StarCoder TPS calculator

Open weights Hugging Face,ServiceNow,Northeastern University,Mila - Quebec AI (originally Montreal Institute for Learning Algorithms),Carnegie Mellon University (CMU),Johns Hopkins University,Leipzig University,ScaDS.AI,Queen Mary University of London,Roblox,Sea AI Lab,Technion - Israel Institute of Technology,Monash University,CSIRO,Data61,McGill University,Saama,University of British Columbia (UBC),Massachusetts Institute of Technology (MIT),Technical University of Munich,IBM,University of Vermont,UnfoldML,SAP,University of Notre Dame,Columbia University,New York University (NYU),University of Allahabad,Discover Dollar,Toloka,Telefonica,Stanford University,Weizmann Institute of Science,Alan Turing Institute,Wellesley College,EleutherAI,Forschungszentrum Julich 15.5B parameters May 2023

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

306 cards that can run it

818 cards we hold specifications for

Smallest card that fits

P102-101

10 GB · Q3_K_M · 20.1 tok/s

Fastest card

B200

219 tok/s · 180 GB

Which GPUs can run StarCoder?

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

306 cards match

Calculating

					Needs	Quantisation	Fit
219 tok/s 131–350 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	17.3 GB	Q8_0	Comfortable
219 tok/s 131–350 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	17.3 GB	Q8_0	Comfortable
175 tok/s 105–279 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	17.3 GB	Q8_0	Comfortable
175 tok/s 105–279 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	17.3 GB	Q8_0	Comfortable
140 tok/s 84–223 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	17.3 GB	Q8_0	Comfortable
134 tok/s 80–214 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	17.3 GB	Q8_0	Comfortable
134 tok/s 80–214 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	17.3 GB	Q8_0	Comfortable
128 tok/s 77–205 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	17.3 GB	Q8_0	Comfortable
115 tok/s 69–184 · low confidence	CMP 170HX 10 GB NVIDIA	10 GB	1,560 GB/s	Sep 2021	8.3 GB	Q3_K_M	Tight
113 tok/s 68–182 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	17.3 GB	Q8_0	Comfortable
113 tok/s 68–182 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	17.3 GB	Q8_0	Comfortable
113 tok/s 68–182 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	17.3 GB	Q8_0	Comfortable
108 tok/s 65–172 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	17.3 GB	Q8_0	Comfortable
91.8 tok/s 55–147 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	17.3 GB	Q8_0	Comfortable
91.8 tok/s 55–147 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	17.3 GB	Q8_0	Comfortable
91.8 tok/s 55–147 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	17.3 GB	Q8_0	Comfortable
91.8 tok/s 55–147 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	17.3 GB	Q8_0	Comfortable
91.8 tok/s 55–147 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	17.3 GB	Q8_0	Comfortable
69.9 tok/s 42–112 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	17.3 GB	Q8_0	Comfortable
69.9 tok/s 42–112 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	17.3 GB	Q8_0	Comfortable
58.3 tok/s 35–93 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	17.3 GB	Q8_0	Comfortable
57.6 tok/s 35–92 · low confidence	GeForce RTX 3080 12 GB NVIDIA	12 GB	912 GB/s	Jan 2022	10.1 GB	Q4_K_M	Tight
57.6 tok/s 35–92 · low confidence	GeForce RTX 3080 Ti NVIDIA	12 GB	912 GB/s	May 2021	10.1 GB	Q4_K_M	Tight
57.0 tok/s 34–91 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	17.3 GB	Q8_0	Comfortable
56.1 tok/s 34–90 · low confidence	CMP 90HX NVIDIA	10 GB	760 GB/s	Jul 2021	8.3 GB	Q3_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: Hugging Face,ServiceNow,Northeastern University,Mila - Quebec AI (originally Montreal Institute for Learning Algorithms),Carnegie Mellon University (CMU),Johns Hopkins University,Leipzig University,ScaDS.AI,Queen Mary University of London,Roblox,Sea AI Lab,Technion - Israel Institute of Technology,Monash University,CSIRO,Data61,McGill University,Saama,University of British Columbia (UBC),Massachusetts Institute of Technology (MIT),Technical University of Munich,IBM,University of Vermont,UnfoldML,SAP,University of Notre Dame,Columbia University,New York University (NYU),University of Allahabad,Discover Dollar,Toloka,Telefonica,Stanford University,Weizmann Institute of Science,Alan Turing Institute,Wellesley College,EleutherAI,Forschungszentrum Julich
Organisation type: Industry,Industry,Academia,Academia,Academia,Academia,Academia,Academia,Industry,Academia,Academia,Government,Government,Academia,Academia,Academia,Academia,Industry,Academia,Industry,Academia,Academia,Academia,Academia,Industry,Industry,Industry,Academia,Academia,Government,Academia,Research collective,Government
Country: United States of America, Canada, Germany, United Kingdom of Great Britain and Northern Ireland, Singapore, Israel, Australia, Sweden, India, Netherlands, Spain
Published: 9 May 2023
Authors: Raymond Li, Loubna Ben Allal, Yangtian Zi, Niklas Muennighoff, Denis Kocetkov, Chenghao Mou, Marc Marone, Christopher Akiki, Jia Li, Jenny Chim, Qian Liu, Evgenii Zheltonozhskii, Terry Yue Zhuo, Thomas Wang, Olivier Dehaene, Mishig Davaadorj, Joel Lamy-Poirier, João Monteiro, Oleh Shliazhko, Nicolas Gontier, Nicholas Meade, Armel Zebaze, Ming-Ho Yee, Logesh Kumar Umapathi, Jian Zhu, Benjamin Lipki…

