Llama 3.2 90B TPS calculator

Open weights Meta AI 88.6B 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

43 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Radeon Instinct MI200

64 GB · Q4_K_M · 14.1 tok/s

Fastest card

B200

38.2 tok/s · 180 GB

Which GPUs can run Llama 3.2 90B?

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

43 cards match

Calculating

					Needs	Quantisation	Fit
38.2 tok/s 23–61 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	95.6 GB	Q8_0	Comfortable
38.2 tok/s 23–61 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	95.6 GB	Q8_0	Comfortable
30.5 tok/s 18–49 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	95.6 GB	Q8_0	Comfortable
30.5 tok/s 18–49 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	95.6 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	64.6 GB	Q5_K_M	Tight
28.7 tok/s 17–46 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	64.6 GB	Q5_K_M	Tight
27.4 tok/s 16–44 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	74.9 GB	Q6_K	Tight
24.4 tok/s 15–39 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	95.6 GB	Q8_0	Comfortable
23.4 tok/s 14–37 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	95.6 GB	Q8_0	Comfortable
23.4 tok/s 14–37 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	95.6 GB	Q8_0	Comfortable
23.3 tok/s 14–37 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	74.9 GB	Q6_K	Tight
23.3 tok/s 14–37 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	74.9 GB	Q6_K	Tight
23.3 tok/s 14–37 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	74.9 GB	Q6_K	Tight
22.4 tok/s 13–36 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	95.6 GB	Q8_0	Comfortable
22.3 tok/s 13–36 · low confidence	H100 SXM5 64 GB NVIDIA	64 GB	2,020 GB/s	Mar 2023	54.3 GB	Q4_K_M	Tight
19.9 tok/s 12–32 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	95.6 GB	Q8_0	Comfortable
19.9 tok/s 12–32 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	95.6 GB	Q8_0	Comfortable
19.9 tok/s 12–32 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	95.6 GB	Q8_0	Comfortable
17.4 tok/s 10–28 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	64.6 GB	Q5_K_M	Tight
17.4 tok/s 10–28 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	64.6 GB	Q5_K_M	Tight
17.4 tok/s 10–28 · low confidence	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	64.6 GB	Q5_K_M	Tight
17.4 tok/s 10–28 · low confidence	H100 CNX NVIDIA	80 GB	2,040 GB/s	Mar 2023	64.6 GB	Q5_K_M	Tight
17.4 tok/s 10–28 · low confidence	H100 PCIe 80 GB NVIDIA	80 GB	2,040 GB/s	Oct 2022	64.6 GB	Q5_K_M	Tight
17.4 tok/s 10–28 · low confidence	H800 PCIe 80 GB NVIDIA	80 GB	2,040 GB/s	Mar 2023	64.6 GB	Q5_K_M	Tight
16.6 tok/s 10–27 · low confidence	A100 PCIe 80 GB NVIDIA	80 GB	1,940 GB/s	Jun 2021	64.6 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: Meta AI
Organisation type: Industry
Country: United States of America
Published: 24 September 2024

What it does

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

Domain: Multimodal, Vision, Language
Task: Visual question answering, Image captioning, Object detection
Base model: Llama 3.1-70B

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: 88.6B
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.

How it was established: Hardware
Fine-tuning compute: 2.5 × 10²⁴ 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
Chip-hours: 1,770,000

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

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
Record confidence: Confident

Sources

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

Reference: Llama 3.2: Revolutionizing edge AI and vision with open, customizable models
Last updated: 28 November 2025

The extremes

The ten fastest GPUs that run Llama 3.2 90B

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 Llama 3.2 90B

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

Radeon Instinct MI200

Memory needed

54.3 GB

Fastest

38.2 tok/s

Llama 3.2 90B reaches a parameter count of 88.6B. 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: 43.

At the low end it is handled by Radeon Instinct MI200, with a memory capacity of 64 GB, running it at a compression of Q4_K_M and producing around 14.1 tokens per second.

The fastest we calculate for it is B200, generating roughly 38.2 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

What this model is

Llama 3.2 90B was published by Meta AI, in the country recorded as United States of America, during September 2024. It comes out of an organisation categorised as industry.

It works in the domain of Multimodal, Vision, Language, and is recorded as performing the task of visual question answering, Image captioning, Object detection.

Its starting point was an existing base model, Llama 3.1-70B. 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.

What decides the speed

The median result is around 17.4 tokens per second. Producing text faster than most people read it: 38 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.

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.

Step by step

How to choose a GPU for Llama 3.2 90B

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

01

Start from the memory column

The table lists every card able to hold Llama 3.2 90B, needing around 54.3 GB at a compression of Q4_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

The conversation occupies memory too, and grows as it goes. Set the slider to the length you expect, because at long context a card that handles short questions easily can be dropped by Llama 3.2 90B.
03

Decide how much compression you will accept

Each card runs the least-compressed copy it can hold, reaching a compression of Q4_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

The speed ordering is effectively an ordering by memory bandwidth, for Llama 3.2 90B. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B200, at 38.2 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 Llama 3.2 90B. 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

Every card name links to its own page, which runs the same calculation across the whole model catalogue. A card is usually bought for more than one model, so it is worth a look before buying for Llama 3.2 90B.

Answers

Llama 3.2 90B — common questions

Llama 3.2 90B— how much VRAM does it need?

It needs about 54.3 GB at a compression of Q4_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.

Llama 3.2 90B— 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.

Llama 3.2 90B— how many parameters does it have?

It has a parameter count of 88.6B. https://huggingface.co/meta-llama/Llama-3.2-90B-Vision. 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.

Llama 3.2 90B— who created it?

It was published by Meta AI, based in United States of America, an organisation categorised as industry.

Llama 3.2 90B— 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.

Llama 3.2 90B— what is it used for?

It works in the domain of Multimodal, Vision, Language, and is recorded as handling the task of visual question answering, Image captioning, Object detection. These are the areas it was designed around; they describe intent rather than a hard boundary.

Llama 3.2 90B— 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.

Llama 3.2 90B— can I run it if it does not fit in my GPU?

Only by offloading, which is usually a false economy: the part held in system memory drags the whole thing down. The nearest miss we calculate falls short by 11.1 GB. Every figure here assumes the whole model is resident on the card.

Llama 3.2 90B— would two GPUs run it faster?

Capacity adds across cards; throughput does not. The number of cards already holding it on their own: 43. So a second card is rarely the answer here.

Llama 3.2 90B— why does the quantisation differ between cards?

Because capacity varies, so does how hard it has to be squeezed. The number of distinct levels in the table above: 4. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

Llama 3.2 90B— how accurate are these speed estimates?

These are estimates with real error bars, and any of them could reasonably land anywhere in its published range depending on which runtime you use. The fastest result here: 23–61 tok/s on B200. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

Llama 3.2 90B— what GPU do I need to run it?

The smallest card in our catalogue that holds it is Radeon Instinct MI200, with a memory capacity of 64 GB. It runs the model at a compression of Q4_K_M using about 54.3 GB, and produces roughly 14.1 tokens per second. The number of cards able to run it in total: 43.

Llama 3.2 90B— how fast is it on a GPU?

It depends on the card. The quickest we calculate is B200, at about 38.2 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: 38.

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