Escape Meridian

· Dec 7, 2023

Joshua Lopez argues in Stability AI’s account for alternatives to GPUs

Markdown · JSON · Original source

Published December 7, 2023, Joshua Lopez describes ’s work with Intel to build an AI supercomputer and argues that alternatives to deserve consideration. He gives three reasons for exploring Intel’s Gaudi2 : availability, price and scalability. Stability AI says its code works on HPUs, and that the Gaudi2 offers more economical performance, 8 bit precision and memory and interconnect advantages. These are the company’s claims; the post provides no methodology or cost figures to assess the comparisons. Lopez also says the system’s faster interconnect enables large model training at almost linear scaling. Stability AI presents the project as a way to meet demand for computing power and offer developers and enterprise clients another choice beyond CUDA and traditional GPU solutions. The post’s argument is that access to alternatives matters when compute is essential to running its models, but it does not quantify the system’s effect on training, operating costs or service for those users.

Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.

Stability AIArticle · stability.aiOpen original source

Original source

· Article

Source details

Original source date: Dec 7, 2023 (2023-12-07T20:45:00+00:00).

Date basis: Date stated on the publisher’s page.

Added to Escape Meridian: 2026-11-22T05:48:12.388Z. First observed: 2026-10-28T12:20:13.167Z. Retrieved: 2026-10-28T12:21:53.291Z.

Referenced by

Recorded citations in public readings. A reference alone does not establish agreement or influence.

No public references recorded here yet.

Reference data (JSON)

Limits

Text as retrieved on October 28, 2026; originally published December 7, 2023. The post gives no methodology, cost figures or comparison conditions for its performance and affordability claims. The post does not quantify the system’s effect on training, operating costs or service for developers and enterprise clients.

Open question

What results and cost comparisons support ’s claims about Gaudi2’s performance and affordability?

Source-stated

Joshua Lopez says Stability AI worked with Intel to build an AI supercomputer and explored alternatives to GPUs for its next-generation models.

In our first post, we’ll talk about our work with Intel to build a new state-of-the-art AI supercomputer.

Source-stated

Stability AI says Intel was ready to supply its Gaudi2 AI accelerator while some silicon providers had lead times extending up to a year.

Unlike many silicon providers with lead times extending up to a year, Intel was ready to hit the ground running with their cutting-edge generative AI chip, the Intel(R) Gaudi(R)2 AI accelerator.

Source-stated

Stability AI says its code for popular image and language models works seamlessly on HPUs.

Code for our most popular image and language models works seamlessly on HPUs (acronym for Intel Gaudi processor units) unlike many other accelerators that are not directly compatible.

Source-stated

Stability AI claims Gaudi2 offers top-tier performance at a more economical price point than competing chips and that 8 bit precision further increases performance.

Simply put, the Gaudi2 offers top-tier performance at a more economical price point than its contemporaries. This affordability is crucial in facilitating rapid training and inference of our models, while managing operating costs. 8 bit precision further increases performance allowing Gaudi2 to

Source-stated

Stability AI says Gaudi2’s memory and interconnect support scalability and large model training at almost linear scaling.

A major advantage is the 96 GB of HBM2e memory, which is 16 GB more than competing chips, enhancing our system’s scalability to meet enterprise demands, particularly running large language and multi-modal models. Additionally, The 2.4 Gb/s interconnect is six times faster than our previous super

Source-stated

Stability AI says it is one of Intel’s largest AI Supercomputer customers and presents the project as offering an alternative for AI computing.

We are proud to be one of the largest customers for Intel’s AI Supercomputer, which marks a significant step in meeting the relentless demand for AI computing power and offering the world a choice and an alternative.

Source and contributor context

Stability AI · Publisher

The publisher is identified here. No individual author has been inferred.

Source media

Media retrieval status: unchecked.

Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.

Graphics processing unit

A chip that performs many calculations at once; the main hardware used to train and run AI models.

AI accelerator

A chip built to speed up AI calculations, such as a graphics processing unit or a tensor processing unit.

Benchmark

A defined set of tests used to compare systems. Results depend on the tasks and measures selected.

Stability AI

Stability AI develops generative models and publishes company proposals about open-model access and oversight.

Editorial provenance

Version 1. Publication ID: c8f5adcc-c03f-21b3-a8a1-1151a98ef374.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-22T05:43:45.103Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.

Observation: a2bc02038f91e1b45954f80711e873beee435d43291506cc25f717ecbaebdf24. Source text hash: da102a03200148acc65acac874a8d6767f3c0b1b985542156c01f734948cfc6a.

