Skip to main content

NVIDIA B300 GPUs Now Available in IBM Cloud Code Engine Serverless Fleets

The AI landscape is moving at a breakneck pace, and today we are shifting into a new gear. We are thrilled to announce that NVIDIA B300 (Blackwell Ultra) GPUs are now also available in IBM Cloud Code Engine Serverless Fleets.

By bringing NVIDIA’s most powerful Blackwell-architecture GPU to our serverless platform, we are giving developers the ability to run the world’s most demanding reasoning models and massive-scale simulations without the burden of infrastructure management.

Why Use B300 with Serverless Fleets?
#

  • “Serverless Fleets" make the B300 accessible and cost-effective. While Blackwell GPUs are power-hungry (1,400W TDP) and expensive to reserve, Code Engine changes the economics:

  • Zero Idle Costs: The B300 is a high-performance asset. In a traditional setup, you pay for it 24/7. With Code Engine, your Fleet scales to zero when your job is done. Billing starts the moment a GPU is initialized, and you are only charged for the GPU‑seconds it is active.

  • Massive Parallelism without the Headaches: Serverless Fleets allow you to submit thousands of tasks to a single endpoint. Code Engine automatically provisions the B300 resources, distributes the tasks, and decommissions the GPUs when the work is complete.

  • Blackwell-Ready Abstraction: No need to configure liquid cooling or 800Gbps networking. We handle the intense infrastructure requirements of the Blackwell architecture so you can focus on your vllM or PyTorch code.

  • Seamless Model Fitting: The massive 288GB VRAM means you can run larger batch sizes or longer context windows (up to 128k+ tokens) on a single node, significantly reducing the latency overhead of inter-GPU communication.

Ready to Scale?
#

Whether you are fine-tuning the latest Granite models or running high-throughput inference for a global application, the B300 on Code Engine is your new high-performance home.

The era of Blackwell has arrived. Scale your intelligence, not your operations.

 Author
Author
Uwe Fassnacht
Product Manager for IBM Cloud Code Engine