PIXART-α TPS calculator

Open weights Huawei Noah's Ark Lab,The University of Hong Kong,Hong Kong University of Science and Technology (HKUST) 600M parameters September 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

818 cards that can run it

818 cards we hold specifications for

Smallest card that fits

Tesla C1080

4 GB · Q8_0 · 61.4 tok/s

Fastest card

B200

5,647 tok/s · 180 GB

Which GPUs can run PIXART-α?

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.

818 cards match

Calculating
Needs Quantisation Fit
5,647 tok/s

3,388–9,035 · low confidence

B200 NVIDIA 180 GB 8,000 GB/s Jan 2024 1.3 GB Q8_0 Comfortable
5,647 tok/s

3,388–9,035 · low confidence

B300 NVIDIA 288 GB 8,000 GB/s Sep 2025 1.3 GB Q8_0 Comfortable
4,509 tok/s

2,706–7,215 · low confidence

Radeon Instinct MI350X AMD 288 GB 8,190 GB/s Jan 2025 1.3 GB Q8_0 Comfortable
4,509 tok/s

2,706–7,215 · low confidence

Radeon Instinct MI355X AMD 288 GB 8,190 GB/s Jan 2025 1.3 GB Q8_0 Comfortable
3,606 tok/s

2,164–5,770 · low confidence

Radeon Instinct MI300 AMD 128 GB 6,550 GB/s Jan 2023 1.3 GB Q8_0 Comfortable
3,452 tok/s

2,071–5,523 · low confidence

H200 NVL NVIDIA 141 GB 4,890 GB/s Nov 2024 1.3 GB Q8_0 Comfortable
3,452 tok/s

2,071–5,523 · low confidence

H200 SXM 141 GB NVIDIA 141 GB 4,890 GB/s Nov 2024 1.3 GB Q8_0 Comfortable
3,304 tok/s

1,982–5,286 · low confidence

Radeon Instinct MI325X AMD 256 GB 6,000 GB/s Oct 2024 1.3 GB Q8_0 Comfortable
2,932 tok/s

1,759–4,691 · low confidence

Radeon Instinct MI300A AMD 128 GB 5,325 GB/s Dec 2023 1.3 GB Q8_0 Comfortable
2,932 tok/s

1,759–4,691 · low confidence

Radeon Instinct MI300X AMD 192 GB 5,325 GB/s Dec 2023 1.3 GB Q8_0 Comfortable
2,932 tok/s

1,759–4,691 · low confidence

Radeon Instinct MI308X AMD 192 GB 5,325 GB/s Dec 2023 1.3 GB Q8_0 Comfortable
2,781 tok/s

1,669–4,450 · low confidence

H100 NVL 94 GB NVIDIA 94 GB 3,940 GB/s Mar 2023 1.3 GB Q8_0 Comfortable
2,372 tok/s

1,423–3,795 · low confidence

H100 PCIe 96 GB NVIDIA 96 GB 3,360 GB/s Mar 2023 1.3 GB Q8_0 Comfortable
2,372 tok/s

1,423–3,795 · low confidence

H100 SXM5 80 GB NVIDIA 80 GB 3,360 GB/s Oct 2022 1.3 GB Q8_0 Comfortable
2,372 tok/s

1,423–3,795 · low confidence

H100 SXM5 94 GB NVIDIA 94 GB 3,360 GB/s Mar 2023 1.3 GB Q8_0 Comfortable
2,372 tok/s

1,423–3,795 · low confidence

H100 SXM5 96 GB NVIDIA 96 GB 3,360 GB/s Mar 2023 1.3 GB Q8_0 Comfortable
2,372 tok/s

1,423–3,795 · low confidence

H800 SXM5 NVIDIA 80 GB 3,360 GB/s Mar 2023 1.3 GB Q8_0 Comfortable
1,806 tok/s

1,084–2,889 · low confidence

Radeon Instinct MI250 AMD 128 GB 3,280 GB/s Nov 2021 1.3 GB Q8_0 Comfortable
1,806 tok/s

