VideoMAE V2 TPS calculator

Open weights Nanjing University,Shenzhen Institute of Advanced Technology,Shanghai AI Lab 1B parameters March 2023

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 · 36.9 tok/s

Fastest card

B200

3,388 tok/s · 180 GB

Which GPUs can run VideoMAE V2?

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
3,388 tok/s 2,033–5,421 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	1.8 GB	Q8_0	Comfortable
3,388 tok/s 2,033–5,421 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	1.8 GB	Q8_0	Comfortable
2,706 tok/s 1,623–4,329 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	1.8 GB	Q8_0	Comfortable
2,706 tok/s 1,623–4,329 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	1.8 GB	Q8_0	Comfortable
2,164 tok/s 1,298–3,462 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	1.8 GB	Q8_0	Comfortable
2,071 tok/s 1,243–3,314 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	1.8 GB	Q8_0	Comfortable
2,071 tok/s 1,243–3,314 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	1.8 GB	Q8_0	Comfortable
1,982 tok/s 1,189–3,171 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	1.8 GB	Q8_0	Comfortable
1,759 tok/s 1,055–2,815 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	1.8 GB	Q8_0	Comfortable
1,759 tok/s 1,055–2,815 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	1.8 GB	Q8_0	Comfortable
1,759 tok/s 1,055–2,815 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	1.8 GB	Q8_0	Comfortable
1,669 tok/s 1,001–2,670 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	1.8 GB	Q8_0	Comfortable
1,423 tok/s 854–2,277 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	1.8 GB	Q8_0	Comfortable
1,423 tok/s 854–2,277 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	1.8 GB	Q8_0	Comfortable
1,423 tok/s 854–2,277 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	1.8 GB	Q8_0	Comfortable
1,423 tok/s 854–2,277 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	1.8 GB	Q8_0	Comfortable
1,423 tok/s 854–2,277 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	1.8 GB	Q8_0	Comfortable
1,084 tok/s 650–1,734 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	1.8 GB	Q8_0	Comfortable
1,084 tok/s 650–1,734 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	1.8 GB	Q8_0	Comfortable
903 tok/s 542–1,445 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	1.8 GB	Q8_0	Comfortable
884 tok/s 530–1,414 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	1.8 GB	Q8_0	Comfortable
864 tok/s 518–1,382 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	1.8 GB	Q8_0	Comfortable
864 tok/s 518–1,382 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	1.8 GB	Q8_0	Comfortable
864 tok/s 518–1,382 · low confidence	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	1.8 GB	Q8_0	Comfortable
864 tok/s 518–1,382 · low confidence	H100 CNX NVIDIA	80 GB	2,040 GB/s	Mar 2023	1.8 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: Nanjing University,Shenzhen Institute of Advanced Technology,Shanghai AI Lab
Organisation type: Academia,Academia
Country: China
Published: 29 March 2023
Authors: Limin Wang, Bingkun Huang, Zhiyu Zhao, Zhan Tong, Yinan He, Yi Wang, Yali Wang, Yu Qiao

What it does

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

Domain: Video
Task: Action recognition
Approach: Self-supervised learning
Base model: ViT-G/14
Numerical format: FP16

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: 1B
Training data: 1,243,350,000 tokens
Epochs: 1,200

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: 9.7 × 10²¹ 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 SXM4 80 GB
Chips used: 64
Wall-clock time: 336 hours (14 days)
Power draw: 51.0 kW
Compute cost: $18,340

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: SOTA improvement
Record confidence: Confident
Citations: 664

Sources

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

Reference: VideoMAE V2: Scaling Video Masked Autoencoders with Dual Masking
Last updated: 25 May 2026

The extremes

The ten fastest GPUs that run VideoMAE V2

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 VideoMAE V2

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 you need to run it

Minimum card

Tesla C1080

Memory needed

1.8 GB

Fastest

3,388 tok/s

VideoMAE V2 is small enough at 1B parameters that hardware is rarely the obstacle — 818 of the cards we track can run it, including cards several years old.

The least hardware that works is a Tesla C1080. Its 4 GB is enough at Q8_0 compression, giving roughly 36.9 tokens per second.

At the other end, a B200 generates roughly 3,388 tokens per second on it, on the strength of 8,000 GB/s of memory bandwidth.

