![]()
Mon Oct 07 23:22:29 UTC 2024: ## Inflection AI Switches to Intel Gaudi 3 for Enterprise AI Platform
**Inflection AI, known for its conversational AI assistant “Pi,” is making a major shift in its enterprise platform, opting for Intel’s Gaudi 3 accelerators over Nvidia GPUs.** This move comes after Inflection 3.0, the latest version of its platform, focuses on fine-tuning models with enterprise-specific data to build custom AI applications.
While Inflection previously relied on Nvidia GPUs for its customer application, Inflection 3.0 will utilize Gaudi 3 with instances hosted either on-premise or in the cloud via Intel’s Tiber AI Cloud service. Intel itself is one of the first customers to adopt the service.
**The decision to switch to Gaudi 3 was driven by its cost-effectiveness and performance.** According to Inflection CEO Sean White, Gaudi 3 offers up to 2x improved price performance compared to current competitive offerings. Notably, Gaudi 3 boasts a 128 GB HBM2e memory with 3.7 Tbps bandwidth and 1,835 teraFLOPS of dense FP8 or BF16 performance, making it particularly efficient for training and fine-tuning workloads.
**This move marks a significant win for Intel in the AI accelerator market.** Despite facing competition from Nvidia’s H100 and AMD’s MI325X GPUs, Intel is aggressively pricing its Gaudi 3 system, which costs roughly two-thirds of an equivalent H100 system.
**The transition to Gaudi 3 will not restrict customers to using the same hardware for model deployment.** While Inflection will be running its platform on Gaudi 3, customers will have the flexibility to choose their preferred hardware for running their finished models.
**Inflection’s commitment to Gaudi 3 also comes at a time when Intel is transitioning to its new Falcon Shores GPU.** This next-generation GPU will integrate Xe graphics DNA with Habana’s technology. While Intel assures seamless migration for high-level frameworks, additional guidelines will be provided for developers working at a lower level.
This strategic partnership between Inflection AI and Intel signifies a growing trend towards more diverse hardware options in the rapidly evolving field of AI.