Gemma 3 QAT 27B TPS calculator

Open weights Google DeepMind 27B 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

241 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Xeon Phi 7120P

16 GB · Q3_K_M · 9.7 tok/s

Fastest card

B200

125 tok/s · 180 GB

Which GPUs can run Gemma 3 QAT 27B?

Set the inputs, read the answer

A longer conversation needs more memory, which can push this model off smaller cards.

Hides cards that would only fit the model by compressing it below this point.

241 cards match

Calculating
Needs Quantisation Fit
125 tok/s

107–151

B200 NVIDIA 180 GB 8,000 GB/s Jan 2024 29.0 GB Q8_0 Comfortable
125 tok/s

107–151

B300 NVIDIA 288 GB 8,000 GB/s Sep 2025 29.0 GB Q8_0 Comfortable
100 tok/s

60–160 · low confidence

Radeon Instinct MI350X AMD 288 GB 8,190 GB/s Jan 2025 29.0 GB Q8_0 Comfortable
100 tok/s

60–160 · low confidence

Radeon Instinct MI355X AMD 288 GB 8,190 GB/s Jan 2025 29.0 GB Q8_0 Comfortable
80.1 tok/s

48–128 · low confidence

Radeon Instinct MI300 AMD 128 GB 6,550 GB/s Jan 2023 29.0 GB Q8_0 Comfortable
76.7 tok/s

65–92

H200 NVL NVIDIA 141 GB 4,890 GB/s Nov 2024 29.0 GB Q8_0 Comfortable
76.7 tok/s

65–92

H200 SXM 141 GB NVIDIA 141 GB 4,890 GB/s Nov 2024 29.0 GB Q8_0 Comfortable
73.4 tok/s

44–117 · low confidence

Radeon Instinct MI325X AMD 256 GB 6,000 GB/s Oct 2024 29.0 GB Q8_0 Comfortable
65.2 tok/s

39–104 · low confidence

Radeon Instinct MI300A AMD 128 GB 5,325 GB/s Dec 2023 29.0 GB Q8_0 Comfortable
65.2 tok/s

39–104 · low confidence

Radeon Instinct MI300X AMD 192 GB 5,325 GB/s Dec 2023 29.0 GB Q8_0 Comfortable
65.2 tok/s

39–104 · low confidence

Radeon Instinct MI308X AMD 192 GB 5,325 GB/s Dec 2023 29.0 GB Q8_0 Comfortable
61.8 tok/s

53–74

H100 NVL 94 GB NVIDIA 94 GB 3,940 GB/s Mar 2023 29.0 GB Q8_0 Comfortable
52.7 tok/s

45–63

H100 PCIe 96 GB NVIDIA 96 GB 3,360 GB/s Mar 2023 29.0 GB Q8_0 Comfortable
52.7 tok/s

45–63

H100 SXM5 80 GB NVIDIA 80 GB 3,360 GB/s Oct 2022 29.0 GB Q8_0 Comfortable
52.7 tok/s

45–63

H100 SXM5 94 GB NVIDIA 94 GB 3,360 GB/s Mar 2023 29.0 GB Q8_0 Comfortable
52.7 tok/s

45–63

H100 SXM5 96 GB NVIDIA 96 GB 3,360 GB/s Mar 2023 29.0 GB Q8_0 Comfortable
52.7 tok/s

45–63

H800 SXM5 NVIDIA 80 GB 3,360 GB/s Mar 2023 29.0 GB Q8_0 Comfortable
47.8 tok/s

41–57

Tesla V100 SXM2 16 GB NVIDIA 16 GB 1,130 GB/s Nov 2019 13.3 GB Q3_K_M Tight
42.6 tok/s

36–51

DRIVE A100 PROD NVIDIA 32 GB 1,870 GB/s May 2020 22.7 GB Q6_K Comfortable
42.6 tok/s

36–51

GRID A100A NVIDIA 32 GB 1,870 GB/s May 2020 22.7 GB Q6_K Comfortable
40.8 tok/s

35–49

GeForce RTX 5090 NVIDIA 32 GB 1,790 GB/s Jan 2025 22.7 GB Q6_K Comfortable
40.8 tok/s

