Skip to content
Updated minutes ago
google

TPU v6e (Trillium)

google · tpu · 32 GB HBM · 200W TDP

VRAM

32 GB

BF16 TFLOPS

460

Bandwidth

1640 GB/s

From

$1.75/hr

Calculate ROI with this GPU →

Spec Sheet

VRAM32 GB HBM
Memory Bandwidth1640 GB/s
BF16 TFLOPS460
FP16 TFLOPS460
FP8 TFLOPS920
INT8 TOPS920
TDP200W
InterconnectPCIE
Max per Node256
PCIe Gen5
Tensor CoresNo

Pricing by Provider

ProviderOn-DemandReservedSpotBadge
gcp$2.50/hr$1.75/hr-Cheapest

Compatible Models (282)

Training Capabilities

Estimated GPU count for full fine-tuning (AdamW, BF16) and QLoRA

Model SizeFull Fine-TuneQLoRA
7B model5 GPUs1 GPU
13B model8 GPUs1 GPU
70B model42 GPUs2 GPUs

Energy Efficiency

Estimated tokens/second per Watt for popular models

Mistral 7B
1.12 t/s/WFP8
Qwen 2.5 7B
1.08 t/s/WFP8
Llama 3.1 8B
1.02 t/s/WFP8
DeepSeek V3
0.22 t/s/WFP8
Llama 3.1 70B
0.12 t/s/WFP8
Qwen 2.5 72B
0.11 t/s/WFP8

Similar GPUs

GPUVRAMBF16 TFLOPSBW (GB/s)From
TPU v432 GB2751200$2.25/hr
TPU v5e16 GB200820$0.85/hr
RTX 509032 GB2101792$0.89/hr
V100 32GB32 GB28.3900$0.19/hr
Instinct MI10032 GB184.61229$0.40/hr

Methodology Note

Performance estimates for the TPU v6e (Trillium)are based on InferenceBench's roofline performance model with CUDA kernel-level optimization including FlashAttention v2 and PagedAttention. Memory calculations account for model weights (32 GB HBM available), KV-cache allocation, and activation memory. Throughput predictions use the TPU v6e (Trillium)'s rated 1640 GB/s memory bandwidth and 460 BF16 TFLOPS compute capacity as roofline ceilings, with empirical correction factors per GPU architecture (tpu). See our full methodology.

Frequently Asked Questions

How many AI models can run on TPU v6e (Trillium)?

The TPU v6e (Trillium) can run 282 AI models from our database within a single node. Compatible models range across various parameter sizes depending on the quantization precision (BF16, FP8, INT4). Smaller models fit on a single GPU while larger models may require multi-GPU setups up to 256x TPU v6e (Trillium).

What is the TPU v6e (Trillium) inference throughput?

The TPU v6e (Trillium) delivers 460 BF16 TFLOPS and 920 FP8 TFLOPS with 1640 GB/s memory bandwidth. Actual inference throughput (tokens/sec) depends on the model size, precision, and batch size. Use our calculator for model-specific throughput estimates.

How much does TPU v6e (Trillium) cost per hour?

The TPU v6e (Trillium) is available starting from $1.75/hour via gcp. Prices vary by provider and pricing tier (on-demand, reserved, spot). Compare pricing across all providers in the table above.