OpenVLA TPS calculator

Open weights Stanford University,University of California (UC) Berkeley,Toyota Research Institute,Google DeepMind,Massachusetts Institute of Technology (MIT),Physical Intelligence 7.2B parameters June 2024

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

589 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Tesla K20c

5 GB · Q3_K_M · 28.1 tok/s

Fastest card

B200

471 tok/s · 180 GB

Which GPUs can run OpenVLA?

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

589 cards match

Calculating

					Needs	Quantisation	Fit
471 tok/s 283–754 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	8.4 GB	Q8_0	Comfortable
471 tok/s 283–754 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	8.4 GB	Q8_0	Comfortable
376 tok/s 226–602 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	8.4 GB	Q8_0	Comfortable
376 tok/s 226–602 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	8.4 GB	Q8_0	Comfortable
301 tok/s 181–482 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	8.4 GB	Q8_0	Comfortable
288 tok/s 173–461 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	8.4 GB	Q8_0	Comfortable
288 tok/s 173–461 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	8.4 GB	Q8_0	Comfortable
276 tok/s 165–441 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	8.4 GB	Q8_0	Comfortable
245 tok/s 147–392 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	8.4 GB	Q8_0	Comfortable
245 tok/s 147–392 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	8.4 GB	Q8_0	Comfortable
245 tok/s 147–392 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	8.4 GB	Q8_0	Comfortable
232 tok/s 139–371 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	8.4 GB	Q8_0	Comfortable
198 tok/s 119–317 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	8.4 GB	Q8_0	Comfortable
198 tok/s 119–317 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	8.4 GB	Q8_0	Comfortable
198 tok/s 119–317 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	8.4 GB	Q8_0	Comfortable
198 tok/s 119–317 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	8.4 GB	Q8_0	Comfortable
198 tok/s 119–317 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	8.4 GB	Q8_0	Comfortable
151 tok/s 90–241 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	8.4 GB	Q8_0	Comfortable
151 tok/s 90–241 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	8.4 GB	Q8_0	Comfortable
128 tok/s 77–204 · low confidence	CMP 170HX 8 GB NVIDIA	8 GB	1,490 GB/s	Sep 2021	6.7 GB	Q6_K	Tight
126 tok/s 75–201 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	8.4 GB	Q8_0	Comfortable
123 tok/s 74–197 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	8.4 GB	Q8_0	Comfortable
120 tok/s 72–192 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	8.4 GB	Q8_0	Comfortable
120 tok/s 72–192 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	8.4 GB	Q8_0	Comfortable
120 tok/s 72–192 · low confidence	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	8.4 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: Stanford University,University of California (UC) Berkeley,Toyota Research Institute,Google DeepMind,Massachusetts Institute of Technology (MIT),Physical Intelligence
Organisation type: Academia,Academia,Industry,Industry,Academia,Industry
Country: United States of America
Published: 13 June 2024
Authors: Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, Ted Xiao, Ashwin Balakrishna, Suraj Nair, Rafael Rafailov, Ethan Foster, Grace Lam, Pannag Sanketi, Quan Vuong, Thomas Kollar, Benjamin Burchfiel, Russ Tedrake, Dorsa Sadigh, Sergey Levine, Percy Liang, Chelsea Finn

What it does

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

Domain: Robotics, Vision, Language
Task: Robotic manipulation
Approach: Supervised
Base model: Llama 2-7B
Numerical format: BF16

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: 7.2B
Training data: tokens
Epochs: 27
Batch size: 2,048

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: 1.1 × 10²³ FLOP
How it was established: Hardware
Fine-tuning compute: 9.7 × 10²¹ FLOP

The training run

What it physically took to train: which chips, how many, for how long, and what that drew from the wall.

Training hardware: NVIDIA A100
Chips used: 64
Wall-clock time: 336 hours (14 days)
Power draw: 50.5 kW

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.

Likely above 10²³ FLOP: Yes
Why it is tracked: SOTA improvement
Record confidence: Confident
Citations: 2,123

Sources

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

Reference: OpenVLA: An Open-Source Vision-Language-Action Mode
Last updated: 25 May 2026

The extremes

The ten fastest GPUs that run OpenVLA

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 OpenVLA

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

The hardware side

Minimum card

Tesla K20c

Memory needed

4.2 GB

Fastest

471 tok/s

OpenVLA is small enough at 7.2B parameters that hardware is rarely the obstacle — 589 of the cards we track can run it, including cards several years old.

At the low end, a Tesla K20c handles it — 5 GB, at Q3_K_M, for about 28.1 tokens per second.

Top of the range is the B200, at roughly 471 tokens per second thanks to 8,000 GB/s of bandwidth.

What this model is

OpenVLA was published by Stanford University,University of California (UC) Berkeley,Toyota Research Institute,Google DeepMind,Massachusetts Institute of Technology (MIT),Physical Intelligence, in United States of America, in June 2024. It comes out of academia,Academia,Industry,Industry,Academia,Industry.