1,084–2,889 · low confidence

Radeon Instinct MI250X AMD 128 GB 3,280 GB/s Nov 2021 1.3 GB Q8_0 Comfortable
1,505 tok/s

903–2,408 · low confidence

Data Center GPU Max 1550 Intel 128 GB 3,280 GB/s Jan 2023 1.3 GB Q8_0 Comfortable
1,473 tok/s

884–2,357 · low confidence

Data Center GPU Max Subsystem Intel 128 GB 3,210 GB/s Jan 2023 1.3 GB Q8_0 Comfortable
1,440 tok/s

864–2,304 · low confidence

A100 SXM4 80 GB NVIDIA 80 GB 2,040 GB/s Nov 2020 1.3 GB Q8_0 Comfortable
1,440 tok/s

864–2,304 · low confidence

A100X NVIDIA 80 GB 2,040 GB/s Jun 2021 1.3 GB Q8_0 Comfortable
1,440 tok/s

864–2,304 · low confidence

A800 SXM4 80 GB NVIDIA 80 GB 2,040 GB/s Aug 2022 1.3 GB Q8_0 Comfortable
1,440 tok/s

864–2,304 · low confidence

H100 CNX NVIDIA 80 GB 2,040 GB/s Mar 2023 1.3 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
Huawei Noah's Ark Lab,The University of Hong Kong,Hong Kong University of Science and Technology (HKUST)
Organisation type
Industry,Academia,Academia
Country
China, Hong Kong
Published
30 September 2023
Authors
Junsong Chen, Jincheng Yu, Chongjian Ge, Lewei Yao, Enze Xie, Yue Wu, Zhongdao Wang, James Kwok, Ping Luo, Huchuan Lu, Zhenguo Li

What it does

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

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

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
600M

0.6B

Training data
626,000,000,000 tokens

Table 2

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
1.5 × 10²¹ FLOP

PixArt-α only takes 12% of Stable Diffusion v1.5's training time (753 vs. 6,250 A100 GPU days), saving nearly $300,000 ($28,000 vs. $320,000) and reducing 90% CO2 emissions. Moreover, compared with a larger SOTA model, RAPHAEL, our training cost is merely 1%. To ensure fairness, we convert the V100 GPU days (1656) of our training to A100 GPU days (753) they compare their compute with Imagen as 7132:753 (see Table 2) Imagen compute was 1.4600000000000002e+22 FLOPS (from Epoch table) then PIXART-…

How it was established
Hardware,Comparison with other models

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 V100

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)

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
PIXART-α: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis
Last updated
28 November 2025

The extremes

What the numbers mean

What you need to run it

Minimum card

Tesla C1080

Memory needed

1.3 GB

Fastest

5,647 tok/s

PIXART-α is small enough at 600M parameters that hardware is rarely the obstacle — 818 of the cards we track can run it, including cards several years old.

The least hardware that works is a Tesla C1080. Its 4 GB is enough at Q8_0 compression, giving roughly 61.4 tokens per second.

At the other end, a B200 generates roughly 5,647 tokens per second on it, on the strength of 8,000 GB/s of memory bandwidth.

About this model

PIXART-α was published by Huawei Noah's Ark Lab,The University of Hong Kong,Hong Kong University of Science and Technology (HKUST), in China, in September 2023. It comes out of industry,Academia,Academia.

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

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

How fast it runs, and why

The median result is around 158.6 tokens per second; 809 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.

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.

What went into building it

Producing it required around 1.5 × 10²¹ FLOP of arithmetic, on NVIDIA V100, which is a statement about the training budget rather than about inference.

The training set ran to roughly 626,000,000,000 tokens.

Step by step

How to choose a GPU for PIXART-α

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

    Look at what PIXART-α actually needs — around 1.3 GB at Q8_0. No amount of processing power compensates for a card that cannot hold it.

  2. 02

    Match the context to your actual use

    Set the context to what you will actually use. The cache grows with the conversation, and it is the usual reason PIXART-α stops fitting a card that seemed fine.