Where it came from

VideoMAE V2 was published by Nanjing University,Shenzhen Institute of Advanced Technology,Shanghai AI Lab, in China, in March 2023. The organisation is categorised as academia,Academia.

It works in Video, and is recorded as doing action recognition.

Its starting point was ViT-G/14 — most models at this scale are adapted from an existing base rather than built from nothing.

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

Understanding the speeds

The median result is around 95.1 tokens per second; 806 cards produce text faster than most people read it.

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

Its internal architecture is not on file, so memory is approximated from the parameter count and marked accordingly. Expect the real figure to differ, more so at long context.

Training and provenance

Producing it required around 9.7 × 10²¹ FLOP of arithmetic, on NVIDIA A100 SXM4 80 GB, which is a statement about the training budget rather than about inference.

The training set ran to roughly 1,243,350,000 tokens.

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

Step by step

How to choose a GPU for VideoMAE V2

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

Look at what VideoMAE V2 actually needs — around 1.8 GB at Q8_0. No amount of processing power compensates for a card that cannot hold it.
02

Decide how long your conversations run

The conversation occupies memory too, and grows as it goes. Set the slider to the length you expect: at long context VideoMAE V2 can slip off a card that handles short questions easily.
03

Set a quality floor

Each card runs the least-compressed copy it can hold — Q8_0 on the smallest card that fits. Setting a floor drops the cards that only manage VideoMAE V2 by squeezing it further than you would want.
04

Rank by throughput rather than spec sheet

The speed ordering for VideoMAE V2 is effectively an ordering by memory bandwidth, which is why the B200 tops it at 3,388 tok/s.
05

Read the fit column last

Tight means VideoMAE V2 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

Check the card from the other side

Following a card through to its own page shows every other model it can hold, which is the question that follows once VideoMAE V2 is settled.

Answers

VideoMAE V2 — common questions

How accurate are these VideoMAE V2 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 2,033–5,421 tok/s on the B200 rather than a single number.

What GPU do I need to run VideoMAE V2?

The smallest card in our catalogue that holds VideoMAE V2 is the Tesla C1080, with 4 GB of memory. It runs the model at Q8_0 using about 1.8 GB, and produces roughly 36.9 tokens per second. 818 cards in total can run it.

How fast is VideoMAE V2 on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 3,388 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 806 of the cards that can run VideoMAE V2 clear that.

How much VRAM does VideoMAE V2 need?

About 1.8 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 VideoMAE V2 on a 8 GB GPU?

Yes. A CMP 170HX 8 GB with 8 GB runs it at Q8_0, using about 1.8 GB and generating roughly 631 tokens per second — a comfortable fit.

Can I run VideoMAE V2 on a 12 GB GPU?

Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q8_0, using about 1.8 GB and generating roughly 386 tokens per second — a comfortable fit.

Can I run VideoMAE V2 on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q8_0, using about 1.8 GB and generating roughly 479 tokens per second — a comfortable fit.

Can I run VideoMAE V2 on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 1.8 GB and generating roughly 568 tokens per second — a comfortable fit.

Is VideoMAE V2 open source?

Its weights are published, so VideoMAE V2 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 VideoMAE V2 have?

VideoMAE V2 has 1B parameters. 1B. 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 VideoMAE V2?

VideoMAE V2 was published by Nanjing University,Shenzhen Institute of Advanced Technology,Shanghai AI Lab, based in China, categorised as academia,Academia.

When was VideoMAE V2 released?

VideoMAE V2 was published in March 2023. 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 VideoMAE V2 used for?

VideoMAE V2 works in Video, and is recorded as handling action recognition. These are the areas it was designed around; they describe intent rather than a hard boundary.

Where can I download VideoMAE V2?

The weights for VideoMAE V2 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 VideoMAE V2?

Around 9.7 × 10²¹ FLOP, on NVIDIA A100 SXM4 80 GB. 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 VideoMAE V2 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 VideoMAE V2 is rarely worth using. Every figure here assumes the whole model is on the card.

Would two GPUs run VideoMAE V2 faster?

Two cards buy memory rather than speed. That matters for VideoMAE V2 only if one card cannot hold it — 818 can, so a second adds little.

Why does the quantisation differ between cards for VideoMAE V2?

Because capacity varies, so does how hard VideoMAE V2 has to be squeezed — 1 distinct levels appear in the table above. Set a minimum quality to compare at one.

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