Revision and source-check history

Version 1 · 2026-11-22T05:48:12.388Z

Joshua Lopez argues in Stability AI’s account for alternatives to GPUs

Published December 7, 2023, Joshua Lopez describes ’s work with Intel to build an AI supercomputer and argues that alternatives to deserve consideration. He gives three reasons for exploring Intel’s Gaudi2 : availability, price and scalability. Stability AI says its code works on HPUs, and that the Gaudi2 offers more economical performance, 8 bit precision and memory and interconnect advantages. These are the company’s claims; the post provides no methodology or cost figures to assess the comparisons. Lopez also says the system’s faster interconnect enables large model training at almost linear scaling. Stability AI presents the project as a way to meet demand for computing power and offer developers and enterprise clients another choice beyond CUDA and traditional GPU solutions. The post’s argument is that access to alternatives matters when compute is essential to running its models, but it does not quantify the system’s effect on training, operating costs or service for those users.

Original source

· Article

Source details

Original source date: Dec 7, 2023 (2023-12-07T20:45:00+00:00).

Date basis: Date stated on the publisher’s page.

Added to Escape Meridian: 2026-11-22T05:48:12.388Z. First observed: 2026-10-28T12:20:13.167Z. Retrieved: 2026-10-28T12:21:53.291Z.

Limits

Text as retrieved on October 28, 2026; originally published December 7, 2023. The post gives no methodology, cost figures or comparison conditions for its performance and affordability claims. The post does not quantify the system’s effect on training, operating costs or service for developers and enterprise clients.

Open question

What results and cost comparisons support ’s claims about Gaudi2’s performance and affordability?

Source-stated

Joshua Lopez says Stability AI worked with Intel to build an AI supercomputer and explored alternatives to GPUs for its next-generation models.

In our first post, we’ll talk about our work with Intel to build a new state-of-the-art AI supercomputer.

Source-stated

Stability AI says Intel was ready to supply its Gaudi2 AI accelerator while some silicon providers had lead times extending up to a year.

Unlike many silicon providers with lead times extending up to a year, Intel was ready to hit the ground running with their cutting-edge generative AI chip, the Intel(R) Gaudi(R)2 AI accelerator.

Source-stated

Stability AI says its code for popular image and language models works seamlessly on HPUs.

Code for our most popular image and language models works seamlessly on HPUs (acronym for Intel Gaudi processor units) unlike many other accelerators that are not directly compatible.

Source-stated

Stability AI claims Gaudi2 offers top-tier performance at a more economical price point than competing chips and that 8 bit precision further increases performance.

Simply put, the Gaudi2 offers top-tier performance at a more economical price point than its contemporaries. This affordability is crucial in facilitating rapid training and inference of our models, while managing operating costs. 8 bit precision further increases performance allowing Gaudi2 to

Source-stated

Stability AI says Gaudi2’s memory and interconnect support scalability and large model training at almost linear scaling.

A major advantage is the 96 GB of HBM2e memory, which is 16 GB more than competing chips, enhancing our system’s scalability to meet enterprise demands, particularly running large language and multi-modal models. Additionally, The 2.4 Gb/s interconnect is six times faster than our previous super

Source-stated

Stability AI says it is one of Intel’s largest AI Supercomputer customers and presents the project as offering an alternative for AI computing.

We are proud to be one of the largest customers for Intel’s AI Supercomputer, which marks a significant step in meeting the relentless demand for AI computing power and offering the world a choice and an alternative.

Source and contributor context

Stability AI · Publisher

The publisher is identified here. No individual author has been inferred.

Source media

Media retrieval status: unchecked.

Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.

Graphics processing unit

A chip that performs many calculations at once; the main hardware used to train and run AI models.

AI accelerator

A chip built to speed up AI calculations, such as a graphics processing unit or a tensor processing unit.

Benchmark

A defined set of tests used to compare systems. Results depend on the tasks and measures selected.

Stability AI

Stability AI develops generative models and publishes company proposals about open-model access and oversight.

Editorial provenance

Version 1. Publication ID: c8f5adcc-c03f-21b3-a8a1-1151a98ef374.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-22T05:43:45.103Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.

Observation: a2bc02038f91e1b45954f80711e873beee435d43291506cc25f717ecbaebdf24. Source text hash: da102a03200148acc65acac874a8d6767f3c0b1b985542156c01f734948cfc6a.

uncertain · 2026-11-22T05:43:45.103Z · anthropic/claude-opus-5.5

Publication: c8f5adcc-c03f-21b3-a8a1-1151a98ef374

attribution — pass: The reading correctly attributes the post to Joshua Lopez at Stability AