Molmo 72B TPS calculator

Open weights Allen Institute for AI,University of Washington 72B parameters September 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

61 cards that can run it

818 cards we hold specifications for

Smallest card that fits

A100 PCIe 40 GB

40 GB · Q3_K_M · 24.8 tok/s

Fastest card

B200

47.1 tok/s · 180 GB

Which GPUs can run Molmo 72B?

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

61 cards match

Calculating

					Needs	Quantisation	Fit
47.1 tok/s 28–75 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	77.8 GB	Q8_0	Comfortable
47.1 tok/s 28–75 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	77.8 GB	Q8_0	Comfortable
37.6 tok/s 23–60 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	77.8 GB	Q8_0	Comfortable
37.6 tok/s 23–60 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	77.8 GB	Q8_0	Comfortable
30.1 tok/s 18–48 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	77.8 GB	Q8_0	Comfortable
28.8 tok/s 17–46 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	77.8 GB	Q8_0	Comfortable
28.8 tok/s 17–46 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	77.8 GB	Q8_0	Comfortable
28.7 tok/s 17–46 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	61.0 GB	Q6_K	Tight
28.7 tok/s 17–46 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	61.0 GB	Q6_K	Tight
27.5 tok/s 17–44 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	77.8 GB	Q8_0	Comfortable
27.0 tok/s 16–43 · low confidence	GRID A100B NVIDIA	48 GB	1,870 GB/s	May 2020	40.1 GB	IQ4_XS	Tight
24.8 tok/s 15–40 · low confidence	A100 PCIe 40 GB NVIDIA	40 GB	1,560 GB/s	Jun 2020	35.9 GB	Q3_K_M	Tight
24.8 tok/s 15–40 · low confidence	A100 SXM4 40 GB NVIDIA	40 GB	1,560 GB/s	May 2020	35.9 GB	Q3_K_M	Tight
24.8 tok/s 15–40 · low confidence	A800 PCIe 40 GB NVIDIA	40 GB	1,560 GB/s	Nov 2022	35.9 GB	Q3_K_M	Tight
24.4 tok/s 15–39 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	77.8 GB	Q8_0	Comfortable
24.4 tok/s 15–39 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	77.8 GB	Q8_0	Comfortable
24.4 tok/s 15–39 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	77.8 GB	Q8_0	Comfortable
23.2 tok/s 14–37 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	77.8 GB	Q8_0	Tight
21.2 tok/s 13–34 · low confidence	H100 SXM5 64 GB NVIDIA	64 GB	2,020 GB/s	Mar 2023	52.6 GB	Q5_K_M	Tight
19.8 tok/s 12–32 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	77.8 GB	Q8_0	Tight
19.8 tok/s 12–32 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	77.8 GB	Q8_0	Tight
19.8 tok/s 12–32 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	77.8 GB	Q8_0	Tight
19.4 tok/s 12–31 · low confidence	RTX PRO 5000 Blackwell NVIDIA	48 GB	1,340 GB/s	Mar 2025	40.1 GB	IQ4_XS	Tight
17.4 tok/s 10–28 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	61.0 GB	Q6_K	Tight
17.4 tok/s 10–28 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	61.0 GB	Q6_K	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: Allen Institute for AI,University of Washington
Organisation type: Research collective,Academia
Country: United States of America
Published: 25 September 2024
Authors: Matt Deitke, Christopher Clark, Sangho Lee, Rohun Tripathi, Yue Yang, Jae Sung Park, Mohammadreza Salehi, Niklas Muennighoff, Kyle Lo, Luca Soldaini, Jiasen Lu, Taira Anderson, Erin Bransom, Kiana Ehsani, Huong Ngo, YenSung Chen, Ajay Patel, Mark Yatskar, Chris Callison-Burch, Andrew Head, Rose Hendrix, Favyen Bastani, Eli VanderBilt, Nathan Lambert, Yvonne Chou, Arnavi Chheda, Jenna Sparks, Sam S…

What it does

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

Domain: Language, Vision, Multimodal
Task: Language modeling/generation, Visual question answering, Question answering
Base model: Qwen2-72B,CLIP (ViT L/14@336px)
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
Training data: tokens

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 × 10²² FLOP
How it was established: Hardware,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.