What it does

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

Domain: Language
Task: Code generation
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: 15.5B
Training data: 203,750,000,000 tokens
Epochs: 1
Batch size: 4,000,000

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: 8.5 × 10²² FLOP
How it was established: Reported,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 A100 SXM4 80 GB
Chips used: 512
Chip-hours: 320,256
Wall-clock time: 626 hours (26.1 days)
Hardware utilisation: MFU 22.7%
Power draw: 408.0 kW
Compute cost: $212,218

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 (restricted use)
Training code: Unreleased

How it is classified

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

Foundation model: Yes
Why it is tracked: SOTA improvement
Record confidence: Confident
Citations: 1,181

Sources

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

Reference: StarCoder: may the source be with you!
Last updated: 25 May 2026

The extremes

The ten fastest GPUs that run StarCoder

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 StarCoder

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

P102-101

Memory needed

8.3 GB

Fastest

219 tok/s

StarCoder reaches a parameter count of 15.5B. 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: 306.

At the low end it is handled by P102-101, with a memory capacity of 10 GB, running it at a compression of Q3_K_M and producing around 20.1 tokens per second.

The quickest result comes from B200, generating roughly 219 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

Background

StarCoder was published by Hugging Face,ServiceNow,Northeastern University,Mila - Quebec AI (originally Montreal Institute for Learning Algorithms),Carnegie Mellon University (CMU),Johns Hopkins University,Leipzig University,ScaDS.AI,Queen Mary University of London,Roblox,Sea AI Lab,Technion - Israel Institute of Technology,Monash University,CSIRO,Data61,McGill University,Saama,University of British Columbia (UBC),Massachusetts Institute of Technology (MIT),Technical University of Munich,IBM,University of Vermont,UnfoldML,SAP,University of Notre Dame,Columbia University,New York University (NYU),University of Allahabad,Discover Dollar,Toloka,Telefonica,Stanford University,Weizmann Institute of Science,Alan Turing Institute,Wellesley College,EleutherAI,Forschungszentrum Julich, in the country recorded as United States of America, during May 2023. The category the publisher falls under is industry,Industry,Academia,Academia,Academia,Academia,Academia,Academia,Industry,Academia,Academia,Government,Government,Academia,Academia,Academia,Academia,Industry,Academia,Industry,Academia,Academia,Academia,Academia,Industry,Industry,Industry,Academia,Academia,Government,Academia,Research collective,Government.

It works in the domain of Language, and is recorded as performing the task of code generation.

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

Reading the throughput figures

Half the cards that hold it manage more than 20.4 tokens per second. Producing text faster than most people read it: 261 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.

