ruDalle: Kandinsky 3.0 TPS calculator

Open weights Sber 11.9B parameters December 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

509 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Xeon Phi 5110P

8 GB · Q3_K_M · 20.0 tok/s

Fastest card

B200

285 tok/s · 180 GB

Which GPUs can run ruDalle: Kandinsky 3.0?

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

509 cards match

Calculating

					Needs	Quantisation	Fit
285 tok/s 171–456 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	13.4 GB	Q8_0	Comfortable
285 tok/s 171–456 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	13.4 GB	Q8_0	Comfortable
227 tok/s 136–364 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	13.4 GB	Q8_0	Comfortable
227 tok/s 136–364 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	13.4 GB	Q8_0	Comfortable
182 tok/s 109–291 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	13.4 GB	Q8_0	Comfortable
174 tok/s 104–278 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	13.4 GB	Q8_0	Comfortable
174 tok/s 104–278 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	13.4 GB	Q8_0	Comfortable
167 tok/s 100–267 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	13.4 GB	Q8_0	Comfortable
148 tok/s 89–237 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	13.4 GB	Q8_0	Comfortable
148 tok/s 89–237 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	13.4 GB	Q8_0	Comfortable
148 tok/s 89–237 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	13.4 GB	Q8_0	Comfortable
143 tok/s 86–229 · low confidence	CMP 170HX 8 GB NVIDIA	8 GB	1,490 GB/s	Sep 2021	6.5 GB	Q3_K_M	Tight
140 tok/s 84–224 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	13.4 GB	Q8_0	Comfortable
128 tok/s 77–205 · low confidence	CMP 170HX 10 GB NVIDIA	10 GB	1,560 GB/s	Sep 2021	7.9 GB	Q4_K_M	Tight
120 tok/s 72–191 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	13.4 GB	Q8_0	Comfortable
120 tok/s 72–191 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	13.4 GB	Q8_0	Comfortable
120 tok/s 72–191 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	13.4 GB	Q8_0	Comfortable
120 tok/s 72–191 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	13.4 GB	Q8_0	Comfortable
120 tok/s 72–191 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	13.4 GB	Q8_0	Comfortable
91.1 tok/s 55–146 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	13.4 GB	Q8_0	Comfortable
91.1 tok/s 55–146 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	13.4 GB	Q8_0	Comfortable
75.9 tok/s 46–121 · low confidence	Data Center GPU Max 1550 Intel	128 GB	3,280 GB/s	Jan 2023	13.4 GB	Q8_0	Comfortable
74.3 tok/s 45–119 · low confidence	Data Center GPU Max Subsystem Intel	128 GB	3,210 GB/s	Jan 2023	13.4 GB	Q8_0	Comfortable
73.8 tok/s 44–118 · low confidence	RTX A5000-8Q NVIDIA	8 GB	768 GB/s	Apr 2021	6.5 GB	Q3_K_M	Tight
72.6 tok/s 44–116 · low confidence	A100 SXM4 80 GB NVIDIA	80 GB	2,040 GB/s	Nov 2020	13.4 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: Sber
Organisation type: Industry,Government
Country: Russia
Published: 11 December 2023
Authors: Vladimir Arkhipkin, Andrei Filatov, Viacheslav Vasilev, Anastasia Maltseva, Said Azizov, Igor Pavlov, Julia Agafonova, Andrey Kuznetsov, Denis Dimitrov

What it does

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

Domain: Image generation
Task: Text-to-image, Image generation

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: 11.9B
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: 2 × 10²⁰ FLOP
How it was established: Operation counting

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

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: ai-forever

How it is classified

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

Record confidence: Speculative

Sources

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

Reference: KANDINSKY 3.0 TECHNICAL REPORT
Last updated: 28 November 2025

The extremes

The ten fastest GPUs that run ruDalle: Kandinsky 3.0

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 ruDalle: Kandinsky 3.0

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

Xeon Phi 5110P

Memory needed

6.5 GB

Fastest

285 tok/s

ruDalle: Kandinsky 3.0 is small enough at 11.9B parameters that hardware is rarely the obstacle — 509 of the cards we track can run it, including cards several years old.

At the low end, a Xeon Phi 5110P handles it — 8 GB, at Q3_K_M, for about 20.0 tokens per second.

A B200 is the fastest we calculate for it: about 285 tokens per second, from 8,000 GB/s of memory bandwidth.

About this model

ruDalle: Kandinsky 3.0 was published by Sber, in Russia, in December 2023. It comes out of industry,Government.

It works in Image generation, and is recorded as doing text-to-image, Image generation.

The weights are published, so it can be downloaded and run on your own hardware indefinitely, offline, with no account attached. It is published under the ai-forever organisation on Hugging Face.

