Phi-4-Reasoning TPS calculator

Open weights Microsoft 14B parameters April 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

306 cards that can run it

818 cards we hold specifications for

Smallest card that fits

P102-101

10 GB · IQ4_XS · 20.2 tok/s

Fastest card

B200

242 tok/s · 180 GB

Which GPUs can run Phi-4-Reasoning?

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

306 cards match

Calculating

					Needs	Quantisation	Fit
242 tok/s 145–387 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	15.7 GB	Q8_0	Comfortable
242 tok/s 145–387 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	15.7 GB	Q8_0	Comfortable
193 tok/s 116–309 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	15.7 GB	Q8_0	Comfortable
193 tok/s 116–309 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	15.7 GB	Q8_0	Comfortable
155 tok/s 93–247 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	15.7 GB	Q8_0	Comfortable
148 tok/s 89–237 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	15.7 GB	Q8_0	Comfortable
148 tok/s 89–237 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	15.7 GB	Q8_0	Comfortable
142 tok/s 85–227 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	15.7 GB	Q8_0	Comfortable
126 tok/s 75–201 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	15.7 GB	Q8_0	Comfortable
126 tok/s 75–201 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	15.7 GB	Q8_0	Comfortable
126 tok/s 75–201 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	15.7 GB	Q8_0	Comfortable
119 tok/s 72–191 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	15.7 GB	Q8_0	Comfortable
116 tok/s 70–185 · low confidence	CMP 170HX 10 GB NVIDIA	10 GB	1,560 GB/s	Sep 2021	8.4 GB	IQ4_XS	Tight
102 tok/s 61–163 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	15.7 GB	Q8_0	Comfortable
102 tok/s 61–163 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	15.7 GB	Q8_0	Comfortable
102 tok/s 61–163 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	15.7 GB	Q8_0	Comfortable
102 tok/s 61–163 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	15.7 GB	Q8_0	Comfortable
102 tok/s 61–163 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	15.7 GB	Q8_0	Comfortable
77.4 tok/s 46–124 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	15.7 GB	Q8_0	Comfortable
77.4 tok/s 46–124 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	15.7 GB	Q8_0	Comfortable
64.5 tok/s 39–103 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	15.7 GB	Q8_0	Comfortable
63.1 tok/s 38–101 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	15.7 GB	Q8_0	Comfortable
61.7 tok/s 37–99 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	15.7 GB	Q8_0	Comfortable
61.7 tok/s 37–99 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	15.7 GB	Q8_0	Comfortable
61.7 tok/s 37–99 · low confidence	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	15.7 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: 30 April 2025
Authors: Marah Abdin, Sahaj Agarwal, Ahmed Awadallah, Vidhisha Balachandran, Harkirat Behl, Lingjiao Chen, Gustavo de Rosa, Suriya Gunasekar, Mojan Javaheripi, Neel Joshi, Piero Kauffmann, Yash Lara, Caio César Teodoro Mendes, Arindam Mitra, Besmira Nushi, Dimitris Papailiopoulos, Olli Saarikivi, Shital Shah, Vaishnavi Shrivastava, Vibhav Vineet, Yue Wu, Safoora Yousefi, Guoqing Zheng

What it does

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

Domain: Language
Task: Language modeling/generation, Quantitative reasoning, Question answering, Mathematical reasoning, Code generation
Base model: Phi-4

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: 14B
Training data: tokens
Epochs: 1.93

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: 9.3 × 10²³ FLOP
How it was established: Operation counting,Hardware
Fine-tuning compute: 1.7 × 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 H100 SXM5 80GB
Chips used: 32
Wall-clock time: 60 hours
Power draw: 43.9 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-reasoning Technical Report
Last updated: 28 November 2025

The extremes

The ten fastest GPUs that run Phi-4-Reasoning

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-Reasoning

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

Hardware requirements in practice

Minimum card

P102-101

Memory needed

8.4 GB

Fastest

242 tok/s

Phi-4-Reasoning is small enough at 14B parameters that hardware is rarely the obstacle — 306 of the cards we track can run it, including cards several years old.

The entry point is the P102-101: 10 GB of memory, IQ4_XS compression, roughly 20.2 tokens per second.

The quickest result comes from a B200 at around 242 tokens per second — its 8,000 GB/s of bandwidth is what buys that.

Where it came from

Phi-4-Reasoning was published by Microsoft, in United States of America, in April 2025. It comes out of industry.