Training hardware: NVIDIA H100 SXM5 80GB

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: Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Multimodal Models
Last updated: 28 November 2025

The extremes

The ten fastest GPUs that run Molmo 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.

The smallest GPUs that still run Molmo 72B

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 PCIe 40 GB

Memory needed

35.9 GB

Fastest

47.1 tok/s

Molmo 72B reaches a parameter count of 72B. 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: 61.

The smallest card that holds it is A100 PCIe 40 GB, with a memory capacity of 40 GB, running it at a compression of Q3_K_M and producing around 24.8 tokens per second.

Top of the range is B200, generating roughly 47.1 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

Where it came from

Molmo 72B was published by Allen Institute for AI,University of Washington, in the country recorded as United States of America, during September 2024. It comes out of an organisation categorised as research collective,Academia.

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

It builds on Qwen2-72B,CLIP (ViT L/14@336px). That is the usual way a specialised model is produced.

The weights being open is what puts this page in the calculator rather than only in the catalogue: it is a model you can actually hold.

Understanding the speeds

The median result is around 16.6 tokens per second. Producing text faster than most people read it: 51 of them.

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.

Its internal architecture is not on file, so memory is approximated from the parameter count and marked accordingly. Expect the real figure to differ, more so at long context.

Training and provenance

Producing it required arithmetic totalling around 1.3 × 10²² FLOP, on hardware recorded as NVIDIA H100 SXM5 80GB. That figure measures what producing the model cost, and has no bearing on how fast it answers.

Step by step

How to choose a GPU for Molmo 72B

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

The table lists every card able to hold Molmo 72B, needing around 35.9 GB at a compression of Q3_K_M. That figure, not the headline performance of a card, is what decides whether it runs.
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 Molmo 72B.
03

Decide how much compression you will accept

The quantisation column varies by card, because a bigger card holds a more accurate copy, 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

Sort by speed to see how cards rank for Molmo 72B. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B200, at 47.1 tok/s.
05

Read the fit column last

The fit column separates cards that just manage it from those with room to spare, in the case of Molmo 72B. 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

Each card page repeats this sweep for every model we hold, answering what else the hardware is good for beyond Molmo 72B.

Answers

Molmo 72B — common questions

Molmo 72B— 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: 61. So a second card is rarely the answer here.

Molmo 72B— 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.

Molmo 72B— how accurate are these speed estimates?

They are calculated from specifications rather than measured, and each carries a range. One example: 28–75 tok/s on B200. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

Molmo 72B— what GPU do I need to run it?

The smallest card in our catalogue that holds it is A100 PCIe 40 GB, with a memory capacity of 40 GB. It runs the model at a compression of Q3_K_M using about 35.9 GB, and produces roughly 24.8 tokens per second. The number of cards able to run it in total: 61.

Molmo 72B— how fast is it on a GPU?

It depends on the card. The quickest we calculate is 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 the number of cards clearing that: 51.

Molmo 72B— how much VRAM does it need?

It needs about 35.9 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.

Molmo 72B— 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.

Molmo 72B— how many parameters does it have?

It has a parameter count of 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.

Molmo 72B— who created it?

It was published by Allen Institute for AI,University of Washington, based in United States of America, an organisation categorised as research collective,Academia.

Molmo 72B— when was it released?

It was published in September 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.

Molmo 72B— what is it used for?

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

Molmo 72B— 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.

Molmo 72B— how much compute was used to train it?

Training consumed around 1.3 × 10²² FLOP, on hardware recorded as NVIDIA H100 SXM5 80GB. 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.

Molmo 72B— 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 15.5 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