Amazon's PyTorch-on-Trainium tool needs one line changed; 22,000 PyTorch tests are skipped
What happened
Amazon Web Services released TorchNeuron on Monday, an open-source backend that lets code written in PyTorch run on its Trainium2 and Trainium3 AI chips after a one-line device change. It is under the Apache 2.0 licence and enters public beta on 20 October at the PyTorch Conference in San Jose. The first release is incomplete: it does not yet support features such as dynamic input shapes, and about 22,000 PyTorch tests are skipped, according to Amazon's PyTorch blog post as cited by The Next Web. Amazon pitches it as a challenge to the CUDA-based software lock-in around Nvidia, but PyTorch code can already be ported to AMD GPUs and Apple chips with a configuration change.
What we know
- TorchNeuron is open source under Apache 2.0, supports Trainium2 and Trainium3, and enters public beta on 20 October, per The Next Web.
- The beta lacks some PyTorch features, including dynamic input shapes, and about 22,000 PyTorch tests are skipped, per The Next Web citing Amazon's PyTorch blog post.
- Amazon says it is contributing changes upstream to PyTorch to remove code that ties the framework to specific chips, per TechCrunch quoting the company's blog post.
- Amazon says about 85% of AI apps are written with PyTorch, and AWS cites Deepgram as processing 70% more audio per dollar after moving to Trainium, per TechCrunch. Both figures are company claims.
- Early users named by The Next Web include Cartesia, Deepgram and the Lieber Institute for Brain Development, and Hugging Face has added Trainium support to its tools.
What we don't know yet
- Independent benchmarks of Trainium against Nvidia GPUs under TorchNeuron have not been published. The only performance evidence so far is Amazon's demo, in which a model reportedly learned at the same rate on both chips, and Amazon's Deepgram figure.
- It is unclear how much of the unsupported feature set and the roughly 22,000 skipped tests will matter for production workloads, or when they will be closed.
- Whether Amazon's upstream PyTorch changes will be accepted, and whether they make PyTorch more hardware-agnostic or add configuration complexity, remains open. TechCrunch flagged both possibilities.
- The Next Web says Amazon now has a $25 billion annual revenue run rate in custom chips. That is Amazon's own figure and its basis was not given in the sources.
Why it matters
Nvidia's lead rests partly on years of CUDA support inside PyTorch. If TorchNeuron's one-line switch holds up beyond the beta's gaps, AI startups choosing between Nvidia-based neoclouds such as CoreWeave and AWS will face lower porting costs.
Claims
- ConfirmedTorchNeuron lets PyTorch code run on Trainium2 and Trainium3 chips by changing a single device setting.
- ConfirmedTorchNeuron is released as open source under the Apache 2.0 licence.
- ConfirmedTorchNeuron enters public beta on 20 October at the PyTorch Conference in San Jose.
- ConfirmedThe first release does not support some PyTorch features, such as dynamic input shapes.
- ConfirmedAbout 22,000 PyTorch tests are skipped in the first release.
- ConfirmedAmazon says it is contributing changes upstream to PyTorch to remove code that ties the framework to specific chips.
- ConfirmedPyTorch code can already be ported to AMD GPUs and Apple chips with a simple configuration change.
- UnconfirmedAmazon says about 85% of AI apps are written with PyTorch.
- UnconfirmedAWS says Deepgram processes 70% more audio per dollar on Trainium at production-grade accuracy.
- UnconfirmedAmazon says its custom chip business has passed a $25 billion annual revenue run rate.
Related coverage
Why you can trust this story
76%Source map · 3 outlets / 3 articles
- The Next Web
- Article 1thenextweb.comreport
- TechCrunch
- Article 1techcrunch.comreport
- GeekWire
- Article 1geekwire.comreport
How this credibility score is calculated
- Source reliability
- Corroboration
- Primary source
- Atom-verified claims
- No contradiction
- Settled
- Claim attribution
- AI disclosure
Atom-verified claims matched cited source text. Weights are fixed and explainable.
Human accountability
- 3 independent origins / 3 sources
- Drafted by the Newsmesis judgment agent (agent-cli); human approval required before publishing.
Verification ledger · 10
- Claim: TorchNeuron lets PyTorch code run on Trainium2 and Trainium3 chips by changing a single device setting.Status: ConfirmedVerification trailVerifiedChecked by verification engine
At least one extracted atom matched cited source full text.
- Claim: TorchNeuron is released as open source under the Apache 2.0 licence.Status: ConfirmedVerification trailVerifiedChecked by verification engine
At least one extracted atom matched cited source full text.
- Claim: TorchNeuron enters public beta on 20 October at the PyTorch Conference in San Jose.Status: ConfirmedVerification trailVerifiedChecked by verification engine
At least one extracted atom matched cited source full text.
- Claim: The first release does not support some PyTorch features, such as dynamic input shapes.Status: ConfirmedVerification trailVerifiedChecked by verification engine
At least one extracted atom matched cited source full text.
- Claim: About 22,000 PyTorch tests are skipped in the first release.Status: ConfirmedVerification trailVerifiedChecked by verification engine
At least one extracted atom matched cited source full text.
- Claim: Amazon says it is contributing changes upstream to PyTorch to remove code that ties the framework to specific chips.Status: ConfirmedVerification trailVerifiedChecked by verification engine
At least one extracted atom matched cited source full text.
- Claim: PyTorch code can already be ported to AMD GPUs and Apple chips with a simple configuration change.Status: ConfirmedVerification trailVerifiedChecked by verification engine
At least one extracted atom matched cited source full text.
- Claim: Amazon says about 85% of AI apps are written with PyTorch.Status: UnconfirmedVerification trailUnverifiedChecked by verification engine
No atom-level source-text verification is available.
- Claim: AWS says Deepgram processes 70% more audio per dollar on Trainium at production-grade accuracy.Status: UnconfirmedVerification trailUnverifiedChecked by verification engine
No atom-level source-text verification is available.
- Claim: Amazon says its custom chip business has passed a $25 billion annual revenue run rate.Status: UnconfirmedVerification trailUnverifiedChecked by verification engine
No atom-level source-text verification is available.
Update log · 0
AI accelerates. High-risk topics wait for a human editor.