TPU v6e (Trillium) specs
Google · Shipping now · first available Oct 2024 (estimated). Dense compute, memory, bandwidth and power from the vendor's documents; each precision stays separate.
All accelerators · Compare with TPU v5p
Specifications
| Dense BF16 · FP32 accumulate | 918 TFLOPS |
|---|---|
| Dense FP8 · FP32 accumulate | 918 TFLOPS |
| Dense FP4 · FP32 accumulate | Not published |
| Memory | 34.36 GB · HBM32 GiB as published |
| Memory bandwidth | 1.64 TB/s |
| Maximum chip power | Not published |
| Interconnect | ICI · 256-chip pods |
| First availability | Oct 2024est — estimated / secondary-source figure |
| Availability | Shipping |
4.7× the peak compute of v5e.
Dense tensor rates with FP32 accumulation, never sparse or FP16-accumulate headline figures. Memory is in decimal GB. The est marker flags a value its vendor-document review did not confirm.
Rent it
$2.70/chip-hrCheapest on-demand · Google Cloud · observed
1 provider: Google Cloud. Quotes checked through .
TPU v6e rental prices from every provider →Value ratios
| $ per dense BF16 PFLOP-hour | $2.94 |
|---|---|
| Dense BF16 TFLOPS per watt | No power figure |
| Bandwidth bytes per 1,000 BF16 FLOP | 1.79 |
| $ per dense FP8 PFLOP-hour | $2.94 |
| Dense FP8 TFLOPS per watt | No power figure |
| Bandwidth bytes per 1,000 FP8 FLOP | 1.79 |
| $ per GB of memory per hour | $0.079 |
Prices divide the cheapest on-demand list price; theoretical ratios do not measure application speed.
Sources
Specifications reviewed . Per-chip architecture comparison; FP8 throughput is explicitly listed, not inferred from BF16.
Rental price: Google Cloud published pricing ↗ (Trillium / TPU v6e; per chip-hour).