Transformer local-attention (NesT-B) TPS calculator

Open weights Google Cloud,Google Research 90.1M parameters May 2021

Each card below is assessed against this model at the context length and minimum quality you choose. Speed is an estimate for a single request, calculated from the card's memory bandwidth and the size of the model once compressed.

Calculated for this model

818 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Tesla C1080

4 GB · Q8_0 · 409 tok/s

Fastest card

B200

37,605 tok/s · 180 GB

Which GPUs can run Transformer local-attention (NesT-B)?

Set the inputs, read the answer

Context length tokens

A longer conversation needs more memory, which can push this model off smaller cards.

Minimum quality

Hides cards that would only fit the model by compressing it below this point.

Search cards

Card memory

Also show cards that cannot run it

818 cards match

Calculating

					Needs	Quantisation	Fit
37,605 tok/s 22,563–60,168 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	0.8 GB	Q8_0	Comfortable
37,605 tok/s 22,563–60,168 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	0.8 GB	Q8_0	Comfortable
30,029 tok/s 18,017–48,046 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	0.8 GB	Q8_0	Comfortable
30,029 tok/s 18,017–48,046 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	0.8 GB	Q8_0	Comfortable
24,016 tok/s 14,409–38,425 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	0.8 GB	Q8_0	Comfortable
22,986 tok/s 13,792–36,778 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	0.8 GB	Q8_0	Comfortable
22,986 tok/s 13,792–36,778 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	0.8 GB	Q8_0	Comfortable
21,999 tok/s 13,199–35,199 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	0.8 GB	Q8_0	Comfortable
19,524 tok/s 11,715–31,239 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	0.8 GB	Q8_0	Comfortable
19,524 tok/s 11,715–31,239 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	0.8 GB	Q8_0	Comfortable
19,524 tok/s 11,715–31,239 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	0.8 GB	Q8_0	Comfortable
18,521 tok/s 11,112–29,633 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	0.8 GB	Q8_0	Comfortable
15,794 tok/s 9,477–25,271 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	0.8 GB	Q8_0	Comfortable
15,794 tok/s 9,477–25,271 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	0.8 GB	Q8_0	Comfortable
15,794 tok/s 9,477–25,271 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	0.8 GB	Q8_0	Comfortable
15,794 tok/s 9,477–25,271 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	0.8 GB	Q8_0	Comfortable
15,794 tok/s 9,477–25,271 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	0.8 GB	Q8_0	Comfortable
12,026 tok/s 7,216–19,242 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	0.8 GB	Q8_0	Comfortable
12,026 tok/s 7,216–19,242 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	0.8 GB	Q8_0	Comfortable
10,022 tok/s 6,013–16,035 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	0.8 GB	Q8_0	Comfortable
9,808 tok/s 5,885–15,693 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	0.8 GB	Q8_0	Comfortable
9,589 tok/s 5,754–15,343 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	0.8 GB	Q8_0	Comfortable
9,589 tok/s 5,754–15,343 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	0.8 GB	Q8_0	Comfortable
9,589 tok/s 5,754–15,343 · low confidence	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	0.8 GB	Q8_0	Comfortable
9,589 tok/s 5,754–15,343 · low confidence	H100 CNX NVIDIA	80 GB	2,040 GB/s	Mar 2023	0.8 GB	Q8_0	Comfortable

Speeds are estimates for a single request — one conversation at a time — calculated from memory bandwidth, model size and quantisation. Real throughput varies with the inference runtime and its version. Figures published by hardware vendors measure many simultaneous requests and are much higher.

On record

Full specification

Everything on record for this model. Most of it describes how it was trained rather than how it runs — useful context for judging how much work went into it, and how it compares with models built at a different scale.

Origin

Who built this model, where, and when it was published.

Organisation: Google Cloud,Google Research
Organisation type: Industry,Industry
Country: United States of America
Published: 26 May 2021
Authors: Zizhao Zhang, Han Zhang, Long Zhao, Ting Chen, Sercan Arık, Tomas Pfister

What it does

The problem areas the model was built for. A model can carry several of each.

