Phi-4-Multimodal TPS calculator

Open weights Microsoft 5.6B parameters March 2025

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 · Q3_K_M · 17.8 tok/s

Fastest card

B200

605 tok/s · 180 GB

Which GPUs can run Phi-4-Multimodal?

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
605 tok/s 363–968 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	6.7 GB	Q8_0	Comfortable
605 tok/s 363–968 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	6.7 GB	Q8_0	Comfortable
483 tok/s 290–773 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	6.7 GB	Q8_0	Comfortable
483 tok/s 290–773 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	6.7 GB	Q8_0	Comfortable
386 tok/s 232–618 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	6.7 GB	Q8_0	Comfortable
370 tok/s 222–592 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	6.7 GB	Q8_0	Comfortable
370 tok/s 222–592 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	6.7 GB	Q8_0	Comfortable
354 tok/s 212–566 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	6.7 GB	Q8_0	Comfortable
314 tok/s 188–503 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	6.7 GB	Q8_0	Comfortable
314 tok/s 188–503 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	6.7 GB	Q8_0	Comfortable
314 tok/s 188–503 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	6.7 GB	Q8_0	Comfortable
298 tok/s 179–477 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	6.7 GB	Q8_0	Comfortable
254 tok/s 152–407 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	6.7 GB	Q8_0	Comfortable
254 tok/s 152–407 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	6.7 GB	Q8_0	Comfortable
254 tok/s 152–407 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	6.7 GB	Q8_0	Comfortable
254 tok/s 152–407 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	6.7 GB	Q8_0	Comfortable
254 tok/s 152–407 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	6.7 GB	Q8_0	Comfortable
193 tok/s 116–310 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	6.7 GB	Q8_0	Comfortable
193 tok/s 116–310 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	6.7 GB	Q8_0	Comfortable
161 tok/s 97–258 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	6.7 GB	Q8_0	Comfortable
158 tok/s 95–252 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	6.7 GB	Q8_0	Comfortable
154 tok/s 93–247 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	6.7 GB	Q8_0	Comfortable
154 tok/s 93–247 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	6.7 GB	Q8_0	Comfortable
154 tok/s 93–247 · low confidence	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	6.7 GB	Q8_0	Comfortable
154 tok/s 93–247 · low confidence	H100 CNX NVIDIA	80 GB	2,040 GB/s	Mar 2023	6.7 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: Microsoft
Organisation type: Industry
Country: United States of America
Published: 3 March 2025
Authors: Abdelrahman Abouelenin, Atabak Ashfaq, Adam Atkinson, Hany Awadalla, Nguyen Bach, Jianmin Bao, Alon Benhaim, Martin Cai, Vishrav Chaudhary, Congcong Chen, Dong Chen, Dongdong Chen, Junkun Chen, Weizhu Chen, Yen-Chun Chen, Yi-ling Chen, Qi Dai, Xiyang Dai, Ruchao Fan, Mei Gao, Min Gao, Amit Garg, Abhishek Goswami, Junheng Hao, Amr Hendy, Yuxuan Hu, Xin Jin, Mahmoud Khademi, Dongwoo Kim, Young Jin K…

What it does

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

Domain: Multimodal, Language, Vision, Speech
Task: Language modeling/generation, Question answering, Visual question answering, Speech recognition (ASR), Translation, Audio question answering, Character recognition (OCR)
Base model: Phi-4 Mini

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: 5.6B
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.

Training compute: 1.2 × 10²³ FLOP
How it was established: Operation counting,Hardware
Fine-tuning compute: 7.2 × 10²¹ 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 A100 SXM4 80 GB
Chips used: 512
Wall-clock time: 672 hours (28 days)
Power draw: 402.0 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: Unreleased
Hugging Face: microsoft

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: Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs
Last updated: 28 November 2025

The extremes

The ten fastest GPUs that run Phi-4-Multimodal

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 Phi-4-Multimodal

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

Tesla C1080

Memory needed

3.4 GB

Fastest

605 tok/s

Phi-4-Multimodal reaches a parameter count of 5.6B. That is small enough that hardware is rarely the obstacle, including on cards several years old. The number of cards we track that can run it: 818.

The smallest card that holds it is Tesla C1080, with a memory capacity of 4 GB, running it at a compression of Q3_K_M and producing around 17.8 tokens per second.

Top of the range is B200, generating roughly 605 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

What this model is

Phi-4-Multimodal was published by Microsoft, in the country recorded as United States of America, during March 2025. The publishing organisation is categorised as industry.

