C2S-Scale TPS calculator

Open weights Google Research,Yale University,Google DeepMind,Brown University,University of Southern California 27B parameters October 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

241 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Xeon Phi 7120P

16 GB · Q3_K_M · 9.7 tok/s

Fastest card

B200

125 tok/s · 180 GB

Which GPUs can run C2S-Scale?

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

241 cards match

Calculating

					Needs	Quantisation	Fit
125 tok/s 75–201 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	29.6 GB	Q8_0	Comfortable
125 tok/s 75–201 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	29.6 GB	Q8_0	Comfortable
100 tok/s 60–160 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	29.6 GB	Q8_0	Comfortable
100 tok/s 60–160 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	29.6 GB	Q8_0	Comfortable
80.1 tok/s 48–128 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	29.6 GB	Q8_0	Comfortable
76.7 tok/s 46–123 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	29.6 GB	Q8_0	Comfortable
76.7 tok/s 46–123 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	29.6 GB	Q8_0	Comfortable
73.4 tok/s 44–117 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	29.6 GB	Q8_0	Comfortable
65.2 tok/s 39–104 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	29.6 GB	Q8_0	Comfortable
65.2 tok/s 39–104 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	29.6 GB	Q8_0	Comfortable
65.2 tok/s 39–104 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	29.6 GB	Q8_0	Comfortable
61.8 tok/s 37–99 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	29.6 GB	Q8_0	Comfortable
52.7 tok/s 32–84 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	29.6 GB	Q8_0	Comfortable
52.7 tok/s 32–84 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	29.6 GB	Q8_0	Comfortable
52.7 tok/s 32–84 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	29.6 GB	Q8_0	Comfortable
52.7 tok/s 32–84 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	29.6 GB	Q8_0	Comfortable
52.7 tok/s 32–84 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	29.6 GB	Q8_0	Comfortable
47.8 tok/s 29–77 · low confidence	Tesla V100 SXM2 16 GB NVIDIA	16 GB	1,130 GB/s	Nov 2019	13.9 GB	Q3_K_M	Tight
42.6 tok/s 26–68 · low confidence	DRIVE A100 PROD NVIDIA	32 GB	1,870 GB/s	May 2020	23.3 GB	Q6_K	Comfortable
42.6 tok/s 26–68 · low confidence	GRID A100A NVIDIA	32 GB	1,870 GB/s	May 2020	23.3 GB	Q6_K	Comfortable
40.8 tok/s 24–65 · low confidence	GeForce RTX 5090 NVIDIA	32 GB	1,790 GB/s	Jan 2025	23.3 GB	Q6_K	Comfortable
40.8 tok/s 24–65 · low confidence	GeForce RTX 5090 D NVIDIA	32 GB	1,790 GB/s	Jan 2025	23.3 GB	Q6_K	Comfortable
40.6 tok/s 24–65 · low confidence	GeForce RTX 5080 NVIDIA	16 GB	960 GB/s	Jan 2025	13.9 GB	Q3_K_M	Tight
40.1 tok/s 24–64 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	29.6 GB	Q8_0	Comfortable
40.1 tok/s 24–64 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	29.6 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 Research,Yale University,Google DeepMind,Brown University,University of Southern California
Organisation type: Industry,Academia,Industry,Academia,Academia
Country: United States of America
Published: 11 October 2025
Authors: Syed Asad Rizvi, Daniel Levine, Aakash Patel, Shiyang Zhang, Eric Wang, Curtis Jamison Perry, Nicole Mayerli Constante, Sizhuang He, David Zhang, Cerise Tang, Zhuoyang Lyu, Rayyan Darji, Chang Li, Emily Sun, David Jeong, Lawrence Zhao, Jennifer Kwan, David Braun, Brian Hafler, Hattie Chung, Rahul M. Dhodapkar, Bryan Perozzi, Jeffrey Ishizuka, Shekoofeh Azizi, David van Dijk

What it does

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

Domain: Language, Biology
Task: Cell Biology, RNA sequence generation, Language modeling/generation, Question answering
Base model: Gemma 2 27B

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: 27B
Training data: 1,000,000,000 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.