Domain: Vision
Task: Image classification, Image generation
Approach: Self-supervised learning

Size

How large the model is and how much data it was trained on. Parameters are the figure that decides whether it fits on a given graphics card.

Parameters: 90.1M
Training data: 1,280,000 tokens

Training compute

The arithmetic performed to train the model, measured in floating-point operations. It is a measure of what the training run cost, not of how fast the finished model answers you.

Training compute: 2.4 × 10¹⁹ FLOP
How it was established: Operation counting

Availability

Whether you can obtain the model and run it on your own hardware, which is what decides if any of the graphics-card figures on this page apply.

Weights: Open — downloadable
Model access: Open weights (unrestricted)
Training code: Open source

How it is classified

Labels the source dataset applies when tracking notable models, and how confident it is in the entry.

Why it is tracked: Highly cited
Record confidence: Confident
Citations: 5,734

Sources

Where this record came from and when it was last checked.

Reference: Nested Hierarchical Transformer: Towards Accurate, Data-Efficient and Interpretable Visual Understanding
Last updated: 11 February 2026

The extremes

The ten fastest GPUs that run Transformer local-attention (NesT-B)

Ranked by estimated tokens per second, newest card first where speeds tie. Because generation is bound by memory bandwidth, this ordering follows bandwidth rather than any gaming benchmark.

The smallest GPUs that still run Transformer local-attention (NesT-B)

The cheapest route in, by memory capacity. A tight fit runs the model but leaves nothing spare for a longer conversation.

What the numbers mean

What it takes to run this model

Minimum card

Tesla C1080

Memory needed

0.8 GB

Fastest

37,605 tok/s

Transformer local-attention (NesT-B) is small enough at 90.1M parameters that hardware is rarely the obstacle — 818 of the cards we track can run it, including cards several years old.

The smallest card that holds it is the Tesla C1080 with 4 GB, running it at Q8_0 and producing around 409 tokens per second.

Top of the range is the B200, at roughly 37,605 tokens per second thanks to 8,000 GB/s of bandwidth.

About this model

Transformer local-attention (NesT-B) was published by Google Cloud,Google Research, in United States of America, in May 2021. It comes out of industry,Industry.

It works in Vision, and is recorded as doing image classification, Image generation.

Because its weights were released, nothing about running it depends on a provider staying available — it is yours once downloaded.

How fast it runs, and why

Across every card that can run it, the middle of the range is about 1,056.0 tokens per second, and 818 of them clear the ten tokens per second that roughly matches reading speed.

Every weight participates in every token here, so bandwidth is the whole story: the ranking below is effectively a ranking of memory throughput.

Memory here is estimated from size rather than computed from the architecture, which is not recorded for this model — the numbers are indicative rather than exact.

How it was trained

Training it took roughly 2.4 × 10¹⁹ FLOP of computation — a measure of what producing the model cost, not of how fast it answers.

It was trained on about 1,280,000 tokens of text.

It is tracked in the underlying dataset for one reason in particular: highly cited.

Step by step

How to choose a GPU for Transformer local-attention (NesT-B)

The table above has already assessed every card we hold specifications for against this model. Getting to your answer takes six steps.

01

Start from the memory column

Look at what Transformer local-attention (NesT-B) actually needs — around 0.8 GB at Q8_0. No amount of processing power compensates for a card that cannot hold it.
02

Decide how long your conversations run

Longer conversations cost memory on top of what the weights need. Move the slider to your real working length before trusting any row for Transformer local-attention (NesT-B).
03

Set a quality floor

The quantisation column varies by card, because a bigger card holds a more accurate copy of Transformer local-attention (NesT-B) — Q8_0 on the smallest card that fits. Set a floor to hold the comparison at one level.
04

Sort by speed

The speed ordering for Transformer local-attention (NesT-B) is effectively an ordering by memory bandwidth, which is why the B200 tops it at 37,605 tok/s.
05

Read the fit column last

Tight means Transformer local-attention (NesT-B) loads and works, with no room to raise the context later. Comfortable means you can. The difference matters more than a few tokens per second.
06

See what else that card runs

Each card page repeats this sweep for every model we hold. It answers what else the hardware is good for, beyond Transformer local-attention (NesT-B).

