4 ms·

My guess is that the Tesla Autopilot Software team is working on language/compiler abstractions to reduce their dependency on semiconductor hardware manufacture

by n00b101 10y ago

My guess is that the Tesla Autopilot Software team is working on language/compiler abstractions to reduce their dependency on semiconductor hardware manufacturers, particularly NVIDIA [1]. NVIDIA hardware has been Tesla cars for many years, and they have a dominating position in both embedded and data center hardware for self-driving AI, and they are not afraid to partner with Tesla's competitors [2]. This whole ecosystem is based on CUDA toolchain, which is built atop LLVM [3]. And Chris Lattner is the original developer of LLVM. [1] https://www.nvidia.com/object/tesla-and-nvidia.html https://www.nvidia.com/object/tesla-and-nvidia.html [2] https://www.nvidia.com/object/audi-and-nvidia.html https://www.nvidia.com/object/audi-and-nvidia.html [3] http://docs.nvidia.com/cuda/nvvm-ir-spec/#axzz4VMqXIKfo http://docs.nvidia.com/cuda/nvvm-ir-spec/#axzz4VMqXIKfo

peller 10y ago
Considering the recent hire of Tesla's new Vice President of Autopilot Hardware Engineering team was Jim Keller[0], the lead architect for AMD's K8 (Athlon 64) and a very influential force behind both Apple's A* chips and AMD's Zen (Ryzen), I think you're right that they're looking to rely less on NVIDIA. My guess would be that Tesla's ambitions are set on controlling the whole stack. Something along the lines of a custom software toolchain for tailor-built silicon. [0] https://en.wikipedia.org/wiki/Jim_Keller_(engineer) https://en.wikipedia.org/wiki/Jim_Keller_(engineer)
localhost 10y ago
Or perhaps they want to go down the road of FPGA / ASIC. There is some LLVM based FPGA tooling around today [1], but perhaps they want to uplevel the core programming language to a new programming language? Chris has done this already from Obj-C to Swift. [1] http://llvm.org/devmtg/2014-10/Slides/Baker-CustomHardwareStateMachines.pdf http://llvm.org/devmtg/2014-10/Slides/Baker-CustomHardwareSt...

by n00b101 10y ago

peller 10y ago

Considering the recent hire of Tesla's new Vice President of Autopilot Hardware Engineering team was Jim Keller[0], the lead architect for AMD's K8 (Athlon 64) and a very influential force behind both Apple's A* chips and AMD's Zen (Ryzen), I think you're right that they're looking to rely less on NVIDIA. My guess would be that Tesla's ambitions are set on controlling the whole stack. Something along the lines of a custom software toolchain for tailor-built silicon. [0] https://en.wikipedia.org/wiki/Jim_Keller_(engineer) https://en.wikipedia.org/wiki/Jim_Keller_(engineer)

localhost 10y ago

Or perhaps they want to go down the road of FPGA / ASIC. There is some LLVM based FPGA tooling around today [1], but perhaps they want to uplevel the core programming language to a new programming language? Chris has done this already from Obj-C to Swift. [1] http://llvm.org/devmtg/2014-10/Slides/Baker-CustomHardwareStateMachines.pdf http://llvm.org/devmtg/2014-10/Slides/Baker-CustomHardwareSt...