Qwen-72B TPS calculator

Open weights Alibaba 72B parameters November 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

43 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Radeon Instinct MI200

64 GB · IQ4_XS · 18.5 tok/s

Fastest card

B200

47.1 tok/s · 180 GB

Which GPUs can run Qwen-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.

43 cards match

Calculating
Needs Quantisation Fit
47.1 tok/s

40–56

B200 NVIDIA 180 GB 8,000 GB/s Jan 2024 93.3 GB Q8_0 Comfortable
47.1 tok/s

40–56

B300 NVIDIA 288 GB 8,000 GB/s Sep 2025 93.3 GB Q8_0 Comfortable
37.6 tok/s

23–60 · low confidence

Radeon Instinct MI350X AMD 288 GB 8,190 GB/s Jan 2025 93.3 GB Q8_0 Comfortable
37.6 tok/s

23–60 · low confidence

Radeon Instinct MI355X AMD 288 GB 8,190 GB/s Jan 2025 93.3 GB Q8_0 Comfortable
35.3 tok/s

30–42

H100 SXM5 80 GB NVIDIA 80 GB 3,360 GB/s Oct 2022 68.1 GB Q5_K_M Tight
35.3 tok/s

30–42

H800 SXM5 NVIDIA 80 GB 3,360 GB/s Mar 2023 68.1 GB Q5_K_M Tight
33.7 tok/s

29–40

H100 NVL 94 GB NVIDIA 94 GB 3,940 GB/s Mar 2023 76.5 GB Q6_K Tight
30.1 tok/s

18–48 · low confidence

Radeon Instinct MI300 AMD 128 GB 6,550 GB/s Jan 2023 93.3 GB Q8_0 Comfortable
29.2 tok/s

25–35

H100 SXM5 64 GB NVIDIA 64 GB 2,020 GB/s Mar 2023 55.6 GB IQ4_XS Tight
28.8 tok/s

24–35

H200 NVL NVIDIA 141 GB 4,890 GB/s Nov 2024 93.3 GB Q8_0 Comfortable
28.8 tok/s

24–35

H200 SXM 141 GB NVIDIA 141 GB 4,890 GB/s Nov 2024 93.3 GB Q8_0 Comfortable
28.7 tok/s

24–34

H100 PCIe 96 GB NVIDIA 96 GB 3,360 GB/s Mar 2023 76.5 GB Q6_K Tight
28.7 tok/s

24–34

H100 SXM5 94 GB NVIDIA 94 GB 3,360 GB/s Mar 2023 76.5 GB Q6_K Tight
28.7 tok/s

24–34

H100 SXM5 96 GB NVIDIA 96 GB 3,360 GB/s Mar 2023 76.5 GB Q6_K Tight
27.5 tok/s

17–44 · low confidence

Radeon Instinct MI325X AMD 256 GB 6,000 GB/s Oct 2024 93.3 GB Q8_0 Comfortable
24.4 tok/s

15–39 · low confidence

Radeon Instinct MI300A AMD 128 GB 5,325 GB/s Dec 2023 93.3 GB Q8_0 Comfortable
24.4 tok/s

15–39 · low confidence

Radeon Instinct MI300X AMD 192 GB 5,325 GB/s Dec 2023 93.3 GB Q8_0 Comfortable
24.4 tok/s

15–39 · low confidence

Radeon Instinct MI308X AMD 192 GB 5,325 GB/s Dec 2023 93.3 GB Q8_0 Comfortable
21.4 tok/s