Answers

Transformer local-attention (NesT-B) — common questions

How fast is Transformer local-attention (NesT-B) on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 37,605 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 818 of the cards that can run Transformer local-attention (NesT-B) clear that.

How much VRAM does Transformer local-attention (NesT-B) need?

About 0.8 GB at Q8_0 compression, which is what the smallest card that runs it uses. Less compression needs more: the figures in the memory column above are recalculated for each card, because each one holds the least-compressed version it can.

Can I run Transformer local-attention (NesT-B) on a 8 GB GPU?

Yes. A CMP 170HX 8 GB with 8 GB runs it at Q8_0, using about 0.8 GB and generating roughly 7,004 tokens per second — a comfortable fit.

Can I run Transformer local-attention (NesT-B) on a 12 GB GPU?

Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q8_0, using about 0.8 GB and generating roughly 4,289 tokens per second — a comfortable fit.

Can I run Transformer local-attention (NesT-B) on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q8_0, using about 0.8 GB and generating roughly 5,312 tokens per second — a comfortable fit.

Can I run Transformer local-attention (NesT-B) on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 0.8 GB and generating roughly 6,299 tokens per second — a comfortable fit.

Is Transformer local-attention (NesT-B) open source?

Its weights are published, so Transformer local-attention (NesT-B) can be downloaded and run on your own hardware. Note that open weights is not the same as open source in the full sense — it says nothing about the training data, the training code, or the commercial terms attached.

How many parameters does Transformer local-attention (NesT-B) have?

Transformer local-attention (NesT-B) has 90.1M parameters. Table A2, NesT-B is the largest size. That figure is the total, and it is what decides how much memory the model needs — roughly half a gigabyte per billion at the compression most people use.

Who created Transformer local-attention (NesT-B)?

Transformer local-attention (NesT-B) was published by Google Cloud,Google Research, based in United States of America, categorised as industry,Industry.

When was Transformer local-attention (NesT-B) released?

Transformer local-attention (NesT-B) was published in May 2021. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

What is Transformer local-attention (NesT-B) used for?

Transformer local-attention (NesT-B) works in Vision, and is recorded as handling image classification, Image generation. These are the areas it was designed around; they describe intent rather than a hard boundary.

Where can I download Transformer local-attention (NesT-B)?

The weights for Transformer local-attention (NesT-B) are published, though we do not hold a repository link for it. This site calculates hardware requirements rather than hosting model files.

How much compute was used to train Transformer local-attention (NesT-B)?

Around 2.4 × 10¹⁹ FLOP. That measures what producing the model cost and says nothing about how quickly it answers once trained — inference speed comes from memory bandwidth, not from the training budget.

Can I run Transformer local-attention (NesT-B) if it does not fit in my GPU?

It can be split between the card and system memory, but Transformer local-attention (NesT-B) generates painfully slowly that way. Nothing on this page assumes offloading.

Would two GPUs run Transformer local-attention (NesT-B) faster?

Capacity adds across cards; throughput does not. Since 818 of the cards we track already hold Transformer local-attention (NesT-B) on their own, a second card is rarely the answer here.

Why does the quantisation differ between cards for Transformer local-attention (NesT-B)?

Each card is shown running the least-compressed copy it can hold, and Transformer local-attention (NesT-B) appears at 1 different compression levels across the cards that fit it. Bigger cards get the more accurate version.

How accurate are these Transformer local-attention (NesT-B) speed estimates?

These are estimates with real error bars. The fastest result here, 22,563–60,168 tok/s on the B200, could reasonably land anywhere in its published range depending on which runtime you use.

What GPU do I need to run Transformer local-attention (NesT-B)?

The smallest card in our catalogue that holds Transformer local-attention (NesT-B) is the Tesla C1080, with 4 GB of memory. It runs the model at Q8_0 using about 0.8 GB, and produces roughly 409 tokens per second. 818 cards in total can run it.

Source

Original publication

Record last updated 11 February 2026

The other direction

Looking at it from the other side?

This page starts from the model. If you already own a card and want to know everything it will run, start from the hardware instead.

All AI models