launch and downstream availability.
- Consult with us: Our Machine Learning Research team has been actively exploring the capabilities of GH200. We're happy to share what we've learned to support your work. You can reach out for Consulting here.
- Exceptional price point: For a limited time, access the NVIDIA GH200 Superchip on Lambda for only $1.49 per hour.
- Get H100 power at a fraction of the cost: Even if Grace Blackwell isn’t your primary focus, this offer allows you to tap into the power of the NVIDIA H100 Tensor Core GPU at an exceptional price. It delivers 3x the performance of the A100, at a very similar cost (A100 SXM 40GB is $1.29/hour on Lambda’s Cloud).
This incredible offer is valid until the end of March 2025. Don't miss this chance to prepare for the future of AI and/or secure the best H100 pricing in the market!
Ready to start using GH200 on-demand with Lambda? Create an account or login.
Want to discuss your Blackwell needs? Contact our team.
Resources
Learn more about GH200: blog
Run a PyTorch-based benchmark on GH200: documentation
Run HuggingFace Transformers and Diffusers on GH200: documentation
Serve Llama 3.1 8B and 70B using vLLM on GH200: documentation
Fine-tune a video generation model (Mochi) on GH200: documentation