Dojo
Tesla's self-developed AI training supercomputer, which uses custom chips to lower the cost of neural network training and reduce dependence on external GPU supply.
Timeline (last 90 days)
Tesla's FSD chip, Dojo supercomputer, pure vision perception solution and vehicle electronic architecture are all developed in-house.
Tesla on the one hand purchases Nvidia GPUs to build clusters, and on the other hand develops its own Dojo supercomputer, seeking to use custom chips to lower training costs and reduce dependence on external suppliers.
Tesla's AI training infrastructure simultaneously supports AI products such as FSD and Optimus, with reusability and economies of scale.
Meta's MTIA, Amazon's Trainium/Inferentia, and Tesla's Dojo are all examples of custom in-house chips developed by major companies for silicon-level customization of their own workloads
Tesla's Dojo uses a custom D1 chip integrating 50 billion transistors per chip, manufactured on TSMC's 7nm process, with 25 D1 chips forming one training module providing approximately 9 PetaFLOPS of computing power
By the end of 2024, Tesla acknowledged that the actual deployment progress of Dojo was slower than expected, with primary training still relying on NVIDIA H100 GPU clusters
All Facts (7)
Musk旗下AI相关业务包括xAI（Grok大语言模型）、特斯拉（自动驾驶AI芯片与Dojo超级计算机）、Neuralink（脑机接口专用芯片）和SpaceX
90%UnverifiedTesla's FSD chip, Dojo supercomputer, pure vision perception solution and vehicle electronic architecture are all developed in-house.
50%UnverifiedTesla on the one hand purchases Nvidia GPUs to build clusters, and on the other hand develops its own Dojo supercomputer, seeking to use custom chips to lower training costs and reduce dependence on external suppliers.
50%UnverifiedTesla's AI training infrastructure simultaneously supports AI products such as FSD and Optimus, with reusability and economies of scale.
50%UnverifiedMeta's MTIA, Amazon's Trainium/Inferentia, and Tesla's Dojo are all examples of custom in-house chips developed by major companies for silicon-level customization of their own workloads
50%UnverifiedBy the end of 2024, Tesla acknowledged that the actual deployment progress of Dojo was slower than expected, with primary training still relying on NVIDIA H100 GPU clusters
50%UnverifiedTesla's Dojo uses a custom D1 chip integrating 50 billion transistors per chip, manufactured on TSMC's 7nm process, with 25 D1 chips forming one training module providing approximately 9 PetaFLOPS of computing power
50%