XLMR-XXL TPS calculator

Open weights Facebook AI Research 10.7B parameters August 2021

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

509 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Xeon Phi 5110P

8 GB · Q4_K_M · 19.0 tok/s

Fastest card

B200

317 tok/s · 180 GB

Which GPUs can run XLMR-XXL?

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.

509 cards match

Calculating
Needs Quantisation Fit
317 tok/s

190–507 · low confidence

B200 NVIDIA 180 GB 8,000 GB/s Jan 2024 12.2 GB Q8_0 Comfortable
317 tok/s

190–507 · low confidence

B300 NVIDIA 288 GB 8,000 GB/s Sep 2025 12.2 GB Q8_0 Comfortable
253 tok/s

152–405 · low confidence

Radeon Instinct MI350X AMD 288 GB 8,190 GB/s Jan 2025 12.2 GB Q8_0 Comfortable
253 tok/s

152–405 · low confidence

Radeon Instinct MI355X AMD 288 GB 8,190 GB/s Jan 2025 12.2 GB Q8_0 Comfortable
202 tok/s

121–324 · low confidence

Radeon Instinct MI300 AMD 128 GB 6,550 GB/s Jan 2023 12.2 GB Q8_0 Comfortable
194 tok/s

116–310 · low confidence

H200 NVL NVIDIA 141 GB 4,890 GB/s Nov 2024 12.2 GB Q8_0 Comfortable
194 tok/s

116–310 · low confidence

H200 SXM 141 GB NVIDIA 141 GB 4,890 GB/s Nov 2024 12.2 GB Q8_0 Comfortable
185 tok/s

111–296 · low confidence

Radeon Instinct MI325X AMD 256 GB 6,000 GB/s Oct 2024 12.2 GB Q8_0 Comfortable
164 tok/s

99–263 · low confidence

Radeon Instinct MI300A AMD 128 GB 5,325 GB/s Dec 2023 12.2 GB Q8_0 Comfortable
164 tok/s

99–263 · low confidence

Radeon Instinct MI300X AMD 192 GB 5,325 GB/s Dec 2023 12.2 GB Q8_0 Comfortable
164 tok/s

99–263 · low confidence

Radeon Instinct MI308X AMD 192 GB 5,325 GB/s Dec 2023 12.2 GB Q8_0 Comfortable
156 tok/s

94–250 · low confidence

H100 NVL 94 GB NVIDIA 94 GB 3,940 GB/s Mar 2023 12.2 GB Q8_0 Comfortable
136 tok/s

82–218 · low confidence

CMP 170HX 8 GB NVIDIA 8 GB 1,490 GB/s Sep 2021 7.2 GB Q4_K_M Tight
133 tok/s

80–213 · low confidence

H100 PCIe 96 GB NVIDIA 96 GB 3,360 GB/s Mar 2023 12.2 GB Q8_0 Comfortable
133 tok/s

80–213 · low confidence

H100 SXM5 80 GB NVIDIA 80 GB 3,360 GB/s Oct 2022 12.2 GB Q8_0 Comfortable
133 tok/s

80–213 · low confidence

H100 SXM5 94 GB NVIDIA 94 GB 3,360 GB/s Mar 2023 12.2 GB Q8_0 Comfortable
133 tok/s

80–213 · low confidence

H100 SXM5 96 GB NVIDIA 96 GB 3,360 GB/s Mar 2023 12.2 GB Q8_0 Comfortable
133 tok/s

80–213 · low confidence

H800 SXM5 NVIDIA 80 GB 3,360 GB/s Mar 2023 12.2 GB Q8_0 Comfortable
110 tok/s

66–176 · low confidence

CMP 170HX 10 GB NVIDIA 10 GB 1,560 GB/s Sep 2021 8.4 GB Q5_K_M Tight
101 tok/s

61–162 · low confidence

Radeon Instinct MI250 AMD 128 GB 3,280 GB/s Nov 2021 12.2 GB Q8_0 Comfortable
101 tok/s

61–162 · low confidence

Radeon Instinct MI250X AMD 128 GB 3,280 GB/s Nov 2021 12.2 GB Q8_0 Comfortable
84.4 tok/s

51–135 · low confidence

Data Center GPU Max 1550 Intel 128 GB 3,280 GB/s Jan 2023 12.2 GB Q8_0 Comfortable
82.6 tok/s

50–132 · low confidence

Data Center GPU Max Subsystem Intel 128 GB 3,210 GB/s Jan 2023 12.2 GB Q8_0 Comfortable
80.8 tok/s

48–129 · low confidence

A100 SXM4 80 GB NVIDIA 80 GB 2,040 GB/s Nov 2020 12.2 GB Q8_0 Comfortable
80.8 tok/s

48–129 · low confidence

A100X NVIDIA 80 GB 2,040 GB/s Jun 2021 12.2 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
Facebook AI Research
Organisation type
Industry
Country
United States of America, France
Published
17 August 2021
Authors
Naman Goyal, Jingfei Du, Myle Ott, Giri Anantharaman, Alexis Conneau

What it does

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

Domain
Language
Task
Translation, Language modeling/generation

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
10.7B

Section 2.1: " ...XLM-RXXL (L= 48, H = 4096, A = 32, 10.7B params)"

Training data
167,000,000,000 tokens

"We pretrain the models on the CC100 dataset, which corresponds to 167B tokens in 100 languages."

Epochs
3.14
Batch size
1,048,576

Batches of 2048 with sequence length of 512

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
3.4 × 10²² FLOP

Trained for 500k steps at a batch size of 2048 with sequence length of 512 = 524,288,000,000 tokens seen. 6 * 10700000000 * 524,288,000,000 = 3.366e22

How it was established
Operation counting

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

https://github.com/facebookresearch/fairseq/tree/main/examples/xlmr

How it is classified

Labels the source dataset applies when tracking notable models, and how confident it is in the entry.

