Intel Gaudi 3 AI Accelerator Launches on IBM Cloud; Outperforms NVIDIA H100 & H200 AI GPUs in Performance Tests

Intel’s Gaudi 3 AI accelerator has recently made its debut on IBM Cloud, marking a significant milestone as the first cloud service provider (CSP) to integrate this cutting-edge technology into its offerings.

IBM Cloud Leads with Intel’s Gaudi 3 AI Accelerator

Despite Intel’s ongoing commitment to AI technologies, the company’s entry into the market has not matched the early momentum of its competitors. Thus far, the adoption of Intel’s Gaudi AI products has been relatively slow. However, the recent unveiling by IBM Cloud of the Gaudi 3 AI accelerators presents a promising strategy to provide affordable AI solutions to businesses.

By bringing Intel Gaudi 3 AI accelerators to IBM Cloud, we’re enabling businesses to help scale generative AI workloads with optimized performance for inferencing and fine-tuning. This collaboration underscores our shared commitment to making AI more accessible and cost-effective for enterprises worldwide.

– Saurabh Kulkarni, VP at Intel

Users can leverage the Gaudi 3 capabilities via IBM’s Cloud Virtual Servers. Notably, Signal65 has conducted tests that compare the performance of Gaudi 3 against NVIDIA’s H100 and H200 GPUs, marking the first independent scrutiny of Intel’s latest accelerator outside of its internal assessments.

LLM Inference Throughput - Input=1024, Output=1024

The tests reveal that Intel’s Gaudi 3 outperformed NVIDIA’s offerings in benchmarks involving IBM’s Granite (8B) and Meta’s Llama-3.1 models. Although the specific parameters for the tests varied, resulting performance may differ in practical applications. These preliminary benchmarks offer critical insights into the potential of Intel in the competitive AI landscape.

Source & Images