It works in the domain of Multimodal, Language, Vision, Speech, and is recorded as performing the task of language modeling/generation, Question answering, Visual question answering, Speech recognition (ASR), Translation, Audio question answering, Character recognition (OCR).

It builds on Phi-4 Mini. That is why it shares the base model's general shape and size.

The weights are published, so it can be downloaded and run on your own hardware indefinitely, offline, with no account attached. On Hugging Face it is published under the organisation microsoft.

What decides the speed

Half the cards that hold it manage more than 25.3 tokens per second. Producing text faster than most people read it: 759 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.

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

Training it took a computation budget of roughly 1.2 × 10²³ FLOP, on hardware recorded as NVIDIA A100 SXM4 80 GB. 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 Phi-4-Multimodal

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 Phi-4-Multimodal, needing around 3.4 GB at a compression of Q3_K_M. That figure, not the headline performance of a card, is what decides whether it runs.
02

Decide how long your conversations run

Longer conversations cost memory on top of what the weights need. Move the slider to your real working length before trusting any row for Phi-4-Multimodal.
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

Sort by speed

The speed ordering is effectively an ordering by memory bandwidth, for Phi-4-Multimodal. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B200, at 605 tok/s.
05

Check the fit verdict before buying

A tight fit runs, but leaves nothing spare for a longer conversation, in the case of Phi-4-Multimodal. 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

Open the card you have settled on

Each card page repeats this sweep for every model we hold, answering what else the hardware is good for beyond Phi-4-Multimodal.

Answers

Phi-4-Multimodal — common questions

Phi-4-Multimodal— who created it?

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

Phi-4-Multimodal— when was it released?

It was published in March 2025.

Phi-4-Multimodal— what is it used for?

It works in the domain of Multimodal, Language, Vision, Speech, and is recorded as handling the task of language modeling/generation, Question answering, Visual question answering, Speech recognition (ASR), Translation, Audio question answering, Character recognition (OCR). These are the areas it was designed around; they describe intent rather than a hard boundary.

Phi-4-Multimodal— where can I download it?

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

Phi-4-Multimodal— how much compute was used to train it?

Training consumed around 1.2 × 10²³ FLOP, on hardware recorded as NVIDIA A100 SXM4 80 GB. 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.

Phi-4-Multimodal— 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. Every figure here assumes the whole model is resident on the card.

Phi-4-Multimodal— would two GPUs run it faster?

Capacity adds across cards; throughput does not. The number of cards already holding it on their own: 818. So a second card is rarely the answer here.

Phi-4-Multimodal— 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: 4. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

Phi-4-Multimodal— 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: 363–968 tok/s on B200. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

Phi-4-Multimodal— what GPU do I need to run it?

The smallest card in our catalogue that holds it is Tesla C1080, with a memory capacity of 4 GB. It runs the model at a compression of Q3_K_M using about 3.4 GB, and produces roughly 17.8 tokens per second. The number of cards able to run it in total: 818.

Phi-4-Multimodal— how fast is it on a GPU?

It depends on the card. The quickest we calculate is B200, at about 605 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: 759.

Phi-4-Multimodal— how much VRAM does it need?

It needs about 3.4 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.

Phi-4-Multimodal— can I run it on a GPU holding 8 GB?

Yes. The card CMP 170HX 8 GB, holding 8 GB, runs it at a compression of Q8_0, using about 6.7 GB and generating roughly 113 tokens per second. The fit is tight.

Phi-4-Multimodal— can I run it on a GPU holding 12 GB?

Yes. The card GeForce RTX 3080 Ti, holding 12 GB, runs it at a compression of Q8_0, using about 6.7 GB and generating roughly 69.0 tokens per second. The fit is comfortable.

Phi-4-Multimodal— can I run it on a GPU holding 16 GB?

Yes. The card Tesla V100 SXM2 16 GB, holding 16 GB, runs it at a compression of Q8_0, using about 6.7 GB and generating roughly 85.5 tokens per second. The fit is comfortable.

Phi-4-Multimodal— can I run it on a GPU holding 24 GB?

Yes. The card GeForce RTX 5090 D V2, holding 24 GB, runs it at a compression of Q8_0, using about 6.7 GB and generating roughly 101 tokens per second. The fit is comfortable.

Phi-4-Multimodal— 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.

Phi-4-Multimodal— how many parameters does it have?

It has a parameter count of 5.6B. 5.6B 1. base: Phi-4 Mini (3.8b parameters) 2. "The audio encoder and projector introduce 460M parameters while LoRAA consumes another 460M parameters." 3. "The image encoder and projector introduce 440M model parameters while the vision adapter LoRAV consumes another 370M model 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.

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