How it was established: Operation counting
Fine-tuning compute: 1.6 × 10²⁰ FLOP

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 v4
Chips used: 256
Power draw: 170.0 kW

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
Hugging Face: vandijklab

How it is classified

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

Record confidence: Likely

Sources

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

Reference: Scaling Large Language Models for Next-Generation Single-Cell Analysis
Last updated: 28 November 2025

The extremes

The ten fastest GPUs that run C2S-Scale

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 C2S-Scale

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 you need to run it

Minimum card

Xeon Phi 7120P

Memory needed

13.9 GB

Fastest

125 tok/s

C2S-Scale reaches a parameter count of 27B. That lands in the range a serious desktop card can handle once the weights are compressed. The number of cards we track that can run it: 241.

The least hardware that works is Xeon Phi 7120P, with a memory capacity of 16 GB, running it at a compression of Q3_K_M and producing around 9.7 tokens per second.

Top of the range is B200, generating roughly 125 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

About this model

C2S-Scale was published by Google Research,Yale University,Google DeepMind,Brown University,University of Southern California, in the country recorded as United States of America, during October 2025. It comes out of an organisation categorised as industry,Academia,Industry,Academia,Academia.

It works in the domain of Language, Biology, and is recorded as performing the task of cell Biology, RNA sequence generation, Language modeling/generation, Question answering.

Rather than being trained from scratch, it is derived from Gemma 2 27B. That is why it shares the base model's general shape and size.

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. On Hugging Face it is published under the organisation vandijklab.

How fast it runs, and why

Half the cards that hold it manage more than 19.0 tokens per second. Clearing the ten tokens per second that roughly matches reading speed: 196 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.

How it was trained

The training set ran to roughly 1,000,000,000 tokens of text.

Step by step

How to choose a GPU for C2S-Scale

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 C2S-Scale, needing around 13.9 GB at a compression of Q3_K_M. That figure, not the headline performance of a card, is what decides whether it runs.
02

Decide how long your conversations run

Longer conversations cost memory on top of what the weights need. Move the slider to your real working length before trusting any row for C2S-Scale.
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 C2S-Scale. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B200, at 125 tok/s.
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 C2S-Scale. 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

See what else that card runs

Each card page repeats this sweep for every model we hold, answering what else the hardware is good for beyond C2S-Scale.

Answers

C2S-Scale — common questions

C2S-Scale— how much VRAM does it need?

It needs about 13.9 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.

C2S-Scale— 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 Q3_K_M, using about 13.9 GB and generating roughly 47.8 tokens per second. The fit is tight.

C2S-Scale— 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 Q5_K_M, using about 20.2 GB and generating roughly 37.5 tokens per second. The fit is tight.

C2S-Scale— 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.

C2S-Scale— how many parameters does it have?

It has a parameter count of 27B. 27B. 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.

C2S-Scale— who created it?

It was published by Google Research,Yale University,Google DeepMind,Brown University,University of Southern California, based in United States of America, an organisation categorised as industry,Academia,Industry,Academia,Academia.

C2S-Scale— when was it released?

It was published in October 2025.

C2S-Scale— what is it used for?

It works in the domain of Language, Biology, and is recorded as handling the task of cell Biology, RNA sequence generation, 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.

C2S-Scale— where can I download it?

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

C2S-Scale— 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 6.2 GB. Every figure here assumes the whole model is resident on the card.

C2S-Scale— would two GPUs run it faster?

A second card roughly doubles the memory available but not the generation rate. The number already able to run it alone: 241. So a second card is rarely the answer here.

C2S-Scale— 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: 5. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

C2S-Scale— 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: 75–201 tok/s on B200. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

C2S-Scale— what GPU do I need to run it?

The smallest card in our catalogue that holds it is Xeon Phi 7120P, with a memory capacity of 16 GB. It runs the model at a compression of Q3_K_M using about 13.9 GB, and produces roughly 9.7 tokens per second. The number of cards able to run it in total: 241.

C2S-Scale— how fast is it on a GPU?

It depends on the card. The quickest we calculate is B200, at about 125 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: 196.

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