Google TPU 8t vs Google TPU v5p 95GB

Side-by-side specifications, performance, and rental pricing for datacenter AI workloads.

TPU 8t: TPU 8, 2026 TPU v5p: TPU v5p, 2023
Memory
216 vs 95 GB
121 GB more for the TPU 8t
Bandwidth
6.5 TB/s vs 2.8 TB/s
TPU 8t moves data faster
Download comparison sheet

Share this page

Google TPU 8t vs Google TPU v5p 95GB

A post with this link shows the card on the Image tab as its preview.

Printable version →

Google TPU 8t

Full specs →

What is the Google TPU 8t?

The TPU 8t is a TPU from Google on the TPU 8 architecture, an accelerator built for AI training and inference. It has been announced but is not yet shipping. TPU 8t full specifications →

Can I rent the TPU 8t in the cloud?

Flopper is not tracking any live rental listings for the TPU 8t yet. The provider directory lists the clouds that rent comparable accelerators on demand. GPU cloud providers →

How does the TPU 8t compare with other accelerators?

This page sets the TPU 8t against the TPU v5p. The TPU 8t is the newer generation of the two. Both are Google parts. The compare tool builds the same comparison for any two parts in the catalogue. All GPU comparisons →

Where do the TPU 8t figures come from?

Google published the TPU 8t figures in marketing material rather than a datasheet, and Flopper has no independent confirmation of them. How Flopper verifies specifications →

Google TPU v5p 95GB

Full specs →

What is the Google TPU v5p?

The TPU v5p is a TPU from Google on the TPU v5p architecture, an accelerator built for AI training and inference. TPU v5p full specifications →

Can I rent the TPU v5p in the cloud?

Flopper is not tracking any live rental listings for the TPU v5p yet. The provider directory lists the clouds that rent comparable accelerators on demand. GPU cloud providers →

How does the TPU v5p compare with other accelerators?

This page sets the TPU v5p against the TPU 8t. The TPU v5p is the older generation of the two. Both are Google parts. Flopper also compares it head to head with the H100 SXM5 80GB, across every published precision, memory and power. TPU v5p vs H100 SXM5 80GB →

Where do the TPU v5p figures come from?

Every TPU v5p figure on this page traces to Google's own datasheet or specification table. How Flopper verifies specifications →

Specifications

TPU 8t
TPU v5p
Architecture
TPU 8TPU v5p
Launch Year
20262023
Form Factor
—Custom
Memory
216 GB95 GB
Memory Bandwidth
6.5 TB/s2.8 TB/s
TDP
—450 W
Process Node
—5nm

Performance (TFLOPS)

TPU 8t
TPU v5p
No verified data available No verified data available
No verified data available No verified data available
No verified data available No verified data available
No verified data available459 TFLOPS
No verified data available No verified data available
No verified data available459 TFLOPS
FP6
No verified data available No verified data available
12,600 TFLOPS No verified data available
No verified data available No verified data available

Figures are dense. Where a vendor publishes a figure it sits under the dense one and never decides a winner. Hover any precision for its definition, or read the GPU glossary.

FLOPS by Precision

What actually differs

The TPU 8t is the newer part: TPU 8 (2026) against the TPU v5p's TPU v5p (2023).

The TPU 8t carries 121 GB more memory (216 GB against 95 GB), so larger models fit on one GPU before you shard. Bandwidth is 6.5 TB/s against 2.8 TB/s, which sets the ceiling on decode speed.

Weight the FP8 and FP16 rows for LLM work and the FP64 row for HPC.

New to the terms? The glossary explains BF16, FP8, FP4, Sparsity and TFLOPS.

Frequently asked questions

Which should you choose, the TPU 8t or the TPU v5p?

Choose the TPU 8t when the model has to fit on one GPU: it carries 121 GB more memory.

Which has more memory, the TPU 8t or the TPU v5p?

The TPU 8t carries 216 GB of memory versus 95 GB for the TPU v5p. Memory bandwidth is 6.5 TB/s for the TPU 8t and 2.8 TB/s for the TPU v5p.

Get Comparison Updates

New GPUs added weekly. Be the first to see how they compare.

© 2026 Flopper.io - Compare the GPUs Powering AI