CollabLLM TPS calculator

Open weights Stanford University,Microsoft,Georgia Institute of Technology 8B parameters June 2025

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 · IQ4_XS · 15.9 tok/s

Fastest card

B200

424 tok/s · 180 GB

Which GPUs can run CollabLLM?

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
424 tok/s 254–678 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	9.3 GB	Q8_0	Comfortable
424 tok/s 254–678 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	9.3 GB	Q8_0	Comfortable
338 tok/s 203–541 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	9.3 GB	Q8_0	Comfortable
338 tok/s 203–541 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	9.3 GB	Q8_0	Comfortable
270 tok/s 162–433 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	9.3 GB	Q8_0	Comfortable
259 tok/s 155–414 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	9.3 GB	Q8_0	Comfortable
259 tok/s 155–414 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	9.3 GB	Q8_0	Comfortable
248 tok/s 149–396 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	9.3 GB	Q8_0	Comfortable
220 tok/s 132–352 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	9.3 GB	Q8_0	Comfortable
220 tok/s 132–352 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	9.3 GB	Q8_0	Comfortable
220 tok/s 132–352 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	9.3 GB	Q8_0	Comfortable
209 tok/s 125–334 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	9.3 GB	Q8_0	Comfortable
178 tok/s 107–285 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	9.3 GB	Q8_0	Comfortable
178 tok/s 107–285 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	9.3 GB	Q8_0	Comfortable
178 tok/s 107–285 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	9.3 GB	Q8_0	Comfortable
178 tok/s 107–285 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	9.3 GB	Q8_0	Comfortable
178 tok/s 107–285 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	9.3 GB	Q8_0	Comfortable
141 tok/s 85–225 · low confidence	CMP 170HX 8 GB NVIDIA	8 GB	1,490 GB/s	Sep 2021	6.5 GB	Q5_K_M	Tight
135 tok/s 81–217 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	9.3 GB	Q8_0	Comfortable
135 tok/s 81–217 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	9.3 GB	Q8_0	Comfortable
120 tok/s 72–192 · low confidence	CMP 170HX 10 GB NVIDIA	10 GB	1,560 GB/s	Sep 2021	7.4 GB	Q6_K	Comfortable
113 tok/s 68–181 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	9.3 GB	Q8_0	Comfortable
110 tok/s 66–177 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	9.3 GB	Q8_0	Comfortable
108 tok/s 65–173 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	9.3 GB	Q8_0	Comfortable
108 tok/s 65–173 · low confidence	A100X NVIDIA	80 GB	2,040 GB/s	Jun 2021	9.3 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: Stanford University,Microsoft,Georgia Institute of Technology
Organisation type: Academia,Industry,Academia
Country: United States of America
Published: 12 June 2025
Authors: Shirley Wu, Michel Galley, Baolin Peng, Hao Cheng, Gavin Li, Yao Dou, Weixin Cai, James Zou, Jure Leskovec, Jianfeng Gao

What it does

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

Domain: Language
Task: Language modeling/generation, Question answering
Base model: Llama 3.1-8B

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: 8B
Training data: tokens

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

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 (non-commercial)
Training code: Open source
Hugging Face: collabllm

How it is classified

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

Record confidence: Confident

Sources

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

Reference: CollabLLM: From Passive Responders to Active Collaborators
Last updated: 28 November 2025

The extremes

The ten fastest GPUs that run CollabLLM

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 CollabLLM

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

5.1 GB

Fastest

424 tok/s

CollabLLM is small enough at 8B parameters that hardware is rarely the obstacle — 582 of the cards we track can run it, including cards several years old.

At the low end, a Quadro 6000 handles it — 6 GB, at IQ4_XS, for about 15.9 tokens per second.

The quickest result comes from a B200 at around 424 tokens per second — its 8,000 GB/s of bandwidth is what buys that.

Where it came from

CollabLLM was published by Stanford University,Microsoft,Georgia Institute of Technology, in United States of America, in June 2025. It comes out of academia,Industry,Academia.

It works in Language, and is recorded as doing language modeling/generation, Question answering.

Its starting point was Llama 3.1-8B — most models at this scale are adapted from an existing base rather than built from nothing.

The weights being open is what puts this page in the calculator rather than only in the catalogue: it is a model you can actually hold. It is published under the collabllm organisation on Hugging Face.

Understanding the speeds

The median result is around 23.8 tokens per second; 551 cards produce text faster than most people read it.

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.

Step by step

How to choose a GPU for CollabLLM

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

The table lists every card that can hold CollabLLM — around 5.1 GB at IQ4_XS. That figure, not the card's headline performance, is what decides whether it runs.
02

Set the context length you will work at

Longer conversations cost memory on top of what the weights need. Move the slider to your real working length before trusting any row for CollabLLM.
03

Set a quality floor

Each card runs the least-compressed copy it can hold — IQ4_XS on the smallest card that fits. Setting a floor drops the cards that only manage CollabLLM by squeezing it further than you would want.
04

Sort by speed

Sort by speed to see how cards rank for CollabLLM. It will not match a gaming ordering — generation is bound by memory bandwidth, which is why the B200 tops it at 424 tok/s.
05

Look at the headroom, not just the fit

The fit column separates cards that just manage CollabLLM from those with room to spare. Buy for the second if the context might grow.
06

See what else that card runs

Following a card through to its own page shows every other model it can hold, which is the question that follows once CollabLLM is settled.

Answers

CollabLLM — common questions

What is CollabLLM used for?

CollabLLM works in Language, and is recorded as handling language modeling/generation, Question answering. 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 CollabLLM?

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

Can I run CollabLLM 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 CollabLLM is rarely worth using — the nearest miss we calculate is short by 1.0 GB. Every figure here assumes the whole model is on the card.

Would two GPUs run CollabLLM faster?

Two cards buy memory rather than speed. That matters for CollabLLM only if one card cannot hold it — 582 can, so a second adds little.

Why does the quantisation differ between cards for CollabLLM?

Because capacity varies, so does how hard CollabLLM has to be squeezed — 4 distinct levels appear in the table above. Set a minimum quality to compare at one.

How accurate are these CollabLLM speed estimates?

These are estimates with real error bars. The fastest result here, 254–678 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 CollabLLM?

The smallest card in our catalogue that holds CollabLLM is the Quadro 6000, with 6 GB of memory. It runs the model at IQ4_XS using about 5.1 GB, and produces roughly 15.9 tokens per second. 582 cards in total can run it.

How fast is CollabLLM on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 424 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 551 of the cards that can run CollabLLM clear that.

How much VRAM does CollabLLM need?

About 5.1 GB at IQ4_XS 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 CollabLLM on a 8 GB GPU?

Yes. A CMP 170HX 8 GB with 8 GB runs it at Q5_K_M, using about 6.5 GB and generating roughly 141 tokens per second — a tight fit.

Can I run CollabLLM on a 12 GB GPU?

Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q8_0, using about 9.3 GB and generating roughly 48.3 tokens per second — a tight fit.

Can I run CollabLLM on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q8_0, using about 9.3 GB and generating roughly 59.8 tokens per second — a comfortable fit.

Can I run CollabLLM on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 9.3 GB and generating roughly 70.9 tokens per second — a comfortable fit.

Is CollabLLM open source?

Its weights are published, so CollabLLM 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 CollabLLM have?

CollabLLM has 8B parameters. same as the base 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.

Who created CollabLLM?

CollabLLM was published by Stanford University,Microsoft,Georgia Institute of Technology, based in United States of America, categorised as academia,Industry,Academia.

When was CollabLLM released?

CollabLLM was published in June 2025.

Source

Original publication

Record last updated 28 November 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.

All AI models