MULAN TPS calculator

Open weights AIRI Artificial Intelligence Research Institute,Skolkovo Institute of Science and Technology,Belozersky Institute of Physio-Chemical Biology,Ligand Pro 35M parameters June 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

818 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Tesla C1080

4 GB · Q8_0 · 1,053 tok/s

Fastest card

B200

96,807 tok/s · 180 GB

Which GPUs can run MULAN?

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
96,807 tok/s 58,084–154,891 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	0.7 GB	Q8_0	Comfortable
96,807 tok/s 58,084–154,891 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	0.7 GB	Q8_0	Comfortable
77,303 tok/s 46,382–123,684 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	0.7 GB	Q8_0	Comfortable
77,303 tok/s 46,382–123,684 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	0.7 GB	Q8_0	Comfortable
61,823 tok/s 37,094–98,917 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	0.7 GB	Q8_0	Comfortable
59,173 tok/s 35,504–94,677 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	0.7 GB	Q8_0	Comfortable
59,173 tok/s 35,504–94,677 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	0.7 GB	Q8_0	Comfortable
56,632 tok/s 33,979–90,611 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	0.7 GB	Q8_0	Comfortable
50,261 tok/s 30,157–80,417 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	0.7 GB	Q8_0	Comfortable
50,261 tok/s 30,157–80,417 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	0.7 GB	Q8_0	Comfortable
50,261 tok/s 30,157–80,417 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	0.7 GB	Q8_0	Comfortable
47,677 tok/s 28,606–76,284 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
40,659 tok/s 24,395–65,054 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
40,659 tok/s 24,395–65,054 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	0.7 GB	Q8_0	Comfortable
40,659 tok/s 24,395–65,054 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
40,659 tok/s 24,395–65,054 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
40,659 tok/s 24,395–65,054 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
30,959 tok/s 18,575–49,534 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	0.7 GB	Q8_0	Comfortable
30,959 tok/s 18,575–49,534 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	0.7 GB	Q8_0	Comfortable
25,799 tok/s 15,479–41,278 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	0.7 GB	Q8_0	Comfortable
25,248 tok/s 15,149–40,397 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	0.7 GB	Q8_0	Comfortable
24,686 tok/s 14,811–39,497 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	0.7 GB	Q8_0	Comfortable
24,686 tok/s 14,811–39,497 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	0.7 GB	Q8_0	Comfortable
24,686 tok/s 14,811–39,497 · low confidence	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	0.7 GB	Q8_0	Comfortable
24,686 tok/s 14,811–39,497 · low confidence	H100 CNX NVIDIA	80 GB	2,040 GB/s	Mar 2023	0.7 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: AIRI Artificial Intelligence Research Institute,Skolkovo Institute of Science and Technology,Belozersky Institute of Physio-Chemical Biology,Ligand Pro
Organisation type: Research collective,Academia,Academia,Industry
Country: Russia
Published: 2 June 2024
Authors: Daria Frolova, Marina A. Pak, Anna Litvin, Ilya Sharov, Dmitry N. Ivankov, Ivan Oseledets

What it does

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

Domain: Biology
Task: Protein or nucleotide language model (pLM/nLM)
Base model: ESM2-35M

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: 35M
Training data: tokens
Epochs: 15

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: 5.2 × 10²⁰ 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 H100 SXM5 80GB
Chips used: 1
Power draw: 760 W

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: DFrolova

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: 7

Sources

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

Reference: MULAN: Multimodal Protein Language Model for Sequence and Structure Encoding
Last updated: 28 November 2025

The extremes

The ten fastest GPUs that run MULAN

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 MULAN

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

The hardware side

Minimum card

Tesla C1080

Memory needed

0.7 GB

Fastest

96,807 tok/s

MULAN is small enough at 35M parameters that hardware is rarely the obstacle — 818 of the cards we track can run it, including cards several years old.

At the low end, a Tesla C1080 handles it — 4 GB, at Q8_0, for about 1,053 tokens per second.

A B200 is the fastest we calculate for it: about 96,807 tokens per second, from 8,000 GB/s of memory bandwidth.

Background

MULAN was published by AIRI Artificial Intelligence Research Institute,Skolkovo Institute of Science and Technology,Belozersky Institute of Physio-Chemical Biology,Ligand Pro, in Russia, in June 2024. It comes out of research collective,Academia,Academia,Industry.

It works in Biology, and is recorded as doing protein or nucleotide language model (pLM/nLM).

It is derived from ESM2-35M rather than trained from scratch, which 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. It is published under the DFrolova organisation on Hugging Face.

Reading the throughput figures

Half the cards that hold it manage more than 2,718.3 tokens per second, and 818 exceed reading speed outright.

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

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.

How it was trained

Producing it required around 5.2 × 10²⁰ FLOP of arithmetic, on NVIDIA H100 SXM5 80GB, which is a statement about the training budget rather than about inference.

Step by step

How to choose a GPU for MULAN

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

01

Check what it needs before anything else

Every card here has been checked against MULAN — around 0.7 GB at Q8_0. Capacity is the gate — a card either holds it or it does not.
02

Set the context length you will work at

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

Set a quality floor

Compression is what makes MULAN fit smaller cards, at some cost in accuracy — Q8_0 on the smallest card that fits. A minimum quality removes the ones that go too far.
04

Sort by speed

Sort by speed to see how cards rank for MULAN. It will not match a gaming ordering — generation is bound by memory bandwidth, which is why the B200 tops it at 96,807 tok/s.
05

Check the fit verdict before buying

Tight means MULAN 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

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

Answers

MULAN — common questions

How accurate are these MULAN speed estimates?

These are estimates with real error bars. The fastest result here, 58,084–154,891 tok/s on the B200, could reasonably land anywhere in its published range depending on which runtime you use.

What GPU do I need to run MULAN?

The smallest card in our catalogue that holds MULAN is the Tesla C1080, with 4 GB of memory. It runs the model at Q8_0 using about 0.7 GB, and produces roughly 1,053 tokens per second. 818 cards in total can run it.

How fast is MULAN on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 96,807 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 MULAN clear that.

How much VRAM does MULAN need?

About 0.7 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 MULAN on a 8 GB GPU?

Yes. A CMP 170HX 8 GB with 8 GB runs it at Q8_0, using about 0.7 GB and generating roughly 18,030 tokens per second — a comfortable fit.

Can I run MULAN on a 12 GB GPU?

Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q8_0, using about 0.7 GB and generating roughly 11,041 tokens per second — a comfortable fit.

Can I run MULAN on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q8_0, using about 0.7 GB and generating roughly 13,674 tokens per second — a comfortable fit.

Can I run MULAN on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 0.7 GB and generating roughly 16,215 tokens per second — a comfortable fit.

Is MULAN open source?

Its weights are published, so MULAN 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 MULAN have?

MULAN has 35M 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 MULAN?

MULAN was published by AIRI Artificial Intelligence Research Institute,Skolkovo Institute of Science and Technology,Belozersky Institute of Physio-Chemical Biology,Ligand Pro, based in Russia, categorised as research collective,Academia,Academia,Industry.

When was MULAN released?

MULAN was published in June 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.

What is MULAN used for?

MULAN works in Biology, and is recorded as handling protein or nucleotide language model (pLM/nLM). Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

Where can I download MULAN?

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

How much compute was used to train MULAN?

Around 5.2 × 10²⁰ FLOP, on 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.

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

Would two GPUs run MULAN faster?

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

Why does the quantisation differ between cards for MULAN?

Each card is shown running the least-compressed copy it can hold, and MULAN 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