New GPU servers with an advanced Intel® Habana® Gaudi® 2 – already on LeaderGPU ®!

Try out the GPU designed specifically for AI training!

The Intel® Habana® Gaudi® 2 is one of the top graphics cards for AI training, deep learning and inference, especially for large language models (LLMs) such as GPT-3, BLOOM, Dolly, and Stable Diffusion.

This GPU is based on Gaudi® 2 architecture, an improved version of Intel's first-generation Gaudi®, designed with 7nm process technology and 24 Tensor Processor Cores. Integrated media control and dual matrix multiplication engines provide fast AI training and deep learning performance.

The Tensor Core supports multiple AI data types such as FP8, FP16, BF16, FP32 and TF32.

It is the first card that provides 24,100 gigabit Ethernet integrated on chip (RoCEv2). This is an remarkable difference compared to the 10,100 gigabit Ethernet ports on the original generation.

The Habana® uses the power of 96GB HBM2E memory. It’s PCIe maximum bandwidth is an impressive 2,45 TB/s.

Want to try these cards in action?

We have great news for you! Servers with Habana® Gaudi® 2 GPUs are available with monthly, weekly, day and minute billing.

Don’t miss out! Order these cutting-edge servers now — there’s already a queue of eager customers waiting for them!

If you have any questions, feel free to contact our support team via chat on our website www.leadergpu.com or email us at info@leadergpu.com.