ERNIE-4.5-VL-28B-A3B TPS calculator

Open weights Baidu 28B parameters June 2025

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

241 of 818 cards that can run it

Smallest card that fits

Xeon Phi 7120P

16 GB · Q3_K_M · 51.9 tok/s

Fastest card

B200

672 tok/s · 180 GB

Which GPUs can run ERNIE-4.5-VL-28B-A3B?

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

241 cards match

Calculating

					Needs	Quantisation	Fit
672 tok/s 403–1,076 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	29.2 GB	Q8_0	Comfortable
672 tok/s 403–1,076 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	29.2 GB	Q8_0	Comfortable
537 tok/s 322–859 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	29.2 GB	Q8_0	Comfortable
537 tok/s 322–859 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	29.2 GB	Q8_0	Comfortable
429 tok/s 258–687 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	29.2 GB	Q8_0	Comfortable
411 tok/s 247–657 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	29.2 GB	Q8_0	Comfortable
411 tok/s 247–657 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	29.2 GB	Q8_0	Comfortable
393 tok/s 236–629 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	29.2 GB	Q8_0	Comfortable
349 tok/s 209–558 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	29.2 GB	Q8_0	Comfortable
349 tok/s 209–558 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	29.2 GB	Q8_0	Comfortable
349 tok/s 209–558 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	29.2 GB	Q8_0	Comfortable
331 tok/s 199–530 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	29.2 GB	Q8_0	Comfortable
282 tok/s 169–452 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	29.2 GB	Q8_0	Comfortable
282 tok/s 169–452 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	29.2 GB	Q8_0	Comfortable
282 tok/s 169–452 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	29.2 GB	Q8_0	Comfortable
282 tok/s 169–452 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	29.2 GB	Q8_0	Comfortable
282 tok/s 169–452 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	29.2 GB	Q8_0	Comfortable
256 tok/s 154–410 · low confidence	Tesla V100 SXM2 16 GB NVIDIA	16 GB	1,130 GB/s	Nov 2019	12.9 GB	Q3_K_M	Tight
228 tok/s 137–365 · low confidence	DRIVE A100 PROD NVIDIA	32 GB	1,870 GB/s	May 2020	22.7 GB	Q6_K	Comfortable
228 tok/s 137–365 · low confidence	GRID A100A NVIDIA	32 GB	1,870 GB/s	May 2020	22.7 GB	Q6_K	Comfortable
219 tok/s 131–350 · low confidence	GeForce RTX 5090 NVIDIA	32 GB	1,790 GB/s	Jan 2025	22.7 GB	Q6_K	Comfortable
219 tok/s 131–350 · low confidence	GeForce RTX 5090 D NVIDIA	32 GB	1,790 GB/s	Jan 2025	22.7 GB	Q6_K	Comfortable
218 tok/s 131–348 · low confidence	GeForce RTX 5080 NVIDIA	16 GB	960 GB/s	Jan 2025	12.9 GB	Q3_K_M	Tight
215 tok/s 129–344 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	29.2 GB	Q8_0	Comfortable
215 tok/s 129–344 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	29.2 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: Baidu
Organisation type: Industry
Country: China
Published: 29 June 2025

What it does

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

Domain: Multimodal, Language, Vision, Video
Task: Language modeling/generation, Visual question answering, Video description, Speech recognition (ASR), Quantitative reasoning, Code generation, Translation, Question answering, Character recognition (OCR)
Base model: ERNIE-4.5-21B-A3B
Numerical format: FP8

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: 28B
Training data: tokens

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: Unreleased
Hugging Face: baidu

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
Record confidence: Confident

Sources

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

Reference: ERNIE 4.5 Technical Report
Last updated: 28 November 2025

The extremes

The ten fastest GPUs for ERNIE-4.5-VL-28B-A3B

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 ERNIE-4.5-VL-28B-A3B

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

Xeon Phi 7120P

Memory needed

12.9 GB

Fastest

672 tok/s

With 28B parameters, ERNIE-4.5-VL-28B-A3B lands in the range a serious desktop card can handle once the weights are compressed. 241 of the cards we track can run it.

The least hardware that works is a Xeon Phi 7120P. Its 16 GB is enough at Q3_K_M compression, giving roughly 51.9 tokens per second.

The quickest result comes from a B200 at around 672 tokens per second — its 8,000 GB/s of bandwidth is what buys that.

Background

ERNIE-4.5-VL-28B-A3B was published by Baidu, in China, in June 2025. It comes out of industry.

