Emu2 TPS calculator

Open weights Beijing Academy of Artificial Intelligence / BAAI,Tsinghua University,Peking University 37B parameters May 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

126 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Tesla M40 24 GB

24 GB · IQ4_XS · 6.9 tok/s

Fastest card

B200

91.6 tok/s · 180 GB

Which GPUs can run Emu2?

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

126 cards match

Calculating

					Needs	Quantisation	Fit
91.6 tok/s 55–147 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	40.3 GB	Q8_0	Comfortable
91.6 tok/s 55–147 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	40.3 GB	Q8_0	Comfortable
73.1 tok/s 44–117 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	40.3 GB	Q8_0	Comfortable
73.1 tok/s 44–117 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	40.3 GB	Q8_0	Comfortable
58.5 tok/s 35–94 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	40.3 GB	Q8_0	Comfortable
56.0 tok/s 34–90 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	40.3 GB	Q8_0	Comfortable
56.0 tok/s 34–90 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	40.3 GB	Q8_0	Comfortable
53.6 tok/s 32–86 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	40.3 GB	Q8_0	Comfortable
47.5 tok/s 29–76 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	40.3 GB	Q8_0	Comfortable
47.5 tok/s 29–76 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	40.3 GB	Q8_0	Comfortable
47.5 tok/s 29–76 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	40.3 GB	Q8_0	Comfortable
45.1 tok/s 27–72 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	40.3 GB	Q8_0	Comfortable
38.5 tok/s 23–62 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	40.3 GB	Q8_0	Comfortable
38.5 tok/s 23–62 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	40.3 GB	Q8_0	Comfortable
38.5 tok/s 23–62 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	40.3 GB	Q8_0	Comfortable
38.5 tok/s 23–62 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	40.3 GB	Q8_0	Comfortable
38.5 tok/s 23–62 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	40.3 GB	Q8_0	Comfortable
38.2 tok/s 23–61 · low confidence	DRIVE A100 PROD NVIDIA	32 GB	1,870 GB/s	May 2020	27.4 GB	Q5_K_M	Tight
38.2 tok/s 23–61 · low confidence	GRID A100A NVIDIA	32 GB	1,870 GB/s	May 2020	27.4 GB	Q5_K_M	Tight
37.7 tok/s 23–60 · low confidence	GeForce RTX 5090 D V2 NVIDIA	24 GB	1,340 GB/s	Aug 2025	20.9 GB	IQ4_XS	Tight
36.6 tok/s 22–59 · low confidence	GeForce RTX 5090 NVIDIA	32 GB	1,790 GB/s	Jan 2025	27.4 GB	Q5_K_M	Tight
36.6 tok/s 22–59 · low confidence	GeForce RTX 5090 D NVIDIA	32 GB	1,790 GB/s	Jan 2025	27.4 GB	Q5_K_M	Tight
34.3 tok/s 21–55 · low confidence	A30X NVIDIA	24 GB	1,220 GB/s	Apr 2021	20.9 GB	IQ4_XS	Tight
29.3 tok/s 18–47 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	40.3 GB	Q8_0	Comfortable
29.3 tok/s 18–47 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	40.3 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: Beijing Academy of Artificial Intelligence / BAAI,Tsinghua University,Peking University
Organisation type: Academia,Academia,Academia
Country: China
Published: 8 May 2024
Authors: Quan Sun, Yufeng Cui, Xiaosong Zhang, Fan Zhang, Qiying Yu, Zhengxiong Luo, Yueze Wang, Yongming Rao, Jingjing Liu, Tiejun Huang, Xinlong Wang

What it does

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

Domain: Multimodal, Language, Vision, Image generation
Task: Language modeling/generation, Question answering, Visual question answering, Image generation, Text-to-image
Base model: LLaMA-33B,Stable Diffusion XL (SDXL)
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: 37B
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 (non-commercial)
Training code: Unreleased
Hugging Face: BAAI

How it is classified

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

Record confidence: Confident

Sources

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

Reference: Generative Multimodal Models are In-Context Learners
Last updated: 28 November 2025

The extremes

The ten fastest GPUs that run Emu2

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 Emu2

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

Hardware requirements in practice

Minimum card

Tesla M40 24 GB

Memory needed

20.9 GB

Fastest

91.6 tok/s

Emu2 reaches a parameter count of 37B. That lands in the range a serious desktop card can handle once the weights are compressed. The number of cards we track that can run it: 126.