It works in Language, and is recorded as doing language modeling/generation, Quantitative reasoning, Question answering, Mathematical reasoning, Code generation.

Its starting point was Phi-4 — most models at this scale are adapted from an existing base rather than built from nothing.

Published weights mean the model runs on your machine rather than someone else's, which is what makes the hardware question below answerable at all. It is published under the microsoft organisation on Hugging Face.

Understanding the speeds

Across every card that can run it, the middle of the range is about 20.4 tokens per second, and 268 of them clear the ten tokens per second that roughly matches reading speed.

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.

What went into building it

Producing it required around 9.3 × 10²³ 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 Phi-4-Reasoning

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

Look at what Phi-4-Reasoning actually needs — around 8.4 GB at IQ4_XS. No amount of processing power compensates for a card that cannot hold it.
02

Decide how long your conversations run

Set the context to what you will actually use. The cache grows with the conversation, and it is the usual reason Phi-4-Reasoning stops fitting a card that seemed fine.
03

Set a quality floor

The quantisation column varies by card, because a bigger card holds a more accurate copy of Phi-4-Reasoning — IQ4_XS on the smallest card that fits. Set a floor to hold the comparison at one level.
04

Rank by throughput rather than spec sheet

Ranking by tokens per second for Phi-4-Reasoning follows memory bandwidth, not core counts, which is why the B200 tops it at 242 tok/s.
05

Look at the headroom, not just the fit

A tight fit runs Phi-4-Reasoning but leaves nothing spare for a longer conversation; comfortable has headroom. If you expect to grow the context, buy for comfortable.
06

Open the card you have settled on

Every card name links to its own page, which runs the same calculation across the whole model catalogue. Worth a look before buying for Phi-4-Reasoning alone — a card is usually bought for more than one model.

Answers

Phi-4-Reasoning — common questions

How much compute was used to train Phi-4-Reasoning?

Around 9.3 × 10²³ 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 Phi-4-Reasoning if it does not fit in my GPU?

Only by offloading, which is usually a false economy: the part in system memory drags the whole thing down — the nearest miss we calculate is short by 2.0 GB. Our figures for Phi-4-Reasoning assume it is fully resident.

Would two GPUs run Phi-4-Reasoning faster?

Two cards buy memory rather than speed. That matters for Phi-4-Reasoning only if one card cannot hold it — 306 can, so a second adds little.

Why does the quantisation differ between cards for Phi-4-Reasoning?

A larger card holds a more accurate copy. Across the cards that run Phi-4-Reasoning, 5 compression levels are used; the floor control above pins it to one.

How accurate are these Phi-4-Reasoning speed estimates?

They are calculated from specifications rather than measured, and each carries a range — 145–387 tok/s on the B200, for instance. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

What GPU do I need to run Phi-4-Reasoning?

The smallest card in our catalogue that holds Phi-4-Reasoning is the P102-101, with 10 GB of memory. It runs the model at IQ4_XS using about 8.4 GB, and produces roughly 20.2 tokens per second. 306 cards in total can run it.

How fast is Phi-4-Reasoning on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 242 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 268 of the cards that can run Phi-4-Reasoning clear that.

How much VRAM does Phi-4-Reasoning need?

About 8.4 GB at IQ4_XS 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 Phi-4-Reasoning on a 12 GB GPU?

Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q5_K_M, using about 10.8 GB and generating roughly 49.3 tokens per second — a tight fit.

Can I run Phi-4-Reasoning on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q6_K, using about 12.4 GB and generating roughly 49.7 tokens per second — a tight fit.

Can I run Phi-4-Reasoning on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 15.7 GB and generating roughly 40.5 tokens per second — a comfortable fit.

Is Phi-4-Reasoning open source?

Its weights are published, so Phi-4-Reasoning 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 Phi-4-Reasoning have?

Phi-4-Reasoning has 14B parameters. 14B. 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 Phi-4-Reasoning?

Phi-4-Reasoning was published by Microsoft, based in United States of America, categorised as industry.

When was Phi-4-Reasoning released?

Phi-4-Reasoning was published in April 2025.

What is Phi-4-Reasoning used for?

Phi-4-Reasoning works in Language, and is recorded as handling language modeling/generation, Quantitative reasoning, Question answering, Mathematical reasoning, Code generation. 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.

Where can I download Phi-4-Reasoning?

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

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