Google TPU · architecture + economics

From systolic arrays to specialized AI systems.

Google’s TPU roadmap increasingly separates training, embedding, post-training and serving needs at the system level. This guide explains the silicon, memory hierarchy, topology and public cloud economics behind that shift.

Evolution of specialization

6
Trillium / v6e
Versatile, cost-oriented training + serving
7
Ironwood / TPU7x
Large HBM, 3D torus, strong scale-up
8t
Training specialization
3D torus, SparseCore, 9,600-chip superpod
8i
Serving specialization
Boardfly, 384 MB SRAM, CAE

Current TPU family

The important 2026 distinction is specialization.

v5e and Trillium are cost/versatility-oriented. v5p and Ironwood emphasize large-scale performance. Google’s eighth generation explicitly splits large-scale pre-training (8t) from reasoning/post-training/serving (8i).

TPUStatusPrimary roleHBMHBM bandwidthScale/topology
TPU v5eAvailableCost-efficient training & inference16 GB819 GB/sUp to 256 chips · 2D torus
TPU v5pAvailableLarge-scale training95 GiB2.765 TB/sUp to 8,960 chips · 3D torus
Trillium / v6eAvailableTraining, fine-tuning & serving32 GiB1.638 TB/s256 chips · 2D torus
Ironwood / TPU7xAvailableHigh-performance training & inference192 GiB7.38 TB/sUp to 9,216 chips · 3D torus
TPU 8tComing soonMassive pre-training & embeddings216 GB6.528 TB/s9,600-chip superpod · 3D torus + Virgo
TPU 8iComing soonPost-training, reasoning & serving288 GB8.601 TB/sBoardfly · up to 1,152 connected chips

Silicon architecture

TPUs are systems for structured tensor dataflow.

The useful mental model is not “a GPU with fewer features.” TPU performance comes from co-design across matrix engines, vector units, on-chip scratchpad, HBM, inter-chip fabric and compiler scheduling.

Memory hierarchy

1
HBM
Large model and activation storage; still a bottleneck for memory-bound operations.
2
VMEM / on-chip SRAM
High-bandwidth scratchpad close to compute; explicit blocking and prefetch strategy matter.
3
MXU + vector units
Matrix work and vector operations must overlap well to keep the chip utilized.

Specialized blocks

SparseCore offloads irregular embedding-related work in several TPU generations. TPU 8t continues that path for embedding-heavy training. TPU 8i replaces SparseCore with a Collectives Acceleration Engine designed to reduce synchronization latency for serving and reasoning.

This is why “TPU” is no longer one architecture profile: the eighth generation deliberately changes specialized blocks and topology by workload.

Conceptual systolic-array demo

This animation illustrates regular data reuse through neighboring processing elements. It is not cycle-accurate hardware behavior.

Cycle: 0

Why compilers matter

The XLA stack can reason globally about tensor placement, sharding and communication. JAX remains central, while PyTorch/XLA is mature and Google has continued expanding native PyTorch support. Portability is much better than early TPU generations, but performance still depends on mapping the model and serving stack to TPU topology.

Pod networking

Network topology follows communication pattern.

2D torus

v5e / Trillium

Compact slices prioritize cost-efficient training, fine-tuning and serving. The topology fits the scale profile of these versatile systems.

3D torus

v5p / Ironwood / 8t

Neighbor-oriented communication scales to very large pods and is well suited to dense training collectives and tensor/model parallel workloads.

Boardfly

TPU 8i

Serving and MoE workloads need lower all-to-all latency. Google says Boardfly reduces network diameter from 16 hops to seven in a representative 1,024-chip configuration.

OCS is not “instant fault healing.” Optical circuit switching gives Google a reconfigurable fabric and supports resilient large-scale topologies, but application-level recovery still depends on the full software/runtime stack.

LLM workloads

Training and reasoning now diverge enough to justify different TPU systems.

TPU 8t

Throughput-oriented training

Google positions 8t for massive-scale pre-training and embedding-heavy workloads. It uses a 3D torus, SparseCore, 216 GB HBM, 128 MB SRAM, native FP4 and a 9,600-chip superpod; Virgo provides large scale-out bandwidth beyond a superpod.

TPU 8i

Reasoning and high-concurrency serving

8i raises on-chip SRAM to 384 MB, uses 288 GB HBM, adds the Collectives Acceleration Engine and switches to Boardfly topology to reduce synchronization/all-to-all latency for MoE and reasoning workloads.

Implication: treat prefill, decode, post-training and training as separate workload classes. A future architecture can route them to different TPU pools instead of assuming one universal accelerator.

Cloud economics

Current public TPU prices are transparent enough for capacity planning.

Google publishes prices per chip-hour. Rates vary by region and consumption model; the table below uses representative U.S. regions from the current Google Cloud pricing page.

TPU / regionOn demandDWS Flex-startDWS Calendar1-year3-year
TPU v5e · us-central1$1.20$0.60$0.84$0.84$0.54
Trillium (v6e) · us-east1$2.70$1.35$1.89$1.89$1.22
TPU v5p · us-east1$4.20$2.10$2.94$2.94$1.89
Ironwood · us-central1$12.00$6.00$8.40$8.40$5.40

Price is not cost per token. Equal chip counts do not imply equal throughput. Use these rates only after measuring the number of chip-hours needed to meet the same quality and SLO.

Interactive worksheet

Estimate TPU capacity spend—not workload performance.

Choose a TPU, consumption mode, chip count and hours. The result is a simple list-price capacity estimate using the representative U.S. rates above.

Estimated capacity spend
$124,416
64 chips × 720 hours × $2.70/chip-hour

This worksheet intentionally does not compare TPU to GPU. A fair cross-platform comparison requires measured workload goodput, not equal accelerator counts.

FAQ

TPU architecture questions

Is Ironwood available now?

Yes. Google Cloud publishes current TPU7x/Ironwood documentation and public per-chip pricing. TPU 8t and 8i are still described by Google as coming soon for Cloud customers.

Is Trillium only for inference?

No. Google documents Trillium/v6e for transformer, text-to-image and CNN training, fine-tuning and serving.

Why did Google split TPU 8 into 8t and 8i?

Google says the operational requirements of large-scale pre-training and high-concurrency reasoning/serving have diverged. 8t emphasizes training throughput and embeddings; 8i emphasizes on-chip SRAM, collectives latency and a serving-oriented network topology.

Can PyTorch run on modern TPUs?

Yes. PyTorch/XLA is a mature TPU path, and Google has continued expanding PyTorch support. Performance still needs workload-specific validation.