A short thread on accelerated computing. How $TSLA's #Dojo, $GOOG's #TPU #Cerebras's #WSE-2, #Graphcore's #IPU, and others compare to $NVDA 's #GPU. This is an extract of work we published a few weeks ago. 👇1/6
It is difficult to compare chips. Designing a chip is all about managing trade-offs between multiple dimensions. A chip has limited potential resources and the architect allocates them over these multiple dimensions. 2/6 
With this framework in mind, one sees alternatives to #GPU have fundamentally different architectures, favoring the flow of data across the chip vs. the flow of data between the memory and the chip. 3/6
These chips, and in particular $TSLA 's #Dojo, can move data across an order of magnitude faster than a #GPU, but this comes at a high cost: lack of flexibility: data can flow that fast only under hard constraints: e.g. neighbor to neighbor or during a data transfer phase) 4/6
Implications: $TSLA 's neural net will run on #Dojo 5-10x faster than on #GPUs. Major advantage for Tesla, learning overnight vs. every week; but Dojo is no threat to $NVDA's #GPU: Too inflexible for others to adopt, and will always materially lag GPUs in terms of ecosystems. 5/6
This is only a very tiny extract of the research we do. If you are an institutional investor and are interested in our work, please reach out! 6/6
• • •
Missing some Tweet in this thread? You can try to
force a refresh
