SambaNova vs Tenstorrent
September 28, 2025
SambaNova ($1.1B)
Tenstorrent ($1B)
SambaNova and Tenstorrent are both startups, creating hardware for artificial intelligence. In addition to hardware accelerators, both companies develop complex software and compilers to optimize the entire system, offering their solutions for both from the cloud and on-premise. Both use low-precision formats (e.g. FP16) to accelerate computations and strive to achieve an optimal ratio of performance / throughput per watt, as well as the density of operations for a given power budget. Both companies focus on open technologies and standards (e.g. RISC-V or open software).
But SambaNova (founded in 2017) is an American startup that uses architectures focused on optimizing dataflow within and between chips, optimized for LLM and generative AI - for example SN40L, a system of RDUs (reconfigurable dataflow units). This architecture supports the "composition of experts" mechanism - when several specialized models work on certain parts of the data/tasks. It also uses advanced memory blocks and networks-on-chip aproach.
Tenstorrent (2016) is a Canadian company focused on using the standard RISC-V architecture and combining its AI accelerators with CPUs. Its chips have a large amount of SRAM, local LPDDR memory, allow to adjust settings for different precision formats (FP8, FP16, BFLOAT16, etc.). The company also produces a line of scalable hardware solutions Wormhole that include network connections between chips, PCIe cards - which are effective for cooling and have low power consumption. In general, Tenstorrent has a more open approach to creating a software stack.
But SambaNova (founded in 2017) is an American startup that uses architectures focused on optimizing dataflow within and between chips, optimized for LLM and generative AI - for example SN40L, a system of RDUs (reconfigurable dataflow units). This architecture supports the "composition of experts" mechanism - when several specialized models work on certain parts of the data/tasks. It also uses advanced memory blocks and networks-on-chip aproach.
Tenstorrent (2016) is a Canadian company focused on using the standard RISC-V architecture and combining its AI accelerators with CPUs. Its chips have a large amount of SRAM, local LPDDR memory, allow to adjust settings for different precision formats (FP8, FP16, BFLOAT16, etc.). The company also produces a line of scalable hardware solutions Wormhole that include network connections between chips, PCIe cards - which are effective for cooling and have low power consumption. In general, Tenstorrent has a more open approach to creating a software stack.
★
See also:
