IMPALA
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
- DeepMind
- Organisation type
- Industry
- Country
- United Kingdom of Great Britain and Northern Ireland
- Published
- 5 February 2018
- Authors
- Lasse Espeholt, Hubert Soyer, Remi Munos, Karen Simonyan, Volodymir Mnih, Tom Ward, Yotam Doron, Vlad Firoiu, Tim Harley, Iain Dunning, Shane Legg, Koray Kavukcuoglu
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Games
- Task
- Atari
- 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
- 1.6M
- Training data
- 11,400,000,000 tokens
"Figure 3 in the paper states that the large architecture has 1.6 million parameters. I am using the large model because it was the only one trained on all the Atari games at once, which seems like the most impressive task in the suite." Source: https://docs.google.com/spreadsheets/d/1Kj4Q5WADcDXtUJLIOfGTCE3tGvxNczEMwyy8QtgSkHk/edit#gid=54587040&fvid=1361937389
From fig 6, there were 1e10 environment frames, and 24 agents. Thus we note down 2.4e11 for the "dataset size"
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
- 1.7 × 10²⁰ FLOP
- How it was established
- Third-party estimation
Source: Ajeya Cotra and Tom Davidson, https://docs.google.com/spreadsheets/d/1Kj4Q5WADcDXtUJLIOfGTCE3tGvxNczEMwyy8QtgSkHk/edit#gid=54587040&fvid=1361937389
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 P100
- Chips used
- 1
- Chip-hours
- 100
- Wall-clock time
- 100 hours
- Power draw
- 286 W
- Compute cost
- $53
Maximum training time for IMPALA is 100 hours according to Figure 6. This seems to refer to the 1 GPU model. The 8 GPU model looks to have been trained about 1/8 as long.
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
- Open source
training code, Apache license: https://github.com/google-deepmind/scalable_agent
How it is classified
Labels the source dataset applies when tracking notable models, and how confident it is in the entry.
- Why it is tracked
- Highly cited,SOTA improvement
- Record confidence
- Confident
- Citations
- 1,829
not an absolute SOTA "IMPALA is able to achieve better performance than previous agents with less data" "We demonstrate the effectiveness of IMPALA for multi-task reinforcement learning on DMLab-30 (a set of 30 tasks from the DeepMind Lab environment (Beattie et al., 2016)) and Atari-57 (all available Atari games in Arcade Learning Environment (Bellemare et al., 2013a)). Our results show that IMPALA is able to achieve better performance than previous agents with less data"
Sources
Where this record came from and when it was last checked.
- Reference
- IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
- Last updated
- 25 May 2026
What the numbers mean
About this model
IMPALA was published by DeepMind, in the country recorded as United Kingdom of Great Britain and Northern Ireland, during February 2018. The category the publisher falls under is industry.
It works in the domain of Games, and is recorded as performing the task of atari.
Its weights were never published, so it can only be reached through its provider. No graphics card changes that.
What went into building it
The training run consumed about 1.7 × 10²⁰ FLOP, on hardware recorded as NVIDIA P100. That figure measures what producing the model cost, and has no bearing on how fast it answers.
It was trained on a corpus of about 11,400,000,000 tokens of text.
Its inclusion criterion: highly cited,SOTA improvement.
Answers
IMPALA — common questions
IMPALA— who created it?
It was published by DeepMind, based in United Kingdom of Great Britain and Northern Ireland, an organisation categorised as industry.
IMPALA— when was it released?
It was published in February 2018. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.
IMPALA— what is it used for?
It works in the domain of Games, and is recorded as handling the task of atari. These are the areas it was designed around; they describe intent rather than a hard boundary.
IMPALA— how much compute was used to train it?
Training consumed around 1.7 × 10²⁰ FLOP, on hardware recorded as NVIDIA P100. 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.
IMPALA— what GPU do I need to run it?
None. This 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.
IMPALA— is it open source?
No. Its weights have not been published, so it exists only as a service controlled by its owner.
IMPALA— how many parameters does it have?
It has a parameter count of 1.6M. "Figure 3 in the paper states that the large architecture has 1.6 million parameters. I am using the large model because it was the only one trained on all the Atari games at once, which seems like the most impressive task in the suite." Source: https://docs.google.com/spreadsheets/d/1Kj4Q5WADcDXtUJLIOfGTCE3tGvxNczEMwyy8QtgSkHk/edit#gid=54587040&fvid=1361937389. 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.
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.