InternVL2_5-38B TPS calculator

Open weights Shanghai AI Lab,SenseTime,Tsinghua University,Nanjing University,Fudan University,Chinese University of Hong Kong (CUHK),Shanghai Jiao Tong University 38.4B parameters December 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 · Q3_K_M · 7.3 tok/s

Fastest card

B200

88.2 tok/s · 180 GB

Which GPUs can run InternVL2_5-38B?

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
88.2 tok/s 53–141 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	41.8 GB	Q8_0	Comfortable
88.2 tok/s 53–141 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	41.8 GB	Q8_0	Comfortable
70.5 tok/s 42–113 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	41.8 GB	Q8_0	Comfortable
70.5 tok/s 42–113 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	41.8 GB	Q8_0	Comfortable
56.4 tok/s 34–90 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	41.8 GB	Q8_0	Comfortable
53.9 tok/s 32–86 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	41.8 GB	Q8_0	Comfortable
53.9 tok/s 32–86 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	41.8 GB	Q8_0	Comfortable
51.6 tok/s 31–83 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	41.8 GB	Q8_0	Comfortable
45.8 tok/s 27–73 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	41.8 GB	Q8_0	Comfortable
45.8 tok/s 27–73 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	41.8 GB	Q8_0	Comfortable
45.8 tok/s 27–73 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	41.8 GB	Q8_0	Comfortable
43.5 tok/s 26–70 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	41.8 GB	Q8_0	Comfortable
39.9 tok/s 24–64 · low confidence	GeForce RTX 5090 D V2 NVIDIA	24 GB	1,340 GB/s	Aug 2025	19.5 GB	Q3_K_M	Tight
37.1 tok/s 22–59 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	41.8 GB	Q8_0	Comfortable
37.1 tok/s 22–59 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	41.8 GB	Q8_0	Comfortable
37.1 tok/s 22–59 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	41.8 GB	Q8_0	Comfortable
37.1 tok/s 22–59 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	41.8 GB	Q8_0	Comfortable
37.1 tok/s 22–59 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	41.8 GB	Q8_0	Comfortable
36.8 tok/s 22–59 · low confidence	DRIVE A100 PROD NVIDIA	32 GB	1,870 GB/s	May 2020	28.4 GB	Q5_K_M	Tight
36.8 tok/s 22–59 · low confidence	GRID A100A NVIDIA	32 GB	1,870 GB/s	May 2020	28.4 GB	Q5_K_M	Tight
36.3 tok/s 22–58 · low confidence	A30X NVIDIA	24 GB	1,220 GB/s	Apr 2021	19.5 GB	Q3_K_M	Tight
35.3 tok/s 21–56 · low confidence	GeForce RTX 5090 NVIDIA	32 GB	1,790 GB/s	Jan 2025	28.4 GB	Q5_K_M	Tight
35.3 tok/s 21–56 · low confidence	GeForce RTX 5090 D NVIDIA	32 GB	1,790 GB/s	Jan 2025	28.4 GB	Q5_K_M	Tight
30.1 tok/s 18–48 · low confidence	GeForce RTX 3090 Ti NVIDIA	24 GB	1,010 GB/s	Jan 2022	19.5 GB	Q3_K_M	Tight
30.1 tok/s 18–48 · low confidence	GeForce RTX 4090 NVIDIA	24 GB	1,010 GB/s	Sep 2022	19.5 GB	Q3_K_M	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: Shanghai AI Lab,SenseTime,Tsinghua University,Nanjing University,Fudan University,Chinese University of Hong Kong (CUHK),Shanghai Jiao Tong University
Organisation type: Academia,Industry,Academia,Academia,Academia,Academia,Academia
Country: China, Hong Kong
Published: 6 December 2024
Authors: Zhe Chen, Weiyun Wang, Yue Cao, Yangzhou Liu, Zhangwei Gao, Erfei Cui, Jinguo Zhu, Shenglong Ye, Hao Tian, Zhaoyang Liu, Lixin Gu, Xuehui Wang, Qingyun Li, Yimin Ren, Zixuan Chen, Jiapeng Luo, Jiahao Wang, Tan Jiang, Bo Wang, Conghui He, Botian Shi, Xingcheng Zhang, Han Lv, Yi Wang, Wenqi Shao, Pei Chu, Zhongying Tu, Tong He, Zhiyong Wu, Huipeng Deng, Jiaye Ge, Kai Chen, Kaipeng Zhang, Limin Wang,…

What it does

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

Domain: Multimodal, Language, Vision
Task: Visual question answering, Language modeling/generation
Base model: Qwen2.5 Instruct (32B),InternViT-6B

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: 38.4B
Training data: 151,000,000,000 tokens
Epochs: 1

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.