Why it is tracked
SOTA improvement

Abstract: "Our model also outperforms the RoBERTa-Large model on several English tasks of the GLUE benchmark by 0.3% on average while handling 99 more languages."

Record confidence
Confident
Citations
159

Sources

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

Reference
Larger-Scale Transformers for Multilingual Masked Language Modeling
Last updated
25 May 2026

The extremes

What the numbers mean

The hardware side

Minimum card

Xeon Phi 5110P

Memory needed

7.2 GB

Fastest

317 tok/s

XLMR-XXL is small enough at 10.7B parameters that hardware is rarely the obstacle — 509 of the cards we track can run it, including cards several years old.

At the low end, a Xeon Phi 5110P handles it — 8 GB, at Q4_K_M, for about 19.0 tokens per second.

A B200 is the fastest we calculate for it: about 317 tokens per second, from 8,000 GB/s of memory bandwidth.

About this model

XLMR-XXL was published by Facebook AI Research, in United States of America, in August 2021. The organisation is categorised as industry.

It works in Language, and is recorded as doing translation, Language modeling/generation.

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.

How fast it runs, and why

Half the cards that hold it manage more than 20.5 tokens per second, and 461 exceed reading speed outright.

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.

Its internal architecture is not on file, so memory is approximated from the parameter count and marked accordingly. Expect the real figure to differ, more so at long context.

Training and provenance

Training it took roughly 3.4 × 10²² FLOP of computation — a measure of what producing the model cost, not of how fast it answers.

It was trained on about 167,000,000,000 tokens of text.

Its inclusion criterion is sOTA improvement.

Step by step

How to choose a GPU for XLMR-XXL

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

  1. 01

    Read the memory figure first

    The table lists every card that can hold XLMR-XXL — around 7.2 GB at Q4_K_M. That figure, not the card's headline performance, is what decides whether it runs.

  2. 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 XLMR-XXL stops fitting a card that seemed fine.

  3. 03

    Choose how far you will compress it

    The quantisation column varies by card, because a bigger card holds a more accurate copy of XLMR-XXL — Q4_K_M on the smallest card that fits. Set a floor to hold the comparison at one level.

  4. 04

    Compare tokens per second, not specifications

    Sort by speed to see how cards rank for XLMR-XXL. It will not match a gaming ordering — generation is bound by memory bandwidth, which is why the B200 tops it at 317 tok/s.

  5. 05

    Check the fit verdict before buying

    Tight means XLMR-XXL 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.

  6. 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 XLMR-XXL alone — a card is usually bought for more than one model.

Answers

XLMR-XXL — common questions

01

How much compute was used to train XLMR-XXL?

Around 3.4 × 10²² FLOP. 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.

02

Can I run XLMR-XXL if it does not fit in my GPU?

It can be split between the card and system memory, but XLMR-XXL generates painfully slowly that way — the nearest miss we calculate is short by 1.8 GB. Nothing on this page assumes offloading.

03

Would two GPUs run XLMR-XXL faster?

Capacity adds across cards; throughput does not. Since 509 of the cards we track already hold XLMR-XXL on their own, a second card is rarely the answer here.

04

Why does the quantisation differ between cards for XLMR-XXL?

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

05

How accurate are these XLMR-XXL speed estimates?

Every figure is derived from memory bandwidth and model size, not benchmarked. That is why each is published as a range such as 190–507 tok/s on the B200 rather than a single number.

06

What GPU do I need to run XLMR-XXL?

The smallest card in our catalogue that holds XLMR-XXL is the Xeon Phi 5110P, with 8 GB of memory. It runs the model at Q4_K_M using about 7.2 GB, and produces roughly 19.0 tokens per second. 509 cards in total can run it.

07

How fast is XLMR-XXL on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 317 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 461 of the cards that can run XLMR-XXL clear that.

08

How much VRAM does XLMR-XXL need?

About 7.2 GB at Q4_K_M 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.

09

Can I run XLMR-XXL on a 8 GB GPU?

Yes. A CMP 170HX 8 GB with 8 GB runs it at Q4_K_M, using about 7.2 GB and generating roughly 136 tokens per second — a tight fit.

10

Can I run XLMR-XXL on a 12 GB GPU?

Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q6_K, using about 9.7 GB and generating roughly 52.5 tokens per second — a tight fit.

11

Can I run XLMR-XXL on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q8_0, using about 12.2 GB and generating roughly 44.7 tokens per second — a tight fit.

12

Can I run XLMR-XXL on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 12.2 GB and generating roughly 53.0 tokens per second — a comfortable fit.

13

Is XLMR-XXL open source?

Its weights are published, so XLMR-XXL 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.

14

How many parameters does XLMR-XXL have?

XLMR-XXL has 10.7B parameters. Section 2.1: " ...XLM-RXXL (L= 48, H = 4096, A = 32, 10.7B params)". 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.

15

Who created XLMR-XXL?

XLMR-XXL was published by Facebook AI Research, based in United States of America, categorised as industry.

16

When was XLMR-XXL released?

XLMR-XXL was published in August 2021. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

17

What is XLMR-XXL used for?

XLMR-XXL works in Language, and is recorded as handling translation, Language modeling/generation. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

18

Where can I download XLMR-XXL?

The weights for XLMR-XXL are published, though we do not hold a repository link for it. This site calculates hardware requirements rather than hosting model files.

Source

Original publication

Record last updated 25 May 2026

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.