The entry point is Tesla M40 24 GB, with a memory capacity of 24 GB, running it at a compression of IQ4_XS and producing around 6.9 tokens per second.

The quickest result comes from B200, generating roughly 91.6 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

What this model is

Emu2 was published by Beijing Academy of Artificial Intelligence / BAAI,Tsinghua University,Peking University, in the country recorded as China, during May 2024. It comes out of an organisation categorised as academia,Academia,Academia.

It works in the domain of Multimodal, Language, Vision, Image generation, and is recorded as performing the task of language modeling/generation, Question answering, Visual question answering, Image generation, Text-to-image.

Its starting point was an existing base model, LLaMA-33B,Stable Diffusion XL (SDXL). That is why it shares the base model's general shape and size.

Because its weights were released, nothing about running it depends on a provider staying available — it is yours once downloaded. On Hugging Face it is published under the organisation BAAI.

What decides the speed

Half the cards that hold it manage more than 19.8 tokens per second. Exceeding reading speed outright: 94 of them.

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.

Step by step

How to choose a GPU for Emu2

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

Every card here has been checked against Emu2, needing around 20.9 GB at a compression of IQ4_XS. Capacity is the gate — a card either holds it or it does not.
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, because at long context a card that handles short questions easily can be dropped by Emu2.
03

Choose how far you will compress it

The quantisation column varies by card, because a bigger card holds a more accurate copy, reaching a compression of IQ4_XS on the smallest card that fits. Setting a minimum quality drops the cards that only manage it by squeezing further than you would want, and holds the comparison at one level.
04

Compare tokens per second, not specifications

The speed ordering is effectively an ordering by memory bandwidth, for Emu2. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B200, at 91.6 tok/s.
05

Look at the headroom, not just the fit

Tight means it loads and works with no room to raise the context later, in the case of Emu2. Comfortable means you can grow the context later. That difference matters more than a few tokens per second, so buy for comfortable if you expect to.
06

See what else that card runs

Following a card through to its own page shows every other model it can hold, which is the question that follows once you have settled on Emu2.

Answers

Emu2 — common questions

Emu2— where can I download it?

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

Emu2— can I run it if it does not fit in my GPU?

It can be split between the card and system memory, but it generates painfully slowly that way. The nearest miss we calculate falls short by 5.1 GB. Every figure here assumes the whole model is resident on the card.

Emu2— would two GPUs run it faster?

Capacity adds across cards; throughput does not. The number of cards already holding it on their own: 126. So a second card is rarely the answer here.

Emu2— why does the quantisation differ between cards?

Because capacity varies, so does how hard it has to be squeezed. The number of distinct levels in the table above: 5. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

Emu2— how accurate are these speed estimates?

They are calculated from specifications rather than measured, and each carries a range. One example: 55–147 tok/s on B200. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

Emu2— what GPU do I need to run it?

The smallest card in our catalogue that holds it is Tesla M40 24 GB, with a memory capacity of 24 GB. It runs the model at a compression of IQ4_XS using about 20.9 GB, and produces roughly 6.9 tokens per second. The number of cards able to run it in total: 126.

Emu2— how fast is it on a GPU?

It depends on the card. The quickest we calculate is B200, at about 91.6 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and the number of cards clearing that: 94.

Emu2— how much VRAM does it need?

It needs about 20.9 GB at a compression of IQ4_XS, 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.

Emu2— can I run it on a GPU holding 24 GB?

Yes. The card GeForce RTX 5090 D V2, holding 24 GB, runs it at a compression of IQ4_XS, using about 20.9 GB and generating roughly 37.7 tokens per second. The fit is tight.

Emu2— is it open source?

Its weights are published, so it 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.

Emu2— how many parameters does it have?

It has a parameter count of 37B. 37B "we leverage pretrained EVA-02-CLIP-E-plus [70], LLaMA-33B [73] and SDXL [58] to initialize the Visual Encoder, Multimodal Modeling, and Visual Decoder, respectively". 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.

Emu2— who created it?

It was published by Beijing Academy of Artificial Intelligence / BAAI,Tsinghua University,Peking University, based in China, an organisation categorised as academia,Academia,Academia.

Emu2— when was it released?

It was published in May 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.

Emu2— what is it used for?

It works in the domain of Multimodal, Language, Vision, Image generation, and is recorded as handling the task of language modeling/generation, Question answering, Visual question answering, Image generation, Text-to-image. 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.

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