Zephyr 141B-A39B TPS calculator

Open weights Hugging Face,Korea Advanced Institute of Science and Technology (KAIST),Argilla 141B parameters April 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

37 cards that can run it

818 cards we hold specifications for

Smallest card that fits

A100 SXM4 80 GB

80 GB · Q3_K_M · 59.8 tok/s

Fastest card

H100 NVL 94 GB

105 tok/s · 94 GB

Which GPUs can run Zephyr 141B-A39B?

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

37 cards match

Calculating

					Needs	Quantisation	Fit
105 tok/s 63–168 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	76.8 GB	IQ4_XS	Tight
99.1 tok/s 59–159 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	101.5 GB	Q5_K_M	Tight
98.5 tok/s 59–158 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	68.6 GB	Q3_K_M	Tight
98.5 tok/s 59–158 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	68.6 GB	Q3_K_M	Tight
89.6 tok/s 54–143 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	76.8 GB	IQ4_XS	Tight
86.9 tok/s 52–139 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	150.7 GB	Q8_0	Tight
86.9 tok/s 52–139 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	150.7 GB	Q8_0	Comfortable
84.2 tok/s 51–135 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	85.0 GB	Q4_K_M	Tight
84.2 tok/s 51–135 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	85.0 GB	Q4_K_M	Tight
80.6 tok/s 48–129 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	101.5 GB	Q5_K_M	Tight
77.2 tok/s 46–123 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	117.9 GB	Q6_K	Tight
77.2 tok/s 46–123 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	117.9 GB	Q6_K	Tight
69.4 tok/s 42–111 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	150.7 GB	Q8_0	Comfortable
69.4 tok/s 42–111 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	150.7 GB	Q8_0	Comfortable
59.8 tok/s 36–96 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	68.6 GB	Q3_K_M	Tight
59.8 tok/s 36–96 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	68.6 GB	Q3_K_M	Tight
59.8 tok/s 36–96 · low confidence	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	68.6 GB	Q3_K_M	Tight
59.8 tok/s 36–96 · low confidence	H100 CNX NVIDIA	80 GB	2,040 GB/s	Mar 2023	68.6 GB	Q3_K_M	Tight
59.8 tok/s 36–96 · low confidence	H100 PCIe 80 GB NVIDIA	80 GB	2,040 GB/s	Oct 2022	68.6 GB	Q3_K_M	Tight
59.8 tok/s 36–96 · low confidence	H800 PCIe 80 GB NVIDIA	80 GB	2,040 GB/s	Mar 2023	68.6 GB	Q3_K_M	Tight
56.9 tok/s 34–91 · low confidence	A100 PCIe 80 GB NVIDIA	80 GB	1,940 GB/s	Jun 2021	68.6 GB	Q3_K_M	Tight
56.9 tok/s 34–91 · low confidence	A800 PCIe 80 GB NVIDIA	80 GB	1,940 GB/s	Nov 2022	68.6 GB	Q3_K_M	Tight
50.8 tok/s 30–81 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	150.7 GB	Q8_0	Comfortable
49.6 tok/s 30–79 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	101.5 GB	Q5_K_M	Tight
49.6 tok/s 30–79 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	101.5 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: Hugging Face,Korea Advanced Institute of Science and Technology (KAIST),Argilla
Organisation type: Industry,Academia,Industry
Country: United States of America, Korea (Republic of), Spain
Published: 10 April 2024
Authors: Alvaro Bartolome, Jiwoo Hong, Noah Lee, Kashif Rasul, Lewis Tunstall

What it does

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

Domain: Language
Task: Language modeling/generation, Question answering
Base model: Mixtral 8x22B

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: 141B
Training data: tokens
Epochs: 3

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: Hardware
Fine-tuning compute: 4.4 × 10¹⁹ FLOP

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 H100 SXM5 80GB
Chips used: 32
Wall-clock time: 1 hours
Power draw: 44.3 kW

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

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: Model Card for Zephyr 141B-A39B
Last updated: 28 November 2025

The extremes

The ten fastest GPUs that run Zephyr 141B-A39B

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 Zephyr 141B-A39B

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

A100 SXM4 80 GB

Memory needed

68.6 GB

Fastest

105 tok/s

Zephyr 141B-A39B reaches a parameter count of 141B. That puts it above consumer hardware, into the range where a card is bought for this purpose rather than repurposed for it. The number of cards we track that can hold it: 37.

