MegaScale (530B)
No estimate
No hardware requirements for this model
The weights for this model have not been published, so it cannot be downloaded or run on your own hardware at any size. It is reachable only through its provider, and no graphics card changes that.
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
- ByteDance,Peking University
- Organisation type
- Industry,Academia
- Country
- China
- Published
- 23 February 2024
- Authors
- Ziheng Jiang, Haibin Lin, Yinmin Zhong, Qi Huang, Yangrui Chen, Zhi Zhang, Yanghua Peng, Xiang Li, Cong Xie, Shibiao Nong, Yulu Jia, Sun He, Hongmin Chen, Zhihao Bai, Qi Hou, Shipeng Yan, Ding Zhou, Yiyao Sheng, Zhuo Jiang, Haohan Xu, Haoran Wei, Zhang Zhang, Pengfei Nie, Leqi Zou, Sida Zhao, Liang Xiang, Zherui Liu, Zhe Li, Xiaoying Jia, Jianxi Ye, Xin Jin, Xin Liu
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Language
- Task
- Language modeling/generation
- Approach
- Self-supervised learning
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
- 530B
- Training data
- 300,000,000,000 tokens
- Epochs
- 1
- Batch size
- 12,582,912
Two models are trained for epoch to evaluate the MegaScale training system; one model with 175B and another with 530B parameters. This entry reports the 530B model. There is a third production model mentioned, with fewer details.
300B tokens
Experiments with both 768 and 6144, multiplied by 2048
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
- 9.7 × 10²³ FLOP
- How it was established
- Comparison with other models
175B models uses 3.2e23 FLOPs (Table 2, bottom row) With constant dataset size and utilization, FLOPs should scale linearly in # parameters, so: 3.2e23 * (530/175) = 9.7e23
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
- 11,200
- Wall-clock time
- 118 hours
- Hardware utilisation
- MFU 54.3%
- Power draw
- 8.9 MW
- Data centre
- ByteDance MegaScale training cluster
175B parameter model is stated to have taken 1.75 days. 500B model used more compute, fewer GPUs, and had slightly worse MFU. Accounting for these: 1.75 days * 24 hours/day * (9.7e23/3.2e23) * (11200/12288) * (0.552/0.543) = 117.9 hours
MFU given in fgure 9, when training with the full 11200 GPUs.
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
- Closed — provider access only
- Model access
- Unreleased
- Training code
- Unreleased
they open-sourced their framework but don't see training code for their big model. https://github.com/volcengine/vescale Model weights are unreleased
How it is classified
Labels the source dataset applies when tracking notable models, and how confident it is in the entry.
- Likely above 10²³ FLOP
- Yes
- Why it is tracked
- SOTA improvement
- Record confidence
- Confident
- Citations
- 302
Improves SOTA in FLOP utilization for distributed LLM training by 1.34X
Sources
Where this record came from and when it was last checked.
- Reference
- MegaScale: Scaling Large Language Model Training to More Than 10,000 GPUs
- Last updated
- 25 May 2026
What the numbers mean
Background
MegaScale (530B) was published by ByteDance,Peking University, in China, in February 2024. industry,Academia is the category the publisher falls under.
It works in Language, and is recorded as doing language modeling/generation.
Because the weights are not available, none of the hardware figures elsewhere on this site apply to it.
How it was trained
Producing it required around 9.7 × 10²³ FLOP of arithmetic, on NVIDIA A100, which is a statement about the training budget rather than about inference.
The training set ran to roughly 300,000,000,000 tokens.
The reason it appears in this catalogue at all is sOTA improvement.
Answers
MegaScale (530B) — common questions
How many parameters does MegaScale (530B) have?
MegaScale (530B) has 530B parameters. Two models are trained for epoch to evaluate the MegaScale training system; one model with 175B and another with 530B parameters. This entry reports the 530B model. There is a third production model mentioned, with fewer details. 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 MegaScale (530B)?
MegaScale (530B) was published by ByteDance,Peking University, based in China, categorised as industry,Academia.
When was MegaScale (530B) released?
MegaScale (530B) was published in February 2024. 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 MegaScale (530B) used for?
MegaScale (530B) works in Language, and is recorded as handling language modeling/generation. 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.
How much compute was used to train MegaScale (530B)?
Around 9.7 × 10²³ 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.
What GPU do I need to run MegaScale (530B)?
None. MegaScale (530B) is a closed model — its weights were never published, so it cannot be downloaded or run on your own hardware at any price. It is reachable only through its provider.
Is MegaScale (530B) open source?
No. MegaScale (530B) has not had its weights published, so it exists only as a service controlled by its owner.
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.