MnasNet-A1 + SSDLite TPS calculator

Open weights Google 4.9M parameters May 2019

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 · 7,523 tok/s

Fastest card

B200

691,477 tok/s · 180 GB

Which GPUs can run MnasNet-A1 + SSDLite?

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
691,477 tok/s 414,886–1,106,363 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	0.7 GB	Q8_0	Comfortable
691,477 tok/s 414,886–1,106,363 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	0.7 GB	Q8_0	Comfortable
552,161 tok/s 331,297–883,458 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	0.7 GB	Q8_0	Comfortable
552,161 tok/s 331,297–883,458 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	0.7 GB	Q8_0	Comfortable
441,594 tok/s 264,957–706,551 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	0.7 GB	Q8_0	Comfortable
422,665 tok/s 253,599–676,264 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	0.7 GB	Q8_0	Comfortable
422,665 tok/s 253,599–676,264 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	0.7 GB	Q8_0	Comfortable
404,514 tok/s 242,708–647,222 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	0.7 GB	Q8_0	Comfortable
359,006 tok/s 215,404–574,410 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	0.7 GB	Q8_0	Comfortable
359,006 tok/s 215,404–574,410 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	0.7 GB	Q8_0	Comfortable
359,006 tok/s 215,404–574,410 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	0.7 GB	Q8_0	Comfortable
340,552 tok/s 204,331–544,884 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
290,420 tok/s 174,252–464,672 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
290,420 tok/s 174,252–464,672 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	0.7 GB	Q8_0	Comfortable
290,420 tok/s 174,252–464,672 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
290,420 tok/s 174,252–464,672 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
290,420 tok/s 174,252–464,672 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	0.7 GB	Q8_0	Comfortable
221,134 tok/s 132,681–353,815 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	0.7 GB	Q8_0	Comfortable
221,134 tok/s 132,681–353,815 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	0.7 GB	Q8_0	Comfortable
184,279 tok/s 110,567–294,846 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	0.7 GB	Q8_0	Comfortable
180,346 tok/s 108,207–288,553 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	0.7 GB	Q8_0	Comfortable
176,327 tok/s 105,796–282,122 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	0.7 GB	Q8_0	Comfortable
176,327 tok/s 105,796–282,122 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	0.7 GB	Q8_0	Comfortable
176,327 tok/s 105,796–282,122 · low confidence	A800 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Aug 2022	0.7 GB	Q8_0	Comfortable
176,327 tok/s 105,796–282,122 · 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: Google
Organisation type: Industry
Country: United States of America
Published: 29 May 2019
Authors: Mingxing Tan, Bo Chen, Ruoming Pang, Vijay Vasudevan, Mark Sandler, Andrew Howard, Quoc V. Le

What it does

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

Domain: Vision
Task: Image classification, Object detection, Neural Architecture Search - NAS
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: 4.9M
Training data: tokens

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: 1.5 × 10²¹ 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: Google TPU v3
Chips used: 256
Wall-clock time: 108 hours
Power draw: 237.0 kW
Compute cost: $9,552

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.

Record confidence: Speculative
Citations: 3,396

Sources

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

Reference: MnasNet: Platform-Aware Neural Architecture Search for Mobile
Last updated: 25 May 2026

The extremes

The ten fastest GPUs that run MnasNet-A1 + SSDLite

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 MnasNet-A1 + SSDLite

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

691,477 tok/s

MnasNet-A1 + SSDLite is small enough at 4.9M parameters that hardware is rarely the obstacle — 818 of the cards we track can run it, including cards several years old.

The smallest card that holds it is the Tesla C1080 with 4 GB, running it at Q8_0 and producing around 7,523 tokens per second.

Top of the range is the B200, at roughly 691,477 tokens per second thanks to 8,000 GB/s of bandwidth.

Where it came from

MnasNet-A1 + SSDLite was published by Google, in United States of America, in May 2019. industry is the category the publisher falls under.

It works in Vision, and is recorded as doing image classification, Object detection, Neural Architecture Search - NAS.

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.

Understanding the speeds

Half the cards that hold it manage more than 19,416.7 tokens per second, and 818 exceed reading speed outright.

Every weight participates in every token here, so bandwidth is the whole story: the ranking below is effectively a ranking of memory throughput.

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.

What went into building it

The training run consumed about 1.5 × 10²¹ FLOP, on Google TPU v3. That figure describes the cost of creating it and has no bearing on how quickly it generates text.

