Pangu-Weather TPS calculator

Open weights Huawei 256M parameters July 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

818 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Tesla C1080

4 GB · Q8_0 · 144 tok/s

Fastest card

B200

13,235 tok/s · 180 GB

Which GPUs can run Pangu-Weather?

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

818 cards match

Calculating

					Needs	Quantisation	Fit
13,235 tok/s 7,941–21,176 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	1.0 GB	Q8_0	Comfortable
13,235 tok/s 7,941–21,176 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	1.0 GB	Q8_0	Comfortable
10,569 tok/s 6,341–16,910 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	1.0 GB	Q8_0	Comfortable
10,569 tok/s 6,341–16,910 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	1.0 GB	Q8_0	Comfortable
8,452 tok/s 5,071–13,524 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	1.0 GB	Q8_0	Comfortable
8,090 tok/s 4,854–12,944 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	1.0 GB	Q8_0	Comfortable
8,090 tok/s 4,854–12,944 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	1.0 GB	Q8_0	Comfortable
7,743 tok/s 4,646–12,388 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	1.0 GB	Q8_0	Comfortable
6,872 tok/s 4,123–10,995 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	1.0 GB	Q8_0	Comfortable
6,872 tok/s 4,123–10,995 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	1.0 GB	Q8_0	Comfortable
6,872 tok/s 4,123–10,995 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	1.0 GB	Q8_0	Comfortable
6,518 tok/s 3,911–10,429 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	1.0 GB	Q8_0	Comfortable
5,559 tok/s 3,335–8,894 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	1.0 GB	Q8_0	Comfortable
5,559 tok/s 3,335–8,894 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	1.0 GB	Q8_0	Comfortable
5,559 tok/s 3,335–8,894 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	1.0 GB	Q8_0	Comfortable
5,559 tok/s 3,335–8,894 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	1.0 GB	Q8_0	Comfortable
5,559 tok/s 3,335–8,894 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	1.0 GB	Q8_0	Comfortable
4,233 tok/s 2,540–6,772 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	1.0 GB	Q8_0	Comfortable
4,233 tok/s 2,540–6,772 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	1.0 GB	Q8_0	Comfortable
3,527 tok/s 2,116–5,644 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	1.0 GB	Q8_0	Comfortable
3,452 tok/s 2,071–5,523 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	1.0 GB	Q8_0	Comfortable
3,375 tok/s 2,025–5,400 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	1.0 GB	Q8_0	Comfortable
3,375 tok/s 2,025–5,400 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	1.0 GB	Q8_0	Comfortable
3,375 tok/s 2,025–5,400 · low confidence	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	1.0 GB	Q8_0	Comfortable
3,375 tok/s 2,025–5,400 · low confidence	H100 CNX NVIDIA	80 GB	2,040 GB/s	Mar 2023	1.0 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: Huawei
Organisation type: Industry
Country: China
Published: 5 July 2023
Authors: Kaifeng Bi, Lingxi Xie, Hengheng Zhang, Xin Chen, Xiaotao Gu, Qi Tian

What it does

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

Domain: Earth science
Task: Weather forecasting

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: 256M
Training data: 24,457,821,696,000 tokens
Epochs: 100

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: 4 × 10²² FLOP
How it was established: Hardware

The training run

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

Training hardware: NVIDIA V100
Chips used: 192
Wall-clock time: 1,536 hours (64 days)
Power draw: 114.6 kW
Compute cost: $51,279

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

How it is classified

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

Why it is tracked: SOTA improvement
Record confidence: Confident
Citations: 197

Sources

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

Reference: Accurate medium-range global weather forecasting with 3D neural networks
Last updated: 28 November 2025

The extremes

The ten fastest GPUs that run Pangu-Weather

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 Pangu-Weather

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 C1080

Memory needed

1.0 GB

Fastest

13,235 tok/s

Pangu-Weather is small enough at 256M parameters that hardware is rarely the obstacle — 818 of the cards we track can run it, including cards several years old.

The entry point is the Tesla C1080: 4 GB of memory, Q8_0 compression, roughly 144 tokens per second.

Top of the range is the B200, at roughly 13,235 tokens per second thanks to 8,000 GB/s of bandwidth.