18–26

A100 SXM4 80 GB NVIDIA 80 GB 2,040 GB/s Nov 2020 68.1 GB Q5_K_M Tight
21.4 tok/s

18–26

A100X NVIDIA 80 GB 2,040 GB/s Jun 2021 68.1 GB Q5_K_M Tight
21.4 tok/s

18–26

A800 SXM4 80 GB NVIDIA 80 GB 2,040 GB/s Aug 2022 68.1 GB Q5_K_M Tight
21.4 tok/s

18–26

H100 CNX NVIDIA 80 GB 2,040 GB/s Mar 2023 68.1 GB Q5_K_M Tight
21.4 tok/s

18–26

H100 PCIe 80 GB NVIDIA 80 GB 2,040 GB/s Oct 2022 68.1 GB Q5_K_M Tight
21.4 tok/s

18–26

H800 PCIe 80 GB NVIDIA 80 GB 2,040 GB/s Mar 2023 68.1 GB Q5_K_M Tight
20.4 tok/s

17–24

A100 PCIe 80 GB NVIDIA 80 GB 1,940 GB/s Jun 2021 68.1 GB Q5_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
Alibaba
Organisation type
Industry
Country
China
Published
30 November 2023
Authors
Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, Binyuan Hui, Luo Ji, Mei Li, Junyang Lin, Runji Lin, Dayiheng Liu, Gao Liu, Chengqiang Lu, Keming Lu, Jianxin Ma, Rui Men, Xingzhang Ren, Xuancheng Ren, Chuanqi Tan, Sinan Tan, Jianhong Tu, Peng Wang, Shijie Wang, Wei Wang, Shengguang Wu, Benfeng Xu, Jin Xu, An Yang, Hao Yang, Jian Yang, Sh…

What it does

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

Domain
Language
Task
Chat, Code generation, Language modeling/generation, Question answering
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

72B

Training data
3,000,000,000,000 tokens

Assuming not trained for multiple epochs.

Batch size
4,000,000

Table 1 https://arxiv.org/abs/2309.16609 (this is uncertain because this table only lists sizes up to 14B. 72B was released after the paper)

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
1.3 × 10²⁴ FLOP

72 billion params, 3 trillion tokens 72b * 3T * 6 = 1.3e24

How it was established
Operation counting

The training run

What it physically took to train: which chips, how many, for how long, and what that drew from the wall.

Data centre
The paper does not mention any hardware, GPUs or any information regarding the hardware used.

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 (restricted use)
Training code
Unreleased

up to 100m active users: https://github.com/QwenLM/Qwen/blob/main/Tongyi%20Qianwen%20LICENSE%20AGREEMENT

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

SOTA on several Chinese benchmarks, with highest average rating overall for Chinese benchmarks: https://opencompass.org.cn/leaderboard-llm "significantly surpasses existing open-source models on multiple Chinese and English downstream evaluation tasks (including commonsense, reasoning, code, mathematics, etc.)" I haven't found confirmations of it being absolute SOTA on any particular benchmarks, only among similarly sized models

Record confidence
Confident

Sources

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

Last updated
28 November 2025

The extremes

The ten fastest GPUs that run Qwen-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 H800 SXM5 80 GB · 3,360 GB/s · Q5_K_M 35.3 tok/s
  6. 06 H100 SXM5 80 GB 80 GB · 3,360 GB/s · Q5_K_M 35.3 tok/s
  7. 07 H100 NVL 94 GB 94 GB · 3,940 GB/s · Q6_K 33.7 tok/s
  8. 08 Radeon Instinct MI300 128 GB · 6,550 GB/s · Q8_0 30.1 tok/s
  9. 09 H100 SXM5 64 GB 64 GB · 2,020 GB/s · IQ4_XS 29.2 tok/s
  10. 10 H200 NVL 141 GB · 4,890 GB/s · Q8_0 28.8 tok/s

The smallest GPUs that still run Qwen-72B

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

  1. 01 Jetson T4000 64 GB · needs 55.6 GB · IQ4_XS · tight 4.0 tok/s
  2. 02 H100 SXM5 64 GB 64 GB · needs 55.6 GB · IQ4_XS · tight 29.2 tok/s
  3. 03 Jetson AGX Orin 64 GB 64 GB · needs 55.6 GB · IQ4_XS · tight 3.0 tok/s
  4. 04 Radeon Instinct MI200 64 GB · needs 55.6 GB · IQ4_XS · tight 18.5 tok/s
  5. 05 Radeon Instinct MI210 64 GB · needs 55.6 GB · IQ4_XS · tight 18.5 tok/s
  6. 06 RTX PRO 5000 72 GB Blackwell 72 GB · needs 59.8 GB · Q4_K_M · tight 18.2 tok/s
  7. 07 H100 CNX 80 GB · needs 68.1 GB · Q5_K_M · tight 21.4 tok/s
  8. 08 H800 PCIe 80 GB 80 GB · needs 68.1 GB · Q5_K_M · tight 21.4 tok/s
  9. 09 H800 SXM5 80 GB · needs 68.1 GB · Q5_K_M · tight 35.3 tok/s
  10. 10 A800 PCIe 80 GB 80 GB · needs 68.1 GB · Q5_K_M · tight 20.4 tok/s

What the numbers mean

The hardware side

Minimum card

Radeon Instinct MI200

Memory needed

55.6 GB

Fastest

47.1 tok/s

Qwen-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. 43 of the cards we track can hold it.