It works in Robotics, Vision, Language, and is recorded as doing robotic manipulation.

Its starting point was Llama 2-7B — most models at this scale are adapted from an existing base rather than built from nothing.

The weights being open is what puts this page in the calculator rather than only in the catalogue: it is a model you can actually hold.

What decides the speed

Across every card that can run it, the middle of the range is about 25.5 tokens per second, and 559 of them clear the ten tokens per second that roughly matches reading speed.

Being dense, it reads all of itself per token, which is why the ordering by speed below follows the ordering by memory bandwidth so closely.

Without the attention layout on record, the memory column is an approximation. It is close enough to choose hardware by, and least reliable at long context.

How it was trained

Producing it required around 1.1 × 10²³ FLOP of arithmetic, on NVIDIA A100, which is a statement about the training budget rather than about inference.

It is tracked in the underlying dataset for one reason in particular: sOTA improvement.

Step by step

How to choose a GPU for OpenVLA

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

01

Check what it needs before anything else

The table lists every card that can hold OpenVLA — around 4.2 GB at Q3_K_M. That figure, not the card's headline performance, is what decides whether it runs.
02

Decide how long your conversations run

The conversation occupies memory too, and grows as it goes. Set the slider to the length you expect: at long context OpenVLA can slip off a card that handles short questions easily.
03

Set a quality floor

Compression is what makes OpenVLA fit smaller cards, at some cost in accuracy — Q3_K_M on the smallest card that fits. A minimum quality removes the ones that go too far.
04

Rank by throughput rather than spec sheet

Ranking by tokens per second for OpenVLA follows memory bandwidth, not core counts, which is why the B200 tops it at 471 tok/s.
05

Look at the headroom, not just the fit

Tight means OpenVLA 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

Open the card you have settled on

Following a card through to its own page shows every other model it can hold, which is the question that follows once OpenVLA is settled.

Answers

OpenVLA — common questions

Why does the quantisation differ between cards for OpenVLA?

A larger card holds a more accurate copy. Across the cards that run OpenVLA, 4 compression levels are used; the floor control above pins it to one.

How accurate are these OpenVLA speed estimates?

Every figure is derived from memory bandwidth and model size, not benchmarked. That is why each is published as a range such as 283–754 tok/s on the B200 rather than a single number.

What GPU do I need to run OpenVLA?

The smallest card in our catalogue that holds OpenVLA is the Tesla K20c, with 5 GB of memory. It runs the model at Q3_K_M using about 4.2 GB, and produces roughly 28.1 tokens per second. 589 cards in total can run it.

How fast is OpenVLA on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 471 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 559 of the cards that can run OpenVLA clear that.

How much VRAM does OpenVLA need?

About 4.2 GB at Q3_K_M 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 OpenVLA on a 8 GB GPU?

Yes. A CMP 170HX 8 GB with 8 GB runs it at Q6_K, using about 6.7 GB and generating roughly 128 tokens per second — a tight fit.

Can I run OpenVLA on a 12 GB GPU?

Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q8_0, using about 8.4 GB and generating roughly 53.8 tokens per second — a comfortable fit.

Can I run OpenVLA on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q8_0, using about 8.4 GB and generating roughly 66.6 tokens per second — a comfortable fit.

Can I run OpenVLA on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 8.4 GB and generating roughly 79.0 tokens per second — a comfortable fit.

Is OpenVLA open source?

Its weights are published, so OpenVLA 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 OpenVLA have?

OpenVLA has 7.2B parameters. Based on a Prismatic-7B VLM backbone, which itself is comprised of 600M parameter vision encoder (DinoV2 + SigLIP) plus Llama-2 7B. Table 1 indicates 7.1881 billion trainable parameters. 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 OpenVLA?

OpenVLA was published by Stanford University,University of California (UC) Berkeley,Toyota Research Institute,Google DeepMind,Massachusetts Institute of Technology (MIT),Physical Intelligence, based in United States of America, categorised as academia,Academia,Industry,Industry,Academia,Industry.

When was OpenVLA released?

OpenVLA was published in June 2024. 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 OpenVLA used for?

OpenVLA works in Robotics, Vision, Language, and is recorded as handling robotic manipulation. A model can carry several of each, so these are the areas it was built for rather than a limit on what it will attempt.

Where can I download OpenVLA?

The weights for OpenVLA 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 OpenVLA?

Around 1.1 × 10²³ FLOP, on NVIDIA A100. 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 OpenVLA if it does not fit in my GPU?

Partly. Layers that do not fit sit in system memory and run at a fraction of the speed, so a mostly-offloaded OpenVLA is rarely worth using — the nearest miss we calculate is short by 1.4 GB. Every figure here assumes the whole model is on the card.

Would two GPUs run OpenVLA faster?

Capacity adds across cards; throughput does not. Since 589 of the cards we track already hold OpenVLA on their own, a second card is rarely the answer here.

Source

Original publication

Record last updated 25 May 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