Google TPU 8t
Overview
Google's eighth-generation training TPU, announced at Google Cloud Next in April 2026. Google publishes a peak of 12.6 PFLOPS of FP4 per chip, with 216 GB of HBM3e at 6,528 GB/s and 128 MB of on-chip Vmem SRAM. Its scale is the notable part: 9,600 chips form a single superpod on a 3D torus with 47 petabits per second of non-blocking bisectional bandwidth, and Google's Virgo network can link over 134,000 of them. FP4 is the only precision Google states for this chip; no FP8, BF16 or INT8 figure is published, and none is inferred from earlier TPU generations. Google publishes no sparse figure for any TPU, so this is a dense rate.
Performance
Peak theoretical throughput by precision type
| Precision | Peak |
|---|---|
FP64 | No verified data available |
FP32 | No verified data available |
TF32 | No verified data available |
BF16 | No verified data available |
FP16 | No verified data available |
FP8 | No verified data available |
FP6 | No verified data available |
FP4 4-bit floating point | 12,600TFLOPS |
INT8 | No verified data available |
Every figure here is dense. The vendor documents no structured (2:4) sparsity mode for this part, so there is no second number to quote.
Specifications
Architecture
TPU 8
Form Factor
No verified data available
Launch Year
2026
Process Node
No verified data available
Memory
216 GB HBM3e
Bandwidth
6,528 GB/s
TDP
No verified data available
Max power
No verified data available
Spec Confidence
Vendor claimed
Full Specifications
| Memory | |
|---|---|
| Memory | 216 GB |
| Memory Type | HBM3e |
| Bandwidth | 6.5 TB/s |
| Interface Width | No verified data available |
| On-Die SRAM | 128 MB |
| Power & Thermal | |
| TDP | No verified data available |
| Max power | No verified data available |
| Enterprise Features | |
| Sparsity | No |
| General | |
| Form Factor | No verified data available |
| Architecture | TPU 8 |
| Process Node | No verified data available |
| Launch Year | 2026 |
Datasheet & Resources
Data Provenance
Every figure traced to a source
Primary Source
- Document
- —
- Publisher
- Published
- No verified data available
Data Quality
- Spec confidence
- Vendor claimed
- Clock basis
- Boost
- Core precisions with figures
- 1 of 9
- Normalization
- All values in TFLOPS
Compare cloud providers offering on-demand GPU instances for AI training, inference, and HPC workloads.
Browse GPU Cloud ProvidersSimilar GPUs
Systems Using This GPU
Pre-configured systems featuring the Google TPU 8t
| System | GPU Count | Peak Performance | Total Power | |
|---|---|---|---|---|
Google TPU 8t Superpod TPU · pod · 2026 | 9600x TPU 8t | 121 EFLOPS | No verified data available | View System |
Frequently Asked Questions
How many TFLOPS does the Google TPU 8t have?
The verified peak figures Flopper holds for the Google TPU 8t are 12,600 TFLOPS FP4. Flopper does not currently have verified FP32, FP16 and FP8 throughput figures for it.
What is the power consumption of the Google TPU 8t?
Flopper does not currently have a verified TDP (Thermal Design Power) figure for the Google TPU 8t.
How much memory does the Google TPU 8t have?
The Google TPU 8t is equipped with 216 GB of memory with 6,528 GB/s of memory bandwidth.
What architecture is the Google TPU 8t based on?
The Google TPU 8t is based on the TPU 8 architecture, launched in 2026.
Stay Updated on GPU Releases
Get notified when new GPUs are added or specifications are updated.
No spam, unsubscribe anytime.