Generative BST TPS calculator

Open weights Facebook AI Research 9.4B parameters March 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

582 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Quadro 6000

6 GB · Q3_K_M · 14.8 tok/s

Fastest card

B200

359 tok/s · 180 GB

Which GPUs can run Generative BST?

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

582 cards match

Calculating

					Needs	Quantisation	Fit
359 tok/s 216–575 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	10.8 GB	Q8_0	Comfortable
359 tok/s 216–575 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	10.8 GB	Q8_0	Comfortable
287 tok/s 172–459 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	10.8 GB	Q8_0	Comfortable
287 tok/s 172–459 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	10.8 GB	Q8_0	Comfortable
229 tok/s 138–367 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	10.8 GB	Q8_0	Comfortable
220 tok/s 132–351 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	10.8 GB	Q8_0	Comfortable
220 tok/s 132–351 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	10.8 GB	Q8_0	Comfortable
210 tok/s 126–336 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	10.8 GB	Q8_0	Comfortable
187 tok/s 112–298 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	10.8 GB	Q8_0	Comfortable
187 tok/s 112–298 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	10.8 GB	Q8_0	Comfortable
187 tok/s 112–298 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	10.8 GB	Q8_0	Comfortable
177 tok/s 106–283 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	10.8 GB	Q8_0	Comfortable
154 tok/s 93–247 · low confidence	CMP 170HX 8 GB NVIDIA	8 GB	1,490 GB/s	Sep 2021	6.4 GB	Q4_K_M	Tight
151 tok/s 91–241 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	10.8 GB	Q8_0	Comfortable
151 tok/s 91–241 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	10.8 GB	Q8_0	Comfortable
151 tok/s 91–241 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	10.8 GB	Q8_0	Comfortable
151 tok/s 91–241 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	10.8 GB	Q8_0	Comfortable
151 tok/s 91–241 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	10.8 GB	Q8_0	Comfortable
115 tok/s 69–184 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	10.8 GB	Q8_0	Comfortable
115 tok/s 69–184 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	10.8 GB	Q8_0	Comfortable
102 tok/s 61–163 · low confidence	CMP 170HX 10 GB NVIDIA	10 GB	1,560 GB/s	Sep 2021	8.6 GB	Q6_K	Tight
95.7 tok/s 57–153 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	10.8 GB	Q8_0	Comfortable
93.7 tok/s 56–150 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	10.8 GB	Q8_0	Comfortable
91.6 tok/s 55–147 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	10.8 GB	Q8_0	Comfortable
91.6 tok/s 55–147 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	10.8 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: 5 March 2021
Authors: Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Kurt Shuster, Eric M. Smith, Y-Lan Boureau, Jason Weston

What it does

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

Domain: Language
Task: Language modeling/generation, Chat, Question answering
Numerical format: FP16

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: 9.4B
Training data: 56,800,000,000 tokens
Epochs: 1.76

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: 1.4 × 10²² FLOP
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: Open source

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
Record confidence: Confident
Citations: 1,123

Sources

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

Reference: Recipes for building an open-domain chatbot
Last updated: 25 May 2026

The extremes

The ten fastest GPUs that run Generative BST

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 Generative BST

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

Quadro 6000

Memory needed

5.3 GB

Fastest

359 tok/s

Generative BST reaches a parameter count of 9.4B. 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: 582.

At the low end it is handled by Quadro 6000, with a memory capacity of 6 GB, running it at a compression of Q3_K_M and producing around 14.8 tokens per second.

The quickest result comes from B200, generating roughly 359 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

Background

Generative BST was published by Facebook AI Research, in the country recorded as United States of America, during March 2021. The publishing organisation is categorised as industry.

It works in the domain of Language, and is recorded as performing the task of language modeling/generation, Chat, Question answering.

Because its weights were released, nothing about running it depends on a provider staying available — it is yours once downloaded.

Reading the throughput figures

Across every card that can run it, the middle of the range sits at 22.6 tokens per second. Clearing the ten tokens per second that roughly matches reading speed: 543 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.

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.

How it was trained

Training it took a computation budget of roughly 1.4 × 10²² FLOP. That figure measures what producing the model cost, and has no bearing on how fast it answers.

Training consumed a corpus of around 56,800,000,000 tokens of text.

Its inclusion criterion: sOTA improvement.

Step by step

How to choose a GPU for Generative BST

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

01

Check what it needs before anything else

Start from what it actually needs, which is the requirement of Generative BST, needing around 5.3 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 Generative BST.
03

Choose how far you will compress it

Compression is what makes a model fit smaller cards, at some cost in accuracy, 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

Rank by throughput rather than spec sheet

The speed ordering is effectively an ordering by memory bandwidth, for Generative BST. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B200, at 359 tok/s.
05

Look at the headroom, not just the fit

Tight means it loads and works with no room to raise the context later, in the case of Generative BST. 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 Generative BST.

Answers

Generative BST — common questions

Generative BST— 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 10.8 GB and generating roughly 50.7 tokens per second. The fit is comfortable.

Generative BST— 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 10.8 GB and generating roughly 60.2 tokens per second. The fit is comfortable.

Generative BST— 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.

Generative BST— how many parameters does it have?

It has a parameter count of 9.4B. The largest model is a transformer with 9.4B parameters (Table 2). 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.

Generative BST— who created it?

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

Generative BST— when was it released?

It was published in March 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.

Generative BST— what is it used for?

It works in the domain of Language, and is recorded as handling the task of language modeling/generation, Chat, Question answering. These are the areas it was designed around; they describe intent rather than a hard boundary.

Generative BST— 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.

Generative BST— how much compute was used to train it?

Training consumed around 1.4 × 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.

Generative BST— can I run it if it does not fit in my GPU?

Partly. Layers that do not fit sit in system memory and run at a fraction of the speed, so a mostly-offloaded model is rarely worth using. The nearest miss we calculate falls short by 1.9 GB. Every figure here assumes the whole model is resident on the card.

Generative BST— would two GPUs run it faster?

Capacity adds across cards; throughput does not. The number of cards already holding it on their own: 582. So a second card is rarely the answer here.

Generative BST— 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.

Generative BST— 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: 216–575 tok/s on B200. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

Generative BST— what GPU do I need to run it?

The smallest card in our catalogue that holds it is Quadro 6000, with a memory capacity of 6 GB. It runs the model at a compression of Q3_K_M using about 5.3 GB, and produces roughly 14.8 tokens per second. The number of cards able to run it in total: 582.

Generative BST— how fast is it on a GPU?

It depends on the card. The quickest we calculate is B200, at about 359 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: 543.

Generative BST— how much VRAM does it need?

It needs about 5.3 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.

Generative BST— 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 Q4_K_M, using about 6.4 GB and generating roughly 154 tokens per second. The fit is tight.

Generative BST— 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 10.8 GB and generating roughly 41.0 tokens per second. The fit is tight.

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.

All AI models