It works in Multimodal, Language, Vision, Video, and is recorded as doing language modeling/generation, Visual question answering, Video description, Speech recognition (ASR), Quantitative reasoning, Code generation, Translation, Question answering, Character recognition (OCR).

Its starting point was ERNIE-4.5-21B-A3B — 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. It is published under the baidu organisation on Hugging Face.

Reading the throughput figures

Across every card that can run it, the middle of the range is about 101.9 tokens per second, and 239 of them clear the ten tokens per second that roughly matches reading speed.

This is a mixture-of-experts model, which routes each token through only part of itself. It therefore generates far faster than its total size suggests — while still needing every parameter resident in memory, so it is quick without being cheap to hold.

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.

Step by step

How to choose a GPU for ERNIE-4.5-VL-28B-A3B

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

The table lists every card that can hold ERNIE-4.5-VL-28B-A3B — around 12.9 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 ERNIE-4.5-VL-28B-A3B can slip off a card that handles short questions easily.
03

Decide how much compression you will accept

Compression is what makes ERNIE-4.5-VL-28B-A3B 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 ERNIE-4.5-VL-28B-A3B follows memory bandwidth, not core counts, which is why the B200 tops it at 672 tok/s.
05

Read the fit column last

The fit column separates cards that just manage ERNIE-4.5-VL-28B-A3B from those with room to spare. Buy for the second if the context might grow.
06

Open the card you have settled on

Every card name links to its own page, which runs the same calculation across the whole model catalogue. Worth a look before buying for ERNIE-4.5-VL-28B-A3B alone — a card is usually bought for more than one model.

Answers

ERNIE-4.5-VL-28B-A3B — common questions

Can I run ERNIE-4.5-VL-28B-A3B on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q5_K_M, using about 19.4 GB and generating roughly 201 tokens per second — a tight fit.

Is ERNIE-4.5-VL-28B-A3B open source?

Its weights are published, so ERNIE-4.5-VL-28B-A3B 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 ERNIE-4.5-VL-28B-A3B have?

ERNIE-4.5-VL-28B-A3B has 28B parameters. MoE total parameters: 28B active parameters: 3B. 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 ERNIE-4.5-VL-28B-A3B?

ERNIE-4.5-VL-28B-A3B was published by Baidu, based in China, categorised as industry.

When was ERNIE-4.5-VL-28B-A3B released?

ERNIE-4.5-VL-28B-A3B was published in June 2025.

What is ERNIE-4.5-VL-28B-A3B used for?

ERNIE-4.5-VL-28B-A3B works in Multimodal, Language, Vision, Video, and is recorded as handling language modeling/generation, Visual question answering, Video description, Speech recognition (ASR), Quantitative reasoning, Code generation, Translation, Question answering, Character recognition (OCR). These are the areas it was designed around; they describe intent rather than a hard boundary.

Where can I download ERNIE-4.5-VL-28B-A3B?

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

Can I run ERNIE-4.5-VL-28B-A3B 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 ERNIE-4.5-VL-28B-A3B is rarely worth using — the nearest miss we calculate is short by 5.4 GB. Every figure here assumes the whole model is on the card.

Would two GPUs run ERNIE-4.5-VL-28B-A3B faster?

A second card roughly doubles the memory available but not the generation rate. With 241 cards already able to run ERNIE-4.5-VL-28B-A3B alone, the case for pairing is weak.

Why does the quantisation differ between cards for ERNIE-4.5-VL-28B-A3B?

A larger card holds a more accurate copy. Across the cards that run ERNIE-4.5-VL-28B-A3B, 5 compression levels are used; the floor control above pins it to one.

How accurate are these ERNIE-4.5-VL-28B-A3B 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 403–1,076 tok/s on the B200 rather than a single number.

What GPU do I need to run ERNIE-4.5-VL-28B-A3B?

The smallest card in our catalogue that holds ERNIE-4.5-VL-28B-A3B is the Xeon Phi 7120P, with 16 GB of memory. It runs the model at Q3_K_M using about 12.9 GB, and produces roughly 51.9 tokens per second. 241 cards in total can run it.

How fast is ERNIE-4.5-VL-28B-A3B on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 672 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 239 of the cards that can run ERNIE-4.5-VL-28B-A3B clear that.

How much VRAM does ERNIE-4.5-VL-28B-A3B need?

About 12.9 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.

Can I run ERNIE-4.5-VL-28B-A3B on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q3_K_M, using about 12.9 GB and generating roughly 256 tokens per second — a tight fit.

Source

Original publication

Record last updated 28 November 2025

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