OREAL 32B TPS calculator

Open weights Shanghai AI Lab,Shanghai Jiao Tong University,Chinese University of Hong Kong (CUHK),InnoHK 32B parameters February 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

132 cards that can run it

818 cards we hold specifications for

Smallest card that fits

RTX A4500

20 GB · Q3_K_M · 22.9 tok/s

Fastest card

B200

106 tok/s · 180 GB

Which GPUs can run OREAL 32B?

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

132 cards match

Calculating

					Needs	Quantisation	Fit
106 tok/s 64–169 · low confidence	B200 NVIDIA	180 GB	8,000 GB/s	Jan 2024	35.0 GB	Q8_0	Comfortable
106 tok/s 64–169 · low confidence	B300 NVIDIA	288 GB	8,000 GB/s	Sep 2025	35.0 GB	Q8_0	Comfortable
84.6 tok/s 51–135 · low confidence	Radeon Instinct MI350X AMD	288 GB	8,190 GB/s	Jan 2025	35.0 GB	Q8_0	Comfortable
84.6 tok/s 51–135 · low confidence	Radeon Instinct MI355X AMD	288 GB	8,190 GB/s	Jan 2025	35.0 GB	Q8_0	Comfortable
67.6 tok/s 41–108 · low confidence	Radeon Instinct MI300 AMD	128 GB	6,550 GB/s	Jan 2023	35.0 GB	Q8_0	Comfortable
64.7 tok/s 39–104 · low confidence	H200 NVL NVIDIA	141 GB	4,890 GB/s	Nov 2024	35.0 GB	Q8_0	Comfortable
64.7 tok/s 39–104 · low confidence	H200 SXM 141 GB NVIDIA	141 GB	4,890 GB/s	Nov 2024	35.0 GB	Q8_0	Comfortable
61.9 tok/s 37–99 · low confidence	Radeon Instinct MI325X AMD	256 GB	6,000 GB/s	Oct 2024	35.0 GB	Q8_0	Comfortable
55.0 tok/s 33–88 · low confidence	Radeon Instinct MI300A AMD	128 GB	5,325 GB/s	Dec 2023	35.0 GB	Q8_0	Comfortable
55.0 tok/s 33–88 · low confidence	Radeon Instinct MI300X AMD	192 GB	5,325 GB/s	Dec 2023	35.0 GB	Q8_0	Comfortable
55.0 tok/s 33–88 · low confidence	Radeon Instinct MI308X AMD	192 GB	5,325 GB/s	Dec 2023	35.0 GB	Q8_0	Comfortable
52.2 tok/s 31–83 · low confidence	H100 NVL 94 GB NVIDIA	94 GB	3,940 GB/s	Mar 2023	35.0 GB	Q8_0	Comfortable
44.5 tok/s 27–71 · low confidence	H100 PCIe 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	35.0 GB	Q8_0	Comfortable
44.5 tok/s 27–71 · low confidence	H100 SXM5 80 GB NVIDIA	80 GB	3,360 GB/s	Oct 2022	35.0 GB	Q8_0	Comfortable
44.5 tok/s 27–71 · low confidence	H100 SXM5 94 GB NVIDIA	94 GB	3,360 GB/s	Mar 2023	35.0 GB	Q8_0	Comfortable
44.5 tok/s 27–71 · low confidence	H100 SXM5 96 GB NVIDIA	96 GB	3,360 GB/s	Mar 2023	35.0 GB	Q8_0	Comfortable
44.5 tok/s 27–71 · low confidence	H800 SXM5 NVIDIA	80 GB	3,360 GB/s	Mar 2023	35.0 GB	Q8_0	Comfortable
40.9 tok/s 25–66 · low confidence	GeForce RTX 5090 D V2 NVIDIA	24 GB	1,340 GB/s	Aug 2025	20.1 GB	Q4_K_M	Tight
37.3 tok/s 22–60 · low confidence	A30X NVIDIA	24 GB	1,220 GB/s	Apr 2021	20.1 GB	Q4_K_M	Tight
36.0 tok/s 22–58 · low confidence	DRIVE A100 PROD NVIDIA	32 GB	1,870 GB/s	May 2020	27.5 GB	Q6_K	Tight
36.0 tok/s 22–58 · low confidence	GRID A100A NVIDIA	32 GB	1,870 GB/s	May 2020	27.5 GB	Q6_K	Tight
34.4 tok/s 21–55 · low confidence	GeForce RTX 5090 NVIDIA	32 GB	1,790 GB/s	Jan 2025	27.5 GB	Q6_K	Tight
34.4 tok/s 21–55 · low confidence	GeForce RTX 5090 D NVIDIA	32 GB	1,790 GB/s	Jan 2025	27.5 GB	Q6_K	Tight
33.9 tok/s 20–54 · low confidence	Radeon Instinct MI250 AMD	128 GB	3,280 GB/s	Nov 2021	35.0 GB	Q8_0	Comfortable
33.9 tok/s 20–54 · low confidence	Radeon Instinct MI250X AMD	128 GB	3,280 GB/s	Nov 2021	35.0 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: Shanghai AI Lab,Shanghai Jiao Tong University,Chinese University of Hong Kong (CUHK),InnoHK
Organisation type: Academia,Academia,Academia,Government
Country: China, Hong Kong
Published: 10 February 2025
Authors: Chengqi Lyu, Songyang Gao, Yuzhe Gu, Wenwei Zhang, Jianfei Gao, Kuikun Liu, Ziyi Wang, Shuaibin Li, Qian Zhao, Haian Huang, Weihan Cao, Jiangning Liu, Hongwei Liu, Junnan Liu, Songyang Zhang, Dahua Lin, Kai Chen