Step by step

How to choose a GPU for MnasNet-A1 + SSDLite

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

Every card here has been checked against MnasNet-A1 + SSDLite — around 0.7 GB at Q8_0. Capacity is the gate — a card either holds it or it does not.
02

Match the context to your actual use

The conversation occupies memory too, and grows as it goes. Set the slider to the length you expect: at long context MnasNet-A1 + SSDLite can slip off a card that handles short questions easily.
03

Decide how much compression you will accept

Each card runs the least-compressed copy it can hold — Q8_0 on the smallest card that fits. Setting a floor drops the cards that only manage MnasNet-A1 + SSDLite by squeezing it further than you would want.
04

Sort by speed

Ranking by tokens per second for MnasNet-A1 + SSDLite follows memory bandwidth, not core counts, which is why the B200 tops it at 691,477 tok/s.
05

Read the fit column last

Tight means MnasNet-A1 + SSDLite 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.
06

Check the card from the other side

Each card page repeats this sweep for every model we hold. It answers what else the hardware is good for, beyond MnasNet-A1 + SSDLite.

Answers

MnasNet-A1 + SSDLite — common questions

Can I run MnasNet-A1 + SSDLite on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 0.7 GB and generating roughly 115,822 tokens per second — a comfortable fit.

Is MnasNet-A1 + SSDLite open source?

Its weights are published, so MnasNet-A1 + SSDLite 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 MnasNet-A1 + SSDLite have?

MnasNet-A1 + SSDLite has 4.9M parameters. From https://arxiv.org/pdf/1807.11626.pdf. 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 MnasNet-A1 + SSDLite?

MnasNet-A1 + SSDLite was published by Google, based in United States of America, categorised as industry.

When was MnasNet-A1 + SSDLite released?

MnasNet-A1 + SSDLite was published in May 2019. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

What is MnasNet-A1 + SSDLite used for?

MnasNet-A1 + SSDLite works in Vision, and is recorded as handling image classification, Object detection, Neural Architecture Search - NAS. 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 MnasNet-A1 + SSDLite?

The weights for MnasNet-A1 + SSDLite are published, though we do not hold a repository link for it. This site calculates hardware requirements rather than hosting model files.

How much compute was used to train MnasNet-A1 + SSDLite?

Around 1.5 × 10²¹ FLOP, on Google TPU v3. 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 MnasNet-A1 + SSDLite if it does not fit in my GPU?

It can be split between the card and system memory, but MnasNet-A1 + SSDLite generates painfully slowly that way. Nothing on this page assumes offloading.

Would two GPUs run MnasNet-A1 + SSDLite faster?

A second card roughly doubles the memory available but not the generation rate. With 818 cards already able to run MnasNet-A1 + SSDLite alone, the case for pairing is weak.

Why does the quantisation differ between cards for MnasNet-A1 + SSDLite?

A larger card holds a more accurate copy. Across the cards that run MnasNet-A1 + SSDLite, 1 compression levels are used; the floor control above pins it to one.

How accurate are these MnasNet-A1 + SSDLite speed estimates?

These are estimates with real error bars. The fastest result here, 414,886–1,106,363 tok/s on the B200, could reasonably land anywhere in its published range depending on which runtime you use.

What GPU do I need to run MnasNet-A1 + SSDLite?

The smallest card in our catalogue that holds MnasNet-A1 + SSDLite is the Tesla C1080, with 4 GB of memory. It runs the model at Q8_0 using about 0.7 GB, and produces roughly 7,523 tokens per second. 818 cards in total can run it.

How fast is MnasNet-A1 + SSDLite on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 691,477 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 818 of the cards that can run MnasNet-A1 + SSDLite clear that.

How much VRAM does MnasNet-A1 + SSDLite need?

About 0.7 GB at Q8_0 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 MnasNet-A1 + SSDLite on a 8 GB GPU?

Yes. A CMP 170HX 8 GB with 8 GB runs it at Q8_0, using about 0.7 GB and generating roughly 128,788 tokens per second — a comfortable fit.

Can I run MnasNet-A1 + SSDLite on a 12 GB GPU?

Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q8_0, using about 0.7 GB and generating roughly 78,863 tokens per second — a comfortable fit.

Can I run MnasNet-A1 + SSDLite on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q8_0, using about 0.7 GB and generating roughly 97,671 tokens per second — a comfortable fit.

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