LLaVA-OV-72B TPS calculator

Open weights ByteDance,Nanyang Technological University,Chinese University of Hong Kong (CUHK),Hong Kong University of Science and Technology (HKUST) 72B parameters August 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

61 cards that can run it

818 cards we hold specifications for

Smallest card that fits

A100 PCIe 40 GB

40 GB · Q3_K_M · 24.8 tok/s

Fastest card

B200

47.1 tok/s · 180 GB

Which GPUs can run LLaVA-OV-72B?

Set the inputs, read the answer

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

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

61 cards match

Calculating
Needs Quantisation Fit
47.1 tok/s

28–75 · low confidence

B200 NVIDIA 180 GB 8,000 GB/s Jan 2024 77.8 GB Q8_0 Comfortable
47.1 tok/s

28–75 · low confidence

B300 NVIDIA 288 GB 8,000 GB/s Sep 2025 77.8 GB Q8_0 Comfortable
37.6 tok/s

23–60 · low confidence

Radeon Instinct MI350X AMD 288 GB 8,190 GB/s Jan 2025 77.8 GB Q8_0 Comfortable
37.6 tok/s

23–60 · low confidence

Radeon Instinct MI355X AMD 288 GB 8,190 GB/s Jan 2025 77.8 GB Q8_0 Comfortable
30.1 tok/s

18–48 · low confidence

Radeon Instinct MI300 AMD 128 GB 6,550 GB/s Jan 2023 77.8 GB Q8_0 Comfortable
28.8 tok/s

17–46 · low confidence

H200 NVL NVIDIA 141 GB 4,890 GB/s Nov 2024 77.8 GB Q8_0 Comfortable
28.8 tok/s

17–46 · low confidence

H200 SXM 141 GB NVIDIA 141 GB 4,890 GB/s Nov 2024 77.8 GB Q8_0 Comfortable
28.7 tok/s

17–46 · low confidence

H100 SXM5 80 GB NVIDIA 80 GB 3,360 GB/s Oct 2022 61.0 GB Q6_K Tight
28.7 tok/s

17–46 · low confidence

H800 SXM5 NVIDIA 80 GB 3,360 GB/s Mar 2023 61.0 GB Q6_K Tight
27.5 tok/s

17–44 · low confidence

Radeon Instinct MI325X AMD 256 GB 6,000 GB/s Oct 2024 77.8 GB Q8_0 Comfortable
27.0 tok/s

16–43 · low confidence

GRID A100B NVIDIA 48 GB 1,870 GB/s May 2020 40.1 GB IQ4_XS Tight
24.8 tok/s

15–40 · low confidence

A100 PCIe 40 GB NVIDIA 40 GB 1,560 GB/s Jun 2020 35.9 GB Q3_K_M Tight
24.8 tok/s

15–40 · low confidence

A100 SXM4 40 GB NVIDIA 40 GB 1,560 GB/s May 2020 35.9 GB Q3_K_M Tight
24.8 tok/s

15–40 · low confidence

A800 PCIe 40 GB NVIDIA 40 GB 1,560 GB/s Nov 2022 35.9 GB Q3_K_M Tight
24.4 tok/s

15–39 · low confidence

Radeon Instinct MI300A AMD 128 GB 5,325 GB/s Dec 2023 77.8 GB Q8_0 Comfortable
24.4 tok/s

15–39 · low confidence

Radeon Instinct MI300X AMD 192 GB 5,325 GB/s Dec 2023 77.8 GB Q8_0 Comfortable
24.4 tok/s

15–39 · low confidence

Radeon Instinct MI308X AMD 192 GB 5,325 GB/s Dec 2023 77.8 GB Q8_0 Comfortable
23.2 tok/s

14–37 · low confidence

H100 NVL 94 GB NVIDIA 94 GB 3,940 GB/s Mar 2023 77.8 GB Q8_0 Tight
21.2 tok/s

13–34 · low confidence

H100 SXM5 64 GB NVIDIA 64 GB 2,020 GB/s Mar 2023 52.6 GB Q5_K_M Tight
19.8 tok/s

12–32 · low confidence

H100 PCIe 96 GB NVIDIA 96 GB 3,360 GB/s Mar 2023 77.8 GB Q8_0 Tight
19.8 tok/s

12–32 · low confidence

H100 SXM5 94 GB NVIDIA 94 GB 3,360 GB/s Mar 2023 77.8 GB Q8_0 Tight
19.8 tok/s

12–32 · low confidence

H100 SXM5 96 GB NVIDIA 96 GB 3,360 GB/s Mar 2023 77.8 GB Q8_0 Tight
19.4 tok/s

12–31 · low confidence

RTX PRO 5000 Blackwell NVIDIA 48 GB 1,340 GB/s Mar 2025 40.1 GB IQ4_XS Tight
17.4 tok/s