35–49

GeForce RTX 5090 D NVIDIA 32 GB 1,790 GB/s Jan 2025 22.7 GB Q6_K Comfortable
40.6 tok/s

35–49

GeForce RTX 5080 NVIDIA 16 GB 960 GB/s Jan 2025 13.3 GB Q3_K_M Tight
40.1 tok/s

24–64 · low confidence

Radeon Instinct MI250 AMD 128 GB 3,280 GB/s Nov 2021 29.0 GB Q8_0 Comfortable
40.1 tok/s

24–64 · low confidence

Radeon Instinct MI250X AMD 128 GB 3,280 GB/s Nov 2021 29.0 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
Google DeepMind
Organisation type
Industry
Country
United States of America
Published
18 April 2025

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, Question answering, Translation, Chat, Quantitative reasoning, Visual question answering, Code generation
Base model
Gemma 3 27B

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
27B

27B

Training data
tokens

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

Gemma license https://huggingface.co/google/gemma-3-27b-it-qat-q4_0-gguf

Hugging Face
google

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
Gemma 3 QAT Models: Bringing state-of-the-Art AI to consumer GPUs
Last updated
28 November 2025

The extremes

The ten fastest GPUs that run Gemma 3 QAT 27B

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.

  1. 01 B300 288 GB · 8,000 GB/s · Q8_0 125 tok/s
  2. 02 B200 180 GB · 8,000 GB/s · Q8_0 125 tok/s
  3. 03 Radeon Instinct MI350X 288 GB · 8,190 GB/s · Q8_0 100 tok/s
  4. 04 Radeon Instinct MI355X 288 GB · 8,190 GB/s · Q8_0 100 tok/s
  5. 05 Radeon Instinct MI300 128 GB · 6,550 GB/s · Q8_0 80.1 tok/s
  6. 06 H200 NVL 141 GB · 4,890 GB/s · Q8_0 76.7 tok/s
  7. 07 H200 SXM 141 GB 141 GB · 4,890 GB/s · Q8_0 76.7 tok/s
  8. 08 Radeon Instinct MI325X 256 GB · 6,000 GB/s · Q8_0 73.4 tok/s
  9. 09 Radeon Instinct MI300A 128 GB · 5,325 GB/s · Q8_0 65.2 tok/s
  10. 10 Radeon Instinct MI300X 192 GB · 5,325 GB/s · Q8_0 65.2 tok/s

The smallest GPUs that still run Gemma 3 QAT 27B

The cheapest route in, by memory capacity. A tight fit runs the model but leaves nothing spare for a longer conversation.

  1. 01 Ryzen AI Z2 Extreme GPU 16 GB · needs 13.3 GB · Q3_K_M · tight 8.5 tok/s
  2. 02 Radeon RX 7700 16 GB · needs 13.3 GB · Q3_K_M · tight 20.5 tok/s
  3. 03 Arc Pro B50 16 GB · needs 13.3 GB · Q3_K_M · tight 6.2 tok/s
  4. 04 RTX PRO 2000 Blackwell 16 GB · needs 13.3 GB · Q3_K_M · tight 12.2 tok/s
  5. 05 Ryzen Z2 Extreme GPU 16 GB · needs 13.3 GB · Q3_K_M · tight 4.2 tok/s
  6. 06 Radeon RX 9060 XT 16 GB 16 GB · needs 13.3 GB · Q3_K_M · tight 10.6 tok/s
  7. 07 GeForce RTX 5060 Ti 16 GB 16 GB · needs 13.3 GB · Q3_K_M · tight 19.0 tok/s
  8. 08 GeForce RTX 5080 Mobile 16 GB · needs 13.3 GB · Q3_K_M · tight 37.9 tok/s
  9. 09 Radeon RX 9070 16 GB · needs 13.3 GB · Q3_K_M · tight 21.3 tok/s
  10. 10 Radeon RX 9070 XT 16 GB · needs 13.3 GB · Q3_K_M · tight 21.3 tok/s

What the numbers mean

The hardware side

Minimum card

Xeon Phi 7120P

Memory needed

13.3 GB

Fastest

125 tok/s

Gemma 3 QAT 27B reaches a parameter count of 27B. That lands in the range a serious desktop card can handle once the weights are compressed. The number of cards we track that can run it: 241.