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.

How it was trained

Producing it required arithmetic totalling around 8.5 × 10²² FLOP, on hardware recorded as NVIDIA A100 SXM4 80 GB. That figure measures what producing the model cost, and has no bearing on how fast it answers.

Training consumed a corpus of around 203,750,000,000 tokens of text.

Its inclusion criterion: sOTA improvement.

Step by step

How to choose a GPU for StarCoder

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

01

Read the memory figure first

Start from what it actually needs, which is the requirement of StarCoder, needing around 8.3 GB at a compression of Q3_K_M. No amount of processing power compensates for a card that cannot hold it.
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 StarCoder.
03

Decide how much compression you will accept

The quantisation column varies by card, because a bigger card holds a more accurate copy, 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

Sort by speed

Ranking by tokens per second follows memory bandwidth rather than core counts, for StarCoder. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B200, at 219 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 StarCoder. 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

Check the card from the other side

Each card page repeats this sweep for every model we hold, answering what else the hardware is good for beyond StarCoder.

Answers

StarCoder — common questions

StarCoder— what is it used for?

It works in the domain of Language, and is recorded as handling the task of code generation. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

StarCoder— 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.

StarCoder— how much compute was used to train it?

Training consumed around 8.5 × 10²² FLOP, on hardware recorded as NVIDIA A100 SXM4 80 GB. 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.

StarCoder— 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 2.9 GB. Every figure here assumes the whole model is resident on the card.

StarCoder— would two GPUs run it faster?

Capacity adds across cards; throughput does not. The number of cards already holding it on their own: 306. So a second card is rarely the answer here.

StarCoder— why does the quantisation differ between cards?

A larger card holds a more accurate copy. The number of compression levels used across the cards that run it: 5. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

StarCoder— 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: 131–350 tok/s on B200. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

StarCoder— what GPU do I need to run it?

The smallest card in our catalogue that holds it is P102-101, with a memory capacity of 10 GB. It runs the model at a compression of Q3_K_M using about 8.3 GB, and produces roughly 20.1 tokens per second. The number of cards able to run it in total: 306.

StarCoder— how fast is it on a GPU?

It depends on the card. The quickest we calculate is B200, at about 219 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: 261.

StarCoder— how much VRAM does it need?

It needs about 8.3 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.

StarCoder— 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 Q4_K_M, using about 10.1 GB and generating roughly 57.6 tokens per second. The fit is tight.

StarCoder— 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 Q6_K, using about 13.7 GB and generating roughly 44.9 tokens per second. The fit is tight.

StarCoder— 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 17.3 GB and generating roughly 36.6 tokens per second. The fit is comfortable.

StarCoder— 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.

StarCoder— how many parameters does it have?

It has a parameter count of 15.5B. "We trained a 15.5B parameter model". 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.

StarCoder— who created it?

It was published by Hugging Face,ServiceNow,Northeastern University,Mila - Quebec AI (originally Montreal Institute for Learning Algorithms),Carnegie Mellon University (CMU),Johns Hopkins University,Leipzig University,ScaDS.AI,Queen Mary University of London,Roblox,Sea AI Lab,Technion - Israel Institute of Technology,Monash University,CSIRO,Data61,McGill University,Saama,University of British Columbia (UBC),Massachusetts Institute of Technology (MIT),Technical University of Munich,IBM,University of Vermont,UnfoldML,SAP,University of Notre Dame,Columbia University,New York University (NYU),University of Allahabad,Discover Dollar,Toloka,Telefonica,Stanford University,Weizmann Institute of Science,Alan Turing Institute,Wellesley College,EleutherAI,Forschungszentrum Julich, based in United States of America, an organisation categorised as industry,Industry,Academia,Academia,Academia,Academia,Academia,Academia,Industry,Academia,Academia,Government,Government,Academia,Academia,Academia,Academia,Industry,Academia,Industry,Academia,Academia,Academia,Academia,Industry,Industry,Industry,Academia,Academia,Government,Academia,Research collective,Government.

StarCoder— when was it released?

It was published in May 2023. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

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