10–28 · low confidence

A100 SXM4 80 GB NVIDIA 80 GB 2,040 GB/s Nov 2020 61.0 GB Q6_K Tight
17.4 tok/s

10–28 · low confidence

A100X NVIDIA 80 GB 2,040 GB/s Jun 2021 61.0 GB Q6_K Tight

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
ByteDance,Nanyang Technological University,Chinese University of Hong Kong (CUHK),Hong Kong University of Science and Technology (HKUST)
Organisation type
Industry,Academia,Academia,Academia
Country
China, Singapore, Hong Kong
Published
6 August 2024
Authors
Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Peiyuan Zhang, Yanwei Li, Ziwei Liu, Chunyuan Li

What it does

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

Domain
Multimodal, Vision, Language, Video
Task
Image captioning, Visual question answering, Video description, Object recognition, Action recognition, Language modeling/generation
Base model
Qwen2-72B
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
72B
Training data
38,314,782,000 tokens

Stage 1: 558000*729=406782000 Stage 1.5: 4M*729*5=14580000000 Stage 2-single: 3.2M*729*5=11664000000 Stage 2-one vision: 1.6M*729*10=11664000000 Total: 38314782000

Epochs
1
Batch size
256

We use a global batch size of 512 for the 0.5B model, and 256 for the 7B and 72B models.

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
3 × 10²⁴ FLOP

FineTune: 6*72000000000*38314782000=1.655199e+22 Base model: 3.02e+24 Total: 3.036552e+24

How it was established
Operation counting
Fine-tuning compute
1.7 × 10²² FLOP

6 FLOP/parameter/token * 72000000000 parameters * 38314782000 tokens * 1 epochs = 1.6551985824e+22 FLOP

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

Apache 2 https://huggingface.co/lmms-lab/llava-onevision-qwen2-72b-ov-sft https://github.com/LLaVA-VL/LLaVA-NeXT

Hugging Face
lmms-lab

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

As shown in Table 4, LLaVA-OneVision (SI) consistently outperforms existing multi-image LMMs in all benchmarks. After additional tuning on multi-image and video data, LLaVA-OneVision shows a marked improvement over GPT-4V in specific areas, with significant margins. This highlights its strong performance in complex tasks such as multi-image reasoning, identifying differences, and understanding 3D environments

Record confidence
Confident
Citations
2,428

Sources

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

Reference
LLaVA-OneVision: Easy Visual Task Transfer
Last updated
25 May 2026

The extremes

The ten fastest GPUs that run LLaVA-OV-72B

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.

  1. 01 B300 288 GB · 8,000 GB/s · Q8_0 47.1 tok/s
  2. 02 B200 180 GB · 8,000 GB/s · Q8_0 47.1 tok/s
  3. 03 Radeon Instinct MI350X 288 GB · 8,190 GB/s · Q8_0 37.6 tok/s
  4. 04 Radeon Instinct MI355X 288 GB · 8,190 GB/s · Q8_0 37.6 tok/s
  5. 05 Radeon Instinct MI300 128 GB · 6,550 GB/s · Q8_0 30.1 tok/s
  6. 06 H200 NVL 141 GB · 4,890 GB/s · Q8_0 28.8 tok/s
  7. 07 H200 SXM 141 GB 141 GB · 4,890 GB/s · Q8_0 28.8 tok/s
  8. 08 H800 SXM5 80 GB · 3,360 GB/s · Q6_K 28.7 tok/s
  9. 09 H100 SXM5 80 GB 80 GB · 3,360 GB/s · Q6_K 28.7 tok/s
  10. 10 Radeon Instinct MI325X 256 GB · 6,000 GB/s · Q8_0 27.5 tok/s

The smallest GPUs that still run LLaVA-OV-72B

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

  1. 01 A800 PCIe 40 GB 40 GB · needs 35.9 GB · Q3_K_M · tight 24.8 tok/s
  2. 02 A100 PCIe 40 GB 40 GB · needs 35.9 GB · Q3_K_M · tight 24.8 tok/s
  3. 03 A100 SXM4 40 GB 40 GB · needs 35.9 GB · Q3_K_M · tight 24.8 tok/s
  4. 04 Radeon PRO W7900D 48 GB · needs 40.1 GB · IQ4_XS · tight 9.7 tok/s
  5. 05 RTX PRO 5000 Blackwell 48 GB · needs 40.1 GB · IQ4_XS · tight 19.4 tok/s
  6. 06 RTX 5880 Ada Generation 48 GB · needs 40.1 GB · IQ4_XS · tight 12.5 tok/s
  7. 07 L20 48 GB · needs 40.1 GB · IQ4_XS · tight 12.5 tok/s
  8. 08 Radeon PRO W7800 48 GB 48 GB · needs 40.1 GB · IQ4_XS · tight 9.7 tok/s
  9. 09 Radeon PRO W7900 48 GB · needs 40.1 GB · IQ4_XS · tight 9.7 tok/s
  10. 10 Data Center GPU Max 1100 48 GB · needs 40.1 GB · IQ4_XS · tight 11.6 tok/s

What the numbers mean

What you need to run it

Minimum card

A100 PCIe 40 GB

Memory needed

35.9 GB

Fastest

47.1 tok/s

LLaVA-OV-72B sits at 72B parameters, which puts it above consumer hardware and into the range where a card is bought for this purpose rather than repurposed for it. 61 of the cards we track can hold it.