How fast it runs, and why

Half the cards that hold it manage more than 20.9 tokens per second, and 455 exceed reading speed outright.

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.

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.

Training and provenance

Producing it required around 2 × 10²⁰ FLOP of arithmetic, on NVIDIA A100, which is a statement about the training budget rather than about inference.

Step by step

How to choose a GPU for ruDalle: Kandinsky 3.0

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 ruDalle: Kandinsky 3.0 — around 6.5 GB at Q3_K_M. That figure, not the card's headline performance, is what decides whether it runs.
02

Decide how long your conversations run

The conversation occupies memory too, and grows as it goes. Set the slider to the length you expect: at long context ruDalle: Kandinsky 3.0 can slip off a card that handles short questions easily.
03

Decide how much compression you will accept

Compression is what makes ruDalle: Kandinsky 3.0 fit smaller cards, at some cost in accuracy — Q3_K_M on the smallest card that fits. A minimum quality removes the ones that go too far.
04

Compare tokens per second, not specifications

Sort by speed to see how cards rank for ruDalle: Kandinsky 3.0. It will not match a gaming ordering — generation is bound by memory bandwidth, which is why the B200 tops it at 285 tok/s.
05

Read the fit column last

Tight means ruDalle: Kandinsky 3.0 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

Open the card you have settled on

Each card page repeats this sweep for every model we hold. It answers what else the hardware is good for, beyond ruDalle: Kandinsky 3.0.

Answers

ruDalle: Kandinsky 3.0 — common questions

Would two GPUs run ruDalle: Kandinsky 3.0 faster?

Capacity adds across cards; throughput does not. Since 509 of the cards we track already hold ruDalle: Kandinsky 3.0 on their own, a second card is rarely the answer here.

Why does the quantisation differ between cards for ruDalle: Kandinsky 3.0?

Each card is shown running the least-compressed copy it can hold, and ruDalle: Kandinsky 3.0 appears at 5 different compression levels across the cards that fit it. Bigger cards get the more accurate version.

How accurate are these ruDalle: Kandinsky 3.0 speed estimates?

They are calculated from specifications rather than measured, and each carries a range — 171–456 tok/s on the B200, for instance. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

What GPU do I need to run ruDalle: Kandinsky 3.0?

The smallest card in our catalogue that holds ruDalle: Kandinsky 3.0 is the Xeon Phi 5110P, with 8 GB of memory. It runs the model at Q3_K_M using about 6.5 GB, and produces roughly 20.0 tokens per second. 509 cards in total can run it.

How fast is ruDalle: Kandinsky 3.0 on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 285 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 455 of the cards that can run ruDalle: Kandinsky 3.0 clear that.

How much VRAM does ruDalle: Kandinsky 3.0 need?

About 6.5 GB at Q3_K_M 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 ruDalle: Kandinsky 3.0 on a 8 GB GPU?

Yes. A CMP 170HX 8 GB with 8 GB runs it at Q3_K_M, using about 6.5 GB and generating roughly 143 tokens per second — a tight fit.

Can I run ruDalle: Kandinsky 3.0 on a 12 GB GPU?

Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q6_K, using about 10.7 GB and generating roughly 47.2 tokens per second — a tight fit.

Can I run ruDalle: Kandinsky 3.0 on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q8_0, using about 13.4 GB and generating roughly 40.2 tokens per second — a tight fit.

Can I run ruDalle: Kandinsky 3.0 on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 13.4 GB and generating roughly 47.7 tokens per second — a comfortable fit.

Is ruDalle: Kandinsky 3.0 open source?

Its weights are published, so ruDalle: Kandinsky 3.0 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 ruDalle: Kandinsky 3.0 have?

ruDalle: Kandinsky 3.0 has 11.9B parameters. 11.9 billion parameters. 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 ruDalle: Kandinsky 3.0?

ruDalle: Kandinsky 3.0 was published by Sber, based in Russia, categorised as industry,Government.

When was ruDalle: Kandinsky 3.0 released?

ruDalle: Kandinsky 3.0 was published in December 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.

What is ruDalle: Kandinsky 3.0 used for?

ruDalle: Kandinsky 3.0 works in Image generation, and is recorded as handling text-to-image, Image generation. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

Where can I download ruDalle: Kandinsky 3.0?

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

How much compute was used to train ruDalle: Kandinsky 3.0?

Around 2 × 10²⁰ FLOP, on NVIDIA A100. 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 ruDalle: Kandinsky 3.0 if it does not fit in my GPU?

It can be split between the card and system memory, but ruDalle: Kandinsky 3.0 generates painfully slowly that way — the nearest miss we calculate is short by 2.5 GB. Nothing on this page assumes offloading.

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