The smallest card that holds it is A100 SXM4 80 GB, with a memory capacity of 80 GB, running it at a compression of Q3_K_M and producing around 59.8 tokens per second.

The quickest result comes from H100 NVL 94 GB, generating roughly 105 tokens per second on the strength of a memory bandwidth of 3,940 GB/s.

What this model is

Zephyr 141B-A39B was published by Hugging Face,Korea Advanced Institute of Science and Technology (KAIST),Argilla, in the country recorded as United States of America, during April 2024. The publishing organisation is categorised as industry,Academia,Industry.

It works in the domain of Language, and is recorded as performing the task of language modeling/generation, Question answering.

Its starting point was an existing base model, Mixtral 8x22B. Most models at this scale are adapted from an existing base rather than built from nothing.

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

What decides the speed

Across every card that can run it, the middle of the range sits at 59.8 tokens per second. Clearing the ten tokens per second that roughly matches reading speed: 35 of them.

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.

Memory here is estimated from size rather than computed from the architecture, which is not recorded for this model — the numbers are indicative rather than exact.

Step by step

How to choose a GPU for Zephyr 141B-A39B

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 Zephyr 141B-A39B, needing around 68.6 GB at a compression of Q3_K_M. No amount of processing power compensates for a card that cannot hold it.
02

Set the context length you will work at

Longer conversations cost memory on top of what the weights need. Move the slider to your real working length before trusting any row for Zephyr 141B-A39B.
03

Set a quality floor

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

Sort by speed

Sort by speed to see how cards rank for Zephyr 141B-A39B. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is H100 NVL 94 GB, at 105 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 Zephyr 141B-A39B. 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 Zephyr 141B-A39B.

Answers

Zephyr 141B-A39B — common questions

Zephyr 141B-A39B— would two GPUs run it faster?

A second card roughly doubles the memory available but not the generation rate. The number already able to run it alone: 37. So a second card is rarely the answer here.

Zephyr 141B-A39B— why does the quantisation differ between cards?

A larger card holds a more accurate copy. The number of compression levels used across the cards that run it: 6. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

Zephyr 141B-A39B— 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: 63–168 tok/s on H100 NVL 94 GB. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

Zephyr 141B-A39B— what GPU do I need to run it?

The smallest card in our catalogue that holds it is A100 SXM4 80 GB, with a memory capacity of 80 GB. It runs the model at a compression of Q3_K_M using about 68.6 GB, and produces roughly 59.8 tokens per second. The number of cards able to run it in total: 37.

Zephyr 141B-A39B— how fast is it on a GPU?

It depends on the card. The quickest we calculate is H100 NVL 94 GB, at about 105 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: 35.

Zephyr 141B-A39B— how much VRAM does it need?

It needs about 68.6 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.

Zephyr 141B-A39B— 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.

Zephyr 141B-A39B— how many parameters does it have?

It has a parameter count of 141B. "A Mixture of Experts (MoE) model with 141B total parameters and 39B active 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.

Zephyr 141B-A39B— who created it?

It was published by Hugging Face,Korea Advanced Institute of Science and Technology (KAIST),Argilla, based in United States of America, an organisation categorised as industry,Academia,Industry.

Zephyr 141B-A39B— when was it released?

It was published in April 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.

Zephyr 141B-A39B— what is it used for?

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

Zephyr 141B-A39B— where can I download it?

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

Zephyr 141B-A39B— can I run it 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 model is rarely worth using. The nearest miss we calculate falls short by 20.2 GB. Every figure here assumes the whole model is resident on the card.

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