  3. 03

    Set a quality floor

    Compression is what makes PIXART-α fit smaller cards, at some cost in accuracy — Q8_0 on the smallest card that fits. A minimum quality removes the ones that go too far.

  4. 04

    Compare tokens per second, not specifications

    The speed ordering for PIXART-α is effectively an ordering by memory bandwidth, which is why the B200 tops it at 5,647 tok/s.

  5. 05

    Check the fit verdict before buying

    A tight fit runs PIXART-α but leaves nothing spare for a longer conversation; comfortable has headroom. If you expect to grow the context, buy for comfortable.

  6. 06

    Check the card from the other side

    Every card name links to its own page, which runs the same calculation across the whole model catalogue. Worth a look before buying for PIXART-α alone — a card is usually bought for more than one model.

Answers

PIXART-α — common questions

01

Can I run PIXART-α on a 16 GB GPU?

Yes. A Tesla V100 SXM2 16 GB with 16 GB runs it at Q8_0, using about 1.3 GB and generating roughly 798 tokens per second — a comfortable fit.

02

Can I run PIXART-α on a 24 GB GPU?

Yes. A GeForce RTX 5090 D V2 with 24 GB runs it at Q8_0, using about 1.3 GB and generating roughly 946 tokens per second — a comfortable fit.

03

Is PIXART-α open source?

Its weights are published, so PIXART-α 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.

04

How many parameters does PIXART-α have?

PIXART-α has 600M parameters. 0.6B. 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.

05

Who created PIXART-α?

PIXART-α was published by Huawei Noah's Ark Lab,The University of Hong Kong,Hong Kong University of Science and Technology (HKUST), based in China, categorised as industry,Academia,Academia.

06

When was PIXART-α released?

PIXART-α was published in September 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.

07

What is PIXART-α used for?

PIXART-α works in Image generation, and is recorded as handling image generation, Text-to-image. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

08

Where can I download PIXART-α?

The weights for PIXART-α are published, though we do not hold a repository link for it. This site calculates hardware requirements rather than hosting model files.

09

How much compute was used to train PIXART-α?

Around 1.5 × 10²¹ FLOP, on NVIDIA V100. 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.

10

Can I run PIXART-α 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 PIXART-α is rarely worth using. Every figure here assumes the whole model is on the card.

11

Would two GPUs run PIXART-α faster?

Two cards buy memory rather than speed. That matters for PIXART-α only if one card cannot hold it — 818 can, so a second adds little.

12

Why does the quantisation differ between cards for PIXART-α?

A larger card holds a more accurate copy. Across the cards that run PIXART-α, 1 compression levels are used; the floor control above pins it to one.

13

How accurate are these PIXART-α speed estimates?

Every figure is derived from memory bandwidth and model size, not benchmarked. That is why each is published as a range such as 3,388–9,035 tok/s on the B200 rather than a single number.

14

What GPU do I need to run PIXART-α?

The smallest card in our catalogue that holds PIXART-α is the Tesla C1080, with 4 GB of memory. It runs the model at Q8_0 using about 1.3 GB, and produces roughly 61.4 tokens per second. 818 cards in total can run it.

15

How fast is PIXART-α on a GPU?

It depends on the card. The quickest we calculate is a B200 at about 5,647 tokens per second; the slowest that still runs it manages considerably less. Reading speed is around ten tokens per second, and 809 of the cards that can run PIXART-α clear that.

16

How much VRAM does PIXART-α need?

About 1.3 GB at Q8_0 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.

17

Can I run PIXART-α on a 8 GB GPU?

Yes. A CMP 170HX 8 GB with 8 GB runs it at Q8_0, using about 1.3 GB and generating roughly 1,052 tokens per second — a comfortable fit.

18

Can I run PIXART-α on a 12 GB GPU?

Yes. A GeForce RTX 3080 Ti with 12 GB runs it at Q8_0, using about 1.3 GB and generating roughly 644 tokens per second — a comfortable fit.

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.