Intel Gaudi Enables a Lower Cost Alternative for AI Compute and GenAI

Intel submits Gaudi 2 results on MLCommons’ newest benchmark by fine-tuning Llama 2 70B using low-rank adapters and training MLPerf GPT-3 model with 1,000+ Gaudi 2s on Intel Tiber Developer Cloud.


  • June 12, 2024

  • Contact Intel PR

  • Follow Intel Newsroom on social:

    Twitter logo
    YouTube Icon



The Intel Gaudi 2 AI accelerator remains the only benchmarked alternative to Nvidia H100 for generative AI performance. (Credit: Intel Corporation)

What’s New: Today, MLCommons published results of its industry AI performance benchmark, MLPerf Training v4.0. Intel’s results demonstrate the choice that Intel® Gaudi® 2 AI accelerators give enterprises and customers. Community-based software simplifies generative AI (GenAI) development and industry-standard Ethernet networking enables flexible scaling of AI systems. For the first time on the MLPerf benchmark, Intel submitted results on a large Gaudi 2 system (1,024 Gaudi 2 accelerators) trained in Intel® Tiber™ Developer Cloud to demonstrate Gaudi 2 performance and scalability and Intel’s cloud capacity for training MLPerf’s GPT-3 175B parameter benchmark model.

“The industry has a clear need: address the gaps in today’s generative AI enterprise offerings with high-performance, high-efficiency compute options. The latest MLPerf results published by MLCommons illustrate the unique value Intel Gaudi brings to market as enterprises and customers seek more cost-efficient, scalable systems with standard networking and open software, making GenAI more accessible to more customers.”

–Zane Ball, Intel corporate vice president and general manager, DCAI Product Management

Why It Matters: More customers want to benefit from GenAI but are unable to because of cost, scale and development requirements. With only 10% of enterprises successfully moving GenAI projects into production last year, Intel's AI offerings address the challenges businesses face in scaling AI initiatives. Intel Gaudi 2 is an accessible, scalable solution that has proven its ability to handily train large language models (LLMs) from 70 billion to 175 billion parameters. The soon-to-be-released Intel® Gaudi® 3 accelerator will bring a leap in performance, as well as openness and choice to enterprise GenAI.

How Intel Gaudi 2 MLPerf Results Demonstrate Transparency: The MLPerf results show Gaudi 2 continues to be the only MLPerf-benchmarked alternative for AI compute to the Nvidia H100. Trained on the Tiber Developer Cloud, Intel’s GPT-3 results for time-to-train (TTT) of 66.9 minutes on an AI system of 1,024 Gaudi accelerators proves strong Gaudi 2 scaling performance on ultra-large LLMs within a developer cloud environment1.

The benchmark suite featured a new measurement: fine-tuning the Llama 2 70B parameter model using low-rank adapters (LoRa). Fine-tuning LLMs is a common task for many customers and AI practitioners, making it a relevant benchmark for everyday applications. Intel’s submission achieved time-to-train of 78.1 minutes on eight Gaudi 2 accelerators. Intel utilized open source software from Optimum Habana for the submission, leveraging Zero-3 from DeepSpeed for optimizing memory efficiency and scaling during large model training, as well as Flash-Attention-2 to accelerate attention mechanisms. The benchmark task force – led by the engineering teams from Intel’s Habana Labs and Hugging Face – are responsible for the reference code and benchmark rules.

How Intel Gaudi Provides Customers with Value in AI: To date, high costs have priced too many enterprises out of the market. Gaudi is starting to change that. At Computex, Intel announced that a standard AI kit including eight Intel Gaudi 2 accelerators with a universal baseboard (UBB) offered to system providers at $65,000 is estimated to be one-third the cost of comparable competitive platforms. A kit including eight Intel Gaudi 3 accelerators with a UBB lists at $125,000, estimated to be two-thirds the cost of comparable competitive platforms2.

The proof is in increased momentum. Customers use Gaudi for the value it brings with price-performance advantages and accessibility, including:

  • Naver, a South Korean cloud service provider and leading search engine catering to more than 600 million users, is building a new AI ecosystem and lowering barriers to enable wide-scale LLM adoption by reducing development costs and project timelines for its customers.
  • AI Sweden, an alliance between the Swedish government and private business, leverages Gaudi for fine-tuning with domain-specific municipal content to improve operational efficiencies and enhance public services for Sweden’s constituents.

How Intel Tiber Developer Cloud Supports Customers Accessing Gaudi: The Tiber Developer Cloud provides customers a unique, managed and cost-efficient platform to develop and deploy AI models, applications and solutions – from single nodes to large cluster-level compute capacity. This platform increases access to Gaudi for AI compute needs. In the Tiber Developer Cloud, Intel makes its accelerators, CPUs, GPUs, an open AI software stack and other services easily accessible. Intel customer Seekr recently launched its new product SeekrFlow, an AI development platform for trusted AI, to serve its customers from  Intel’s developer cloud.

According to, Seekr cited cost savings of 40% up to 400% from the Tiber Developer Cloud for select AI workloads compared to on-premise systems with another vendor’s GPUs and with another cloud service provider, along with 20% faster AI training and 50% faster AI inference than on-premise3.

What’s Next: Intel will submit MLPerf results based on the Intel Gaudi 3 AI accelerator in the upcoming inference benchmark. Intel Gaudi 3 accelerators are projected to provide a leap in performance for AI training and inference on popular LLMs and multimodal models and will be generally available from original equipment manufacturers in fall of 2024.

More ContextMLCommons Announcement

The Small Print:

For workloads and configurations, visit Results may vary.

MLPerf's GPT-3 measurement is conducted on a 1% representative slice of the entire model as determined by the participating companies who collectively devise the MLCommons benchmark.

Pricing guidance for cards and systems is for modeling purposes only. Please consult your original equipment manufacturer (OEM) of choice for final pricing. Results may vary based upon volumes and lead times.