How it was established: Operation counting
Fine-tuning compute: 1 × 10²² 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
Hugging Face: OpenGVLab

How it is classified

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

Record confidence: Confident
Citations: 1,468

Sources

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

Reference: Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling
Last updated: 25 May 2026

The extremes

The ten fastest GPUs that run InternVL2_5-38B

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 InternVL2_5-38B

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 it takes to run this model

Minimum card

Tesla M40 24 GB

Memory needed

19.5 GB

Fastest

88.2 tok/s

InternVL2_5-38B reaches a parameter count of 38.4B. 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 smallest card that holds it is Tesla M40 24 GB, with a memory capacity of 24 GB, running it at a compression of Q3_K_M and producing around 7.3 tokens per second.

The quickest result comes from B200, generating roughly 88.2 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

About this model

InternVL2_5-38B was published by Shanghai AI Lab,SenseTime,Tsinghua University,Nanjing University,Fudan University,Chinese University of Hong Kong (CUHK),Shanghai Jiao Tong University, in the country recorded as China, during December 2024. The category the publisher falls under is academia,Industry,Academia,Academia,Academia,Academia,Academia.

It works in the domain of Multimodal, Language, Vision, and is recorded as performing the task of visual question answering, Language modeling/generation.

Rather than being trained from scratch, it is derived from Qwen2.5 Instruct (32B),InternViT-6B. That is the usual way a specialised model is produced.

The weights are published, so it can be downloaded and run on your own hardware indefinitely, offline, with no account attached. On Hugging Face it is published under the organisation OpenGVLab.

How fast it runs, and why

Across every card that can run it, the middle of the range sits at 19.7 tokens per second. Clearing the ten tokens per second that roughly matches reading speed: 92 of them.

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

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

It was trained on a corpus of about 151,000,000,000 tokens of text.

Step by step

How to choose a GPU for InternVL2_5-38B

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

01

Read the memory figure first

Start from what it actually needs, which is the requirement of InternVL2_5-38B, needing around 19.5 GB at a compression of Q3_K_M. That figure, not the headline performance of a card, is what decides whether it runs.
02

Decide how long your conversations run

Set the context to what you will actually use. The cache grows with the conversation, and it is the usual reason a card that seemed fine stops fitting InternVL2_5-38B.
03

Choose how far you will compress it

Each card runs the least-compressed copy it can hold, reaching a compression of Q3_K_M 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

Rank by throughput rather than spec sheet

Ranking by tokens per second follows memory bandwidth rather than core counts, for InternVL2_5-38B. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B200, at 88.2 tok/s.
05

Look at the headroom, not just the fit

A tight fit runs, but leaves nothing spare for a longer conversation, in the case of InternVL2_5-38B. 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

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. A card is usually bought for more than one model, so it is worth a look before buying for InternVL2_5-38B.

Answers

InternVL2_5-38B — common questions

InternVL2_5-38B— how fast is it on a GPU?

It depends on the card. The quickest we calculate is B200, at about 88.2 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: 92.

InternVL2_5-38B— how much VRAM does it need?

It needs about 19.5 GB at a compression of Q3_K_M, 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.

InternVL2_5-38B— 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 Q3_K_M, using about 19.5 GB and generating roughly 39.9 tokens per second. The fit is tight.

InternVL2_5-38B— 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.

InternVL2_5-38B— how many parameters does it have?

It has a parameter count of 38.4B. 38.4B. 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.

InternVL2_5-38B— who created it?

It was published by Shanghai AI Lab,SenseTime,Tsinghua University,Nanjing University,Fudan University,Chinese University of Hong Kong (CUHK),Shanghai Jiao Tong University, based in China, an organisation categorised as academia,Industry,Academia,Academia,Academia,Academia,Academia.

InternVL2_5-38B— when was it released?

It was published in December 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.

InternVL2_5-38B— what is it used for?

It works in the domain of Multimodal, Language, Vision, and is recorded as handling the task of visual question answering, Language modeling/generation. These are the areas it was designed around; they describe intent rather than a hard boundary.

InternVL2_5-38B— where can I download it?

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

InternVL2_5-38B— 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 5.9 GB. Every figure here assumes the whole model is resident on the card.

InternVL2_5-38B— would two GPUs run it faster?

Two cards buy memory rather than speed, which matters only if one card cannot hold it. The number that can: 126. So a second card is rarely the answer here.

InternVL2_5-38B— why does the quantisation differ between cards?

Each card is shown running the least-compressed copy it can hold. The number of distinct compression levels across the cards that fit it: 5. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

InternVL2_5-38B— how accurate are these speed estimates?

Every figure is derived from memory bandwidth and model size, not benchmarked, which is why each is published as a range rather than a single number. One example: 53–141 tok/s on B200. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

InternVL2_5-38B— 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 Q3_K_M using about 19.5 GB, and produces roughly 7.3 tokens per second. The number of cards able to run it in total: 126.

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