The least hardware that works is a A100 PCIe 40 GB. Its 40 GB is enough at Q3_K_M compression, giving roughly 24.8 tokens per second.

Top of the range is the B200, at roughly 47.1 tokens per second thanks to 8,000 GB/s of bandwidth.

Background

LLaVA-OV-72B was published by ByteDance,Nanyang Technological University,Chinese University of Hong Kong (CUHK),Hong Kong University of Science and Technology (HKUST), in China, in August 2024. industry,Academia,Academia,Academia is the category the publisher falls under.

It works in Multimodal, Vision, Language, Video, and is recorded as doing image captioning, Visual question answering, Video description, Object recognition, Action recognition, Language modeling/generation.

It builds on Qwen2-72B, which is why it shares that model's general shape and size.

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. It is published under the lmms-lab organisation on Hugging Face.

Reading the throughput figures

The median result is around 16.6 tokens per second; 51 cards produce text faster than most people read it.

It is a dense model, so every parameter is read for every token produced. That makes speed track memory bandwidth almost exactly — a card with twice the bandwidth generates roughly twice as fast.

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.

What went into building it

Producing it required around 3 × 10²⁴ FLOP of arithmetic, which is a statement about the training budget rather than about inference.

The training set ran to roughly 38,314,782,000 tokens.

The reason it appears in this catalogue at all is sOTA improvement.

Step by step

How to choose a GPU for LLaVA-OV-72B

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

  1. 01

    Read the memory figure first

    The table lists every card that can hold LLaVA-OV-72B — around 35.9 GB at Q3_K_M. That figure, not the card's headline performance, is what decides whether it runs.

  2. 02

    Set the context length you will work at

    Set the context to what you will actually use. The cache grows with the conversation, and it is the usual reason LLaVA-OV-72B stops fitting a card that seemed fine.

  3. 03

    Decide how much compression you will accept

    The quantisation column varies by card, because a bigger card holds a more accurate copy of LLaVA-OV-72B — Q3_K_M on the smallest card that fits. Set a floor to hold the comparison at one level.

  4. 04

    Sort by speed

    Ranking by tokens per second for LLaVA-OV-72B follows memory bandwidth, not core counts, which is why the B200 tops it at 47.1 tok/s.

  5. 05

    Look at the headroom, not just the fit

    The fit column separates cards that just manage LLaVA-OV-72B from those with room to spare. Buy for the second if the context might grow.

  6. 06

    Check the card from the other side

    Every card name links to its own page, which runs the same calculation across the whole model catalogue. Worth a look before buying for LLaVA-OV-72B alone — a card is usually bought for more than one model.

Answers

LLaVA-OV-72B — common questions

01

Is LLaVA-OV-72B open source?

Its weights are published, so LLaVA-OV-72B 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.

02

How many parameters does LLaVA-OV-72B have?

LLaVA-OV-72B has 72B 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.

03

Who created LLaVA-OV-72B?

LLaVA-OV-72B was published by ByteDance,Nanyang Technological University,Chinese University of Hong Kong (CUHK),Hong Kong University of Science and Technology (HKUST), based in China, categorised as industry,Academia,Academia,Academia.

04

When was LLaVA-OV-72B released?

LLaVA-OV-72B was published in August 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.

05

What is LLaVA-OV-72B used for?

LLaVA-OV-72B works in Multimodal, Vision, Language, Video, and is recorded as handling image captioning, Visual question answering, Video description, Object recognition, Action recognition, Language modeling/generation. These are the areas it was designed around; they describe intent rather than a hard boundary.

06

Where can I download LLaVA-OV-72B?

Its weights are published under the lmms-lab organisation on Hugging Face. We do not host model files — this site calculates what hardware is needed to run them.

07

How much compute was used to train LLaVA-OV-72B?

Around 3 × 10²⁴ 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.

08

Can I run LLaVA-OV-72B 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 LLaVA-OV-72B is rarely worth using — the nearest miss we calculate is short by 15.5 GB. Every figure here assumes the whole model is on the card.

09

Would two GPUs run LLaVA-OV-72B faster?

Capacity adds across cards; throughput does not. Since 61 of the cards we track already hold LLaVA-OV-72B on their own, a second card is rarely the answer here.

10

Why does the quantisation differ between cards for LLaVA-OV-72B?

Each card is shown running the least-compressed copy it can hold, and LLaVA-OV-72B appears at 5 different compression levels across the cards that fit it. Bigger cards get the more accurate version.

11

How accurate are these LLaVA-OV-72B speed estimates?

These are estimates with real error bars. The fastest result here, 28–75 tok/s on the B200, could reasonably land anywhere in its published range depending on which runtime you use.

12

What GPU do I need to run LLaVA-OV-72B?

The smallest card in our catalogue that holds LLaVA-OV-72B is the A100 PCIe 40 GB, with 40 GB of memory. It runs the model at Q3_K_M using about 35.9 GB, and produces roughly 24.8 tokens per second. 61 cards in total can run it.

13

How fast is LLaVA-OV-72B on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 47.1 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 51 of the cards that can run LLaVA-OV-72B clear that.

14

How much VRAM does LLaVA-OV-72B need?

About 35.9 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.

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