RFdiffusion
No estimate
No hardware requirements for this model
This model's weights are open, but no parameter count has been published for it. Every memory and speed figure starts from that number, so we would rather show nothing than a fabricated estimate.
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
- University of Washington,Columbia University,Ecole Normale Supèrieure,University of Cambridge,Massachusetts Institute of Technology (MIT),Seoul National University
- Organisation type
- Academia,Academia,Academia,Academia,Academia,Academia
- Country
- United States of America, France, United Kingdom of Great Britain and Northern Ireland, Korea (Republic of)
- Published
- 23 July 2023
- Authors
- Joseph L. Watson, David Juergens, Nathaniel R. Bennett, Brian L. Trippe, Jason Yim, Helen E. Eisenach, Woody Ahern, Andrew J. Borst, Robert J. Ragotte, Lukas F. Milles, Basile I. M. Wicky, Nikita Hanikel, Samuel J. Pellock, Alexis Courbet, William Sheffler, Jue Wang, Preetham Venkatesh, Isaac Sappington, Susana Vázquez Torres, Anna Lauko, Valentin De Bortoli, Emile Mathieu, Sergey Ovchinnikov, Reg…
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Biology
- Task
- Protein generation, Protein folding prediction
- Base model
- RoseTTAFold All-Atom (RFAA)
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.
- Training data
- tokens
Table 3 (Supplementary materials): Initial Training: crop size 256 25600 examples per epoch 200 epochs Fine tuning: crop size 384 25600 examples per epoch 100 epochs 256*25600*200+384*25600*100 = 2293760000 tokens ~ 2.3B 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
- Hardware
- Fine-tuning compute
- 5.8 × 10²¹ FLOP
125000000000000*64*672*3600*0.3=5.80608e+21
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
- Chips used
- 64
- Wall-clock time
- 672 hours (28 days)
- Power draw
- 38.2 kW
"RoseTTAFold was trained for 4 weeks on 64 V100 GPUs on Microsoft Azure." 4*7*24=672 hours
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
Code for running RFdiffusion has been released on GitHub, free for academic, personal and commercial use at https://github.com/Rosetta- Commons/RFdiffusion. It is also available as a Google Colab notebook, accessible through GitHub.
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
- De novo design of protein structure and function with RFdiffusion
- Last updated
- 28 November 2025
What the numbers mean
What this model is
RFdiffusion was published by University of Washington,Columbia University,Ecole Normale Supèrieure,University of Cambridge,Massachusetts Institute of Technology (MIT),Seoul National University, in United States of America, in July 2023. academia,Academia,Academia,Academia,Academia,Academia is the category the publisher falls under.
It works in Biology, and is recorded as doing protein generation, Protein folding prediction.
Its starting point was RoseTTAFold All-Atom (RFAA) — most models at this scale are adapted from an existing base rather than built from nothing.
Because its weights were released, nothing about running it depends on a provider staying available — it is yours once downloaded.
Answers
RFdiffusion — common questions
Is RFdiffusion open source?
Its weights are published, so RFdiffusion 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 RFdiffusion have?
No parameter count has been published for RFdiffusion, which is why no memory or speed figure appears on this page.
Who created RFdiffusion?
RFdiffusion was published by University of Washington,Columbia University,Ecole Normale Supèrieure,University of Cambridge,Massachusetts Institute of Technology (MIT),Seoul National University, based in United States of America, categorised as academia,Academia,Academia,Academia,Academia,Academia.
When was RFdiffusion released?
RFdiffusion was published in July 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 RFdiffusion used for?
RFdiffusion works in Biology, and is recorded as handling protein generation, Protein folding prediction. 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.
Where can I download RFdiffusion?
The weights for RFdiffusion are published, though we do not hold a repository link for it. This site calculates hardware requirements rather than hosting model files.
What GPU do I need to run RFdiffusion?
We cannot say. RFdiffusion has open weights, but no parameter count has been published for it, and every memory and speed calculation starts from that number. We would rather show nothing than a fabricated estimate.
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.