Once for All TPS calculator

Open weights MIT-IBM Watson AI Lab,Massachusetts Institute of Technology (MIT),IBM 7.7M parameters April 2020

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 · Q8_0 · 4,788 tok/s

Fastest card

B200

440,031 tok/s · 180 GB

Which GPUs can run Once for All?

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
440,031 tok/s 264,018–704,049 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	0.7 GB	Q8_0	Comfortable
440,031 tok/s 264,018–704,049 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	0.7 GB	Q8_0	Comfortable
351,375 tok/s 210,825–562,201 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	0.7 GB	Q8_0	Comfortable
351,375 tok/s 210,825–562,201 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	0.7 GB	Q8_0	Comfortable
281,015 tok/s 168,609–449,623 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	0.7 GB	Q8_0	Comfortable
268,969 tok/s 161,381–430,350 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	0.7 GB	Q8_0	Comfortable
268,969 tok/s 161,381–430,350 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	0.7 GB	Q8_0	Comfortable
257,418 tok/s 154,451–411,869 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	0.7 GB	Q8_0	Comfortable
228,458 tok/s 137,075–365,533 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	0.7 GB	Q8_0	Comfortable
228,458 tok/s 137,075–365,533 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	0.7 GB	Q8_0	Comfortable
228,458 tok/s 137,075–365,533 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	0.7 GB	Q8_0	Comfortable
216,715 tok/s 130,029–346,744 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
184,813 tok/s 110,888–295,701 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
184,813 tok/s 110,888–295,701 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	0.7 GB	Q8_0	Comfortable
184,813 tok/s 110,888–295,701 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
184,813 tok/s 110,888–295,701 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
184,813 tok/s 110,888–295,701 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
140,722 tok/s 84,433–225,155 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	0.7 GB	Q8_0	Comfortable
140,722 tok/s 84,433–225,155 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	0.7 GB	Q8_0	Comfortable
117,268 tok/s 70,361–187,629 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	0.7 GB	Q8_0	Comfortable
114,765 tok/s 68,859–183,625 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	0.7 GB	Q8_0	Comfortable
112,208 tok/s 67,325–179,532 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	0.7 GB	Q8_0	Comfortable
112,208 tok/s 67,325–179,532 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	0.7 GB	Q8_0	Comfortable
112,208 tok/s 67,325–179,532 · low confidence	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	0.7 GB	Q8_0	Comfortable
112,208 tok/s 67,325–179,532 · low confidence	H100 CNX NVIDIA	80 GB	2,040 GB/s	Mar 2023	0.7 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: MIT-IBM Watson AI Lab,Massachusetts Institute of Technology (MIT),IBM
Organisation type: Academia,Industry,Academia,Industry
Country: United States of America
Published: 29 April 2020
Authors: Han Cai, Chuang Gan, Tianzhe Wang, Zhekai Zhang, and Song Han

What it does

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

Domain: Vision
Task: Image classification
Numerical format: FP32

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: 7.7M
Training data: 1,280,000 tokens
Epochs: 180

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: 6.2 × 10²⁰ FLOP
How it was established: Hardware

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 V100
Chip-hours: 4,200
Compute cost: $1,754
Cloud vendor: Amazon Web Services

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: 1,536

Sources

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

Reference: Once for all: Train one network and specialize it for efficient deployment.
Last updated: 25 May 2026

The extremes

The ten fastest GPUs that run Once for All

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 Once for All

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

0.7 GB

Fastest

440,031 tok/s

Once for All reaches a parameter count of 7.7M. 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 Q8_0 and producing around 4,788 tokens per second.

At the other end sits B200, generating roughly 440,031 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

Where it came from

Once for All was published by MIT-IBM Watson AI Lab,Massachusetts Institute of Technology (MIT),IBM, in the country recorded as United States of America, during April 2020. The publishing organisation is categorised as academia,Industry,Academia,Industry.

It works in the domain of Vision, and is recorded as performing the task of image classification.

The weights are published, so it can be downloaded and run on your own hardware indefinitely, offline, with no account attached.

Understanding the speeds

The median result is around 12,356.1 tokens per second. Producing text faster than most people read it: 818 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 6.2 × 10²⁰ FLOP, on hardware recorded as NVIDIA V100. That figure measures what producing the model cost, and has no bearing on how fast it answers.

It was trained on a corpus of about 1,280,000 tokens of text.

It is tracked in the underlying dataset for one reason in particular: sOTA improvement.

Step by step

How to choose a GPU for Once for All

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

Start from what it actually needs, which is the requirement of Once for All, needing around 0.7 GB at a compression of Q8_0. That figure, not the headline performance of a card, is what decides whether it runs.
02

Set the context length you will work at

The conversation occupies memory too, and grows as it goes. Set the slider to the length you expect, because at long context a card that handles short questions easily can be dropped by Once for All.
03

Choose how far you will compress it

The quantisation column varies by card, because a bigger card holds a more accurate copy, reaching a compression of Q8_0 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

Compare tokens per second, not specifications

Sort by speed to see how cards rank for Once for All. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B200, at 440,031 tok/s.
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 Once for All. 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

Following a card through to its own page shows every other model it can hold, which is the question that follows once you have settled on Once for All.

Answers

Once for All — common questions

Once for All— 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 0.7 GB and generating roughly 73,705 tokens per second. The fit is comfortable.

Once for All— 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.

Once for All— how many parameters does it have?

It has a parameter count of 7.7M. "Since all of these sub-networks share the same weights (i.e., Wo) (Cheung et al., 2019), we only require 7.7M parameters to store all of them. Without sharing, the total model size will be prohibitive". 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.

Once for All— who created it?

It was published by MIT-IBM Watson AI Lab,Massachusetts Institute of Technology (MIT),IBM, based in United States of America, an organisation categorised as academia,Industry,Academia,Industry.

Once for All— when was it released?

It was published in April 2020. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

Once for All— what is it used for?

It works in the domain of Vision, and is recorded as handling the task of image classification. These are the areas it was designed around; they describe intent rather than a hard boundary.

Once for All— 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.

Once for All— how much compute was used to train it?

Training consumed around 6.2 × 10²⁰ FLOP, on hardware recorded as NVIDIA V100. 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.

Once for All— 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.

Once for All— 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.

Once for All— why does the quantisation differ between cards?

Because capacity varies, so does how hard it has to be squeezed. The number of distinct levels in the table above: 1. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

Once for All— 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: 264,018–704,049 tok/s on B200. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

Once for All— 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 Q8_0 using about 0.7 GB, and produces roughly 4,788 tokens per second. The number of cards able to run it in total: 818.

Once for All— how fast is it on a GPU?

It depends on the card. The quickest we calculate is B200, at about 440,031 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: 818.

Once for All— how much VRAM does it need?

It needs about 0.7 GB at a compression of Q8_0, 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.

Once for All— 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 0.7 GB and generating roughly 81,956 tokens per second. The fit is comfortable.

Once for All— 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 0.7 GB and generating roughly 50,185 tokens per second. The fit is comfortable.

Once for All— 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 0.7 GB and generating roughly 62,154 tokens per second. The fit is comfortable.

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