At the low end, a Radeon Instinct MI200 handles it — 64 GB, at IQ4_XS, for about 18.5 tokens per second.

At the other end, a B200 generates roughly 47.1 tokens per second on it, on the strength of 8,000 GB/s of memory bandwidth.

Where it came from

Qwen-72B was published by Alibaba, in China, in November 2023. industry is the category the publisher falls under.

It works in Language, and is recorded as doing chat, Code generation, Language modeling/generation, Question answering.

The weights are published, so it can be downloaded and run on your own hardware indefinitely, offline, with no account attached.

Understanding the speeds

Half the cards that hold it manage more than 21.4 tokens per second, and 39 exceed reading speed outright.

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.

Because the architecture is recorded, the memory column is derived rather than estimated.

What went into building it

Producing it required around 1.3 × 10²⁴ FLOP of arithmetic, which is a statement about the training budget rather than about inference.

It was trained on about 3,000,000,000,000 tokens of text.

Its inclusion criterion is sOTA improvement.

Step by step

How to choose a GPU for Qwen-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

    Check what it needs before anything else

    Look at what Qwen-72B actually needs — around 55.6 GB at IQ4_XS. No amount of processing power compensates for a card that cannot hold it.

  2. 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 Qwen-72B stops fitting a card that seemed fine.

  3. 03

    Choose how far you will compress it

    Each card runs the least-compressed copy it can hold — IQ4_XS on the smallest card that fits. Setting a floor drops the cards that only manage Qwen-72B by squeezing it further than you would want.

  4. 04

    Compare tokens per second, not specifications

    Sort by speed to see how cards rank for Qwen-72B. It will not match a gaming ordering — generation is bound by memory bandwidth, which is why the B200 tops it at 47.1 tok/s.

  5. 05

    Read the fit column last

    The fit column separates cards that just manage Qwen-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

    Each card page repeats this sweep for every model we hold. It answers what else the hardware is good for, beyond Qwen-72B.

Answers

Qwen-72B — common questions

01

Who created Qwen-72B?

Qwen-72B was published by Alibaba, based in China, categorised as industry.

02

When was Qwen-72B released?

Qwen-72B was published in November 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.

03

What is Qwen-72B used for?

Qwen-72B works in Language, and is recorded as handling chat, Code generation, Language modeling/generation, Question answering. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

04

Where can I download Qwen-72B?

The weights for Qwen-72B are published, though we do not hold a repository link for it. This site calculates hardware requirements rather than hosting model files.

05

How much compute was used to train Qwen-72B?

Around 1.3 × 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.

06

Can I run Qwen-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 Qwen-72B is rarely worth using — the nearest miss we calculate is short by 16.6 GB. Every figure here assumes the whole model is on the card.

07

Would two GPUs run Qwen-72B faster?

A second card roughly doubles the memory available but not the generation rate. With 43 cards already able to run Qwen-72B alone, the case for pairing is weak.

08

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

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

09

How accurate are these Qwen-72B 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 40–56 tok/s on the B200 rather than a single number.

10

What GPU do I need to run Qwen-72B?

The smallest card in our catalogue that holds Qwen-72B is the Radeon Instinct MI200, with 64 GB of memory. It runs the model at IQ4_XS using about 55.6 GB, and produces roughly 18.5 tokens per second. 43 cards in total can run it.

11

How fast is Qwen-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 39 of the cards that can run Qwen-72B clear that.

12

How much VRAM does Qwen-72B need?

About 55.6 GB at IQ4_XS 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.

13

Is Qwen-72B open source?

Its weights are published, so Qwen-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.

14

How many parameters does Qwen-72B have?

Qwen-72B has 72B parameters. 72B. 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.

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