Nemotron-4 340B TPS calculator

Open weights NVIDIA 340B parameters June 2024

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

6 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Radeon Instinct MI300X

192 GB · Q3_K_M · 14.0 tok/s

Fastest card

B300

17.8 tok/s · 288 GB

Which GPUs can run Nemotron-4 340B?

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.

6 cards match

Calculating
Needs Quantisation Fit
17.8 tok/s

11–28 · low confidence

B300 NVIDIA 288 GB 8,000 GB/s Sep 2025 246.0 GB Q5_K_M Tight
14.2 tok/s

9–23 · low confidence

Radeon Instinct MI350X AMD 288 GB 8,190 GB/s Jan 2025 246.0 GB Q5_K_M Tight
14.2 tok/s

9–23 · low confidence

Radeon Instinct MI355X AMD 288 GB 8,190 GB/s Jan 2025 246.0 GB Q5_K_M Tight
14.0 tok/s

8–22 · low confidence

Radeon Instinct MI300X AMD 192 GB 5,325 GB/s Dec 2023 166.8 GB Q3_K_M Tight
14.0 tok/s

8–22 · low confidence

Radeon Instinct MI308X AMD 192 GB 5,325 GB/s Dec 2023 166.8 GB Q3_K_M Tight
13.5 tok/s

8–22 · low confidence

Radeon Instinct MI325X AMD 256 GB 6,000 GB/s Oct 2024 206.4 GB Q4_K_M Tight

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
NVIDIA
Organisation type
Industry
Country
United States of America
Published
14 June 2024
Authors
Bo Adler, Niket Agarwal, Ashwath Aithal, Dong H. Anh, Pallab Bhattacharya, Annika Brundyn, Jared Casper, Bryan Catanzaro, Sharon Clay, Jonathan Cohen, Sirshak Das, Ayush Dattagupta, Olivier Delalleau, Leon Derczynski, Yi Dong, Daniel Egert, Ellie Evans, Aleksander Ficek, Denys Fridman, Shaona Ghosh, Boris Ginsburg, Igor Gitman, Tomasz Grzegorzek, Robert Hero, Jining Huang, Vibhu Jawa, Joseph Jenni…

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
BF16

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
340B

340B

Training data
9,000,000,000,000 tokens

9T training tokens. They first train on an 8T token dataset and then an additional 1T tokens, it's slightly unclear if that's more data or a partial second epoch 6.75T words using 1 token = 0.75 words

Batch size
9,437,184

2304 * 4096

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
1.8 × 10²⁵ FLOP

9 trillion tokens for training 6 * 340B * 9T = 1.8E25 alternatively, can do a hardware estimate with a few extra steps: According to the technical report, Nemotron-4 340B was trained using up to 6144 H100 GPUs. Helpfully, they also report the model FLOP utilization (MFU), which was 41-42% (Table 2). This is the ratio of the actual output of their GPUs, in FLOP used for training, relative to their theoretical max of 989 teraFLOP/s per GPU. Unfortunately, the report omits the last ingredient, w…

How it was established
Operation counting,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 H100 SXM5 80GB
Chips used
6,144
Wall-clock time
2,200 hours (91.7 days)

see training compute notes, this is an inferred estimate

Hardware utilisation
MFU 41.1%

Table 2 indicates MFU at different stages of training. ((42.4% * 200B) + (42.3% * 200B) + (41.0% * 7600B)) / (200B + 200B + 7600B) = 0.410675, averaged over training.

Power draw
8.5 MW
Compute cost
$21,271,018

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

Permissive commercial license: https://developer.download.nvidia.com/licenses/nvidia-open-model-license-agreement-june-2024.pdf

Hugging Face
nvidia

How it is classified

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

Frontier model
Yes
Likely above 10²³ FLOP
Yes
Why it is tracked
Training cost

~2e25 FLOP, so high training cost, likely >5M

Record confidence
Confident

Sources

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

Reference
NVIDIA Releases Open Synthetic Data Generation Pipeline for Training Large Language Models
Last updated
18 December 2025

What the numbers mean

What it takes to run this model

Minimum card

Radeon Instinct MI300X

Memory needed

166.8 GB

Fastest

17.8 tok/s

Nemotron-4 340B reaches a parameter count of 340B. That is beyond what any single graphics card holds. Running it means either splitting it across several cards or renting hardware built for the job, and every card able to hold it alone is a datacentre part. The number that can: 6.

The smallest card that holds it is Radeon Instinct MI300X, with a memory capacity of 192 GB, running it at a compression of Q3_K_M and producing around 14.0 tokens per second.

The fastest we calculate for it is B300, generating roughly 17.8 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

What this model is

Nemotron-4 340B was published by NVIDIA, in the country recorded as United States of America, during June 2024. The category the publisher falls under is 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. On Hugging Face it is published under the organisation nvidia.

What decides the speed

Half the cards that hold it manage more than 14.1 tokens per second. Producing text faster than most people read it: 6 of them.

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

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

The training run consumed about 1.8 × 10²⁵ FLOP, on hardware recorded as NVIDIA H100 SXM5 80GB. 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 9,000,000,000,000 tokens of text.

Its inclusion criterion: training cost.

Step by step

How to choose a GPU for Nemotron-4 340B

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

    Every card here has been checked against Nemotron-4 340B, needing around 166.8 GB at a compression of Q3_K_M. Capacity is the gate — a card either holds it or it does not.

  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 a card that seemed fine stops fitting Nemotron-4 340B.

  3. 03

    Set a quality floor

    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.

  4. 04

    Rank by throughput rather than spec sheet

    The speed ordering is effectively an ordering by memory bandwidth, for Nemotron-4 340B. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B300, at 17.8 tok/s.

  5. 05

    Look at the headroom, not just the fit

    A tight fit runs, but leaves nothing spare for a longer conversation, in the case of Nemotron-4 340B. 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

    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 Nemotron-4 340B.

Answers

Nemotron-4 340B — common questions

01

Nemotron-4 340B— 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 44.4 GB. Every figure here assumes the whole model is resident on the card.

02

Nemotron-4 340B— would two GPUs run it faster?

Capacity adds across cards; throughput does not. The number of cards already holding it on their own: 6. So a second card is rarely the answer here.

03

Nemotron-4 340B— 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: 3. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

04

Nemotron-4 340B— how accurate are these speed estimates?

They are calculated from specifications rather than measured, and each carries a range. One example: 11–28 tok/s on B300. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

05

Nemotron-4 340B— what GPU do I need to run it?

The smallest card in our catalogue that holds it is Radeon Instinct MI300X, with a memory capacity of 192 GB. It runs the model at a compression of Q3_K_M using about 166.8 GB, and produces roughly 14.0 tokens per second. The number of cards able to run it in total: 6.

06

Nemotron-4 340B— how fast is it on a GPU?

It depends on the card. The quickest we calculate is B300, at about 17.8 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: 6.

07

Nemotron-4 340B— how much VRAM does it need?

It needs about 166.8 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.

08

Nemotron-4 340B— 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.

09

Nemotron-4 340B— how many parameters does it have?

It has a parameter count of 340B. 340B. 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.

10

Nemotron-4 340B— who created it?

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

11

Nemotron-4 340B— when was it released?

It was published in June 2024. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

12

Nemotron-4 340B— 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.

13

Nemotron-4 340B— where can I download it?

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

14

Nemotron-4 340B— how much compute was used to train it?

Training consumed around 1.8 × 10²⁵ FLOP, on hardware recorded as NVIDIA H100 SXM5 80GB. 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.

Source

Original publication

Record last updated 18 December 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.