At the low end it is handled by Xeon Phi 7120P, with a memory capacity of 16 GB, running it at a compression of Q3_K_M and producing around 9.7 tokens per second.

The quickest result comes from B200, generating roughly 125 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

Where it came from

Gemma 3 QAT 27B was published by Google DeepMind, in the country recorded as United States of America, during April 2025. The category the publisher falls under is industry.

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

It builds on Gemma 3 27B. 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. On Hugging Face it is published under the organisation google.

Understanding the speeds

Half the cards that hold it manage more than 19.0 tokens per second. Clearing the ten tokens per second that roughly matches reading speed: 196 of them.

Being dense, it reads all of itself per token, which is why the ordering by speed below follows the ordering by memory bandwidth so closely.

Because the architecture is recorded, the memory column is derived rather than estimated.

Step by step

How to choose a GPU for Gemma 3 QAT 27B

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

  1. 01

    Start from the memory column

    Start from what it actually needs, which is the requirement of Gemma 3 QAT 27B, needing around 13.3 GB at a compression of Q3_K_M. No amount of processing power compensates for a card that cannot hold it.

  2. 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 Gemma 3 QAT 27B.

  3. 03

    Choose how far you will compress it

    Each card runs the least-compressed copy it can hold, 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.

  4. 04

    Compare tokens per second, not specifications

    The speed ordering is effectively an ordering by memory bandwidth, for Gemma 3 QAT 27B. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B200, at 125 tok/s.

  5. 05

    Check the fit verdict before buying

    Tight means it loads and works with no room to raise the context later, in the case of Gemma 3 QAT 27B. 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.

  6. 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 Gemma 3 QAT 27B.

Answers

Gemma 3 QAT 27B — common questions

01

Gemma 3 QAT 27B— what GPU do I need to run it?

The smallest card in our catalogue that holds it is Xeon Phi 7120P, with a memory capacity of 16 GB. It runs the model at a compression of Q3_K_M using about 13.3 GB, and produces roughly 9.7 tokens per second. The number of cards able to run it in total: 241.

02

Gemma 3 QAT 27B— how fast is it on a GPU?

It depends on the card. The quickest we calculate is B200, at about 125 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: 196.

03

Gemma 3 QAT 27B— how much VRAM does it need?

It needs about 13.3 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.

04

Gemma 3 QAT 27B— 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 Q3_K_M, using about 13.3 GB and generating roughly 47.8 tokens per second. The fit is tight.

05

Gemma 3 QAT 27B— 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 Q5_K_M, using about 19.5 GB and generating roughly 37.5 tokens per second. The fit is tight.

06

Gemma 3 QAT 27B— 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.

07

Gemma 3 QAT 27B— how many parameters does it have?

It has a parameter count of 27B. 27B. 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.

08

Gemma 3 QAT 27B— who created it?

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

09

Gemma 3 QAT 27B— when was it released?

It was published in April 2025.

10

Gemma 3 QAT 27B— 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, Question answering, Translation, Chat, Quantitative reasoning, Visual question answering, Code generation. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

11

Gemma 3 QAT 27B— where can I download it?

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

12

Gemma 3 QAT 27B— 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 5.6 GB. Every figure here assumes the whole model is resident on the card.

13

Gemma 3 QAT 27B— 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: 241. So a second card is rarely the answer here.

14

Gemma 3 QAT 27B— why does the quantisation differ between cards?

A larger card holds a more accurate copy. The number of compression levels used across the cards that run it: 5. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

15

Gemma 3 QAT 27B— how accurate are these speed estimates?

Every figure is derived from memory bandwidth and model size, not benchmarked, which is why each is published as a range rather than a single number. One example: 107–151 tok/s on B200. The same model and card vary by thirty to fifty per cent depending on the inference software and its 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