Background

Pangu-Weather was published by Huawei, in China, in July 2023. It comes out of industry.

It works in Earth science, and is recorded as doing weather forecasting.

Published weights mean the model runs on your machine rather than someone else's, which is what makes the hardware question below answerable at all.

Reading the throughput figures

The median result is around 371.7 tokens per second; 818 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 4 × 10²² FLOP of arithmetic, on NVIDIA V100, which is a statement about the training budget rather than about inference.

It was trained on about 24,457,821,696,000 tokens of text.

Its inclusion criterion is sOTA improvement.

Step by step

How to choose a GPU for Pangu-Weather

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

Look at what Pangu-Weather actually needs — around 1.0 GB at Q8_0. No amount of processing power compensates for a card that cannot hold it.
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 Pangu-Weather stops fitting a card that seemed fine.
03

Set a quality floor

The quantisation column varies by card, because a bigger card holds a more accurate copy of Pangu-Weather — Q8_0 on the smallest card that fits. Set a floor to hold the comparison at one level.
04

Compare tokens per second, not specifications

Sort by speed to see how cards rank for Pangu-Weather. It will not match a gaming ordering — generation is bound by memory bandwidth, which is why the B200 tops it at 13,235 tok/s.
05

Look at the headroom, not just the fit

Tight means Pangu-Weather loads and works, with no room to raise the context later. Comfortable means you can. The difference matters more than a few tokens per second.
06

See what else that card runs

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

Answers

Pangu-Weather — common questions

How accurate are these Pangu-Weather speed estimates?

They are calculated from specifications rather than measured, and each carries a range — 7,941–21,176 tok/s on the B200, for instance. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

What GPU do I need to run Pangu-Weather?

The smallest card in our catalogue that holds Pangu-Weather is the Tesla C1080, with 4 GB of memory. It runs the model at Q8_0 using about 1.0 GB, and produces roughly 144 tokens per second. 818 cards in total can run it.

How fast is Pangu-Weather on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 13,235 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 818 of the cards that can run Pangu-Weather clear that.

How much VRAM does Pangu-Weather need?

About 1.0 GB at Q8_0 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 Pangu-Weather on a 8 GB GPU?

Yes. A CMP 170HX 8 GB with 8 GB runs it at Q8_0, using about 1.0 GB and generating roughly 2,465 tokens per second — a comfortable fit.

Can I run Pangu-Weather on a 12 GB GPU?

Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q8_0, using about 1.0 GB and generating roughly 1,509 tokens per second — a comfortable fit.

Can I run Pangu-Weather on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q8_0, using about 1.0 GB and generating roughly 1,869 tokens per second — a comfortable fit.

Can I run Pangu-Weather on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 1.0 GB and generating roughly 2,217 tokens per second — a comfortable fit.

Is Pangu-Weather open source?

Its weights are published, so Pangu-Weather 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 Pangu-Weather have?

Pangu-Weather has 256M parameters. 4*64 million = 256M params "We trained four deep networks with lead times (the time difference between input and output) at 1 h, 3 h, 6 h and 24 h, respectively... This modification increases the number of bias parameters by a factor of 527, with each 3D deep network containing approximately 64 million 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.

Who created Pangu-Weather?

Pangu-Weather was published by Huawei, based in China, categorised as industry.

When was Pangu-Weather released?

Pangu-Weather was published in July 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.

What is Pangu-Weather used for?

Pangu-Weather works in Earth science, and is recorded as handling weather forecasting. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

Where can I download Pangu-Weather?

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

How much compute was used to train Pangu-Weather?

Around 4 × 10²² FLOP, on NVIDIA V100. 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.

Can I run Pangu-Weather 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 Pangu-Weather is rarely worth using. Every figure here assumes the whole model is on the card.

Would two GPUs run Pangu-Weather faster?

Two cards buy memory rather than speed. That matters for Pangu-Weather only if one card cannot hold it — 818 can, so a second adds little.

Why does the quantisation differ between cards for Pangu-Weather?

Each card is shown running the least-compressed copy it can hold, and Pangu-Weather appears at 1 different compression levels across the cards that fit it. Bigger cards get the more accurate version.

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