Parts/Accelerators/Google TPU v5e
Accelerators · v5e · 2023
Google TPU v5e
197 TFLOPS BF16 per chip, 16 GB HBM, 256-chip pods on a 2D torus. Offered only as rented Cloud TPU capacity.
Why this part
Single TensorCore per chip. Google positions v5e as a combined training and serving part.
Systems using this part
No systems in the current dataset record this part. It is carried in the taxonomy so the graph is complete, and so edges can attach to it as coverage grows.