What it does

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

Domain: Mathematics, Language
Task: Mathematical reasoning, Quantitative reasoning, Language modeling/generation, Question answering
Base model: Qwen2.5-32B

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: 32B
Training data: tokens
Epochs: 1

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: 2.9 × 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: NVIDIA A100
Chips used: 128
Power draw: 100.5 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: Open source
Hugging Face: internlm

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: Exploring the Limit of Outcome Reward for Learning Mathematical Reasoning
Last updated: 28 November 2025

The extremes

The ten fastest GPUs that run OREAL 32B

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 OREAL 32B

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 it takes to run this model

Minimum card

RTX A4500

Memory needed

16.3 GB

Fastest

106 tok/s

OREAL 32B reaches a parameter count of 32B. 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: 132.

The smallest card that holds it is RTX A4500, with a memory capacity of 20 GB, running it at a compression of Q3_K_M and producing around 22.9 tokens per second.

At the other end sits B200, generating roughly 106 tokens per second on the strength of a memory bandwidth of 8,000 GB/s.

Background

OREAL 32B was published by Shanghai AI Lab,Shanghai Jiao Tong University,Chinese University of Hong Kong (CUHK),InnoHK, in the country recorded as China, during February 2025. The category the publisher falls under is academia,Academia,Academia,Government.

It works in the domain of Mathematics, Language, and is recorded as performing the task of mathematical reasoning, Quantitative reasoning, Language modeling/generation, Question answering.

Rather than being trained from scratch, it is derived from Qwen2.5-32B. Most models at this scale are adapted from an existing base rather than built from nothing.

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

Reading the throughput figures

Half the cards that hold it manage more than 20.7 tokens per second. Producing text faster than most people read it: 103 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.

Step by step

How to choose a GPU for OREAL 32B

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

01

Start from the memory column

The table lists every card able to hold OREAL 32B, needing around 16.3 GB at a compression of Q3_K_M. No amount of processing power compensates for a card that cannot hold it.
02

Match the context to your actual use

Longer conversations cost memory on top of what the weights need. Move the slider to your real working length before trusting any row for OREAL 32B.
03

Choose how far you will compress it

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

Rank by throughput rather than spec sheet

Sort by speed to see how cards rank for OREAL 32B. It will not match a gaming ordering, because generation is bound by memory bandwidth. The card topping the list is B200, at 106 tok/s.
05

Look at the headroom, not just the fit

The fit column separates cards that just manage it from those with room to spare, in the case of OREAL 32B. 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

Following a card through to its own page shows every other model it can hold, which is the question that follows once you have settled on OREAL 32B.

Answers

OREAL 32B — common questions

OREAL 32B— how much VRAM does it need?

It needs about 16.3 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.

OREAL 32B— 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 Q4_K_M, using about 20.1 GB and generating roughly 40.9 tokens per second. The fit is tight.

OREAL 32B— 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.

OREAL 32B— how many parameters does it have?

It has a parameter count of 32B. 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.

OREAL 32B— who created it?

It was published by Shanghai AI Lab,Shanghai Jiao Tong University,Chinese University of Hong Kong (CUHK),InnoHK, based in China, an organisation categorised as academia,Academia,Academia,Government.

OREAL 32B— when was it released?

It was published in February 2025.

OREAL 32B— what is it used for?

It works in the domain of Mathematics, Language, and is recorded as handling the task of mathematical reasoning, Quantitative reasoning, Language modeling/generation, Question answering. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

OREAL 32B— where can I download it?

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

OREAL 32B— can I run it if it does not fit in my GPU?

It can be split between the card and system memory, but it generates painfully slowly that way. The nearest miss we calculate falls short by 5.7 GB. Every figure here assumes the whole model is resident on the card.

OREAL 32B— 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: 132. So a second card is rarely the answer here.

OREAL 32B— why does the quantisation differ between cards?

Each card is shown running the least-compressed copy it can hold. The number of distinct compression levels across the cards that fit it: 5. Bigger cards get the more accurate version, and the quality floor above pins the comparison to one level.

OREAL 32B— how accurate are these speed estimates?

They are calculated from specifications rather than measured, and each carries a range. One example: 64–169 tok/s on B200. The same model and card vary by thirty to fifty per cent depending on the inference software and its version.

OREAL 32B— what GPU do I need to run it?

The smallest card in our catalogue that holds it is RTX A4500, with a memory capacity of 20 GB. It runs the model at a compression of Q3_K_M using about 16.3 GB, and produces roughly 22.9 tokens per second. The number of cards able to run it in total: 132.

OREAL 32B— how fast is it on a GPU?

It depends on the card. The quickest we calculate is B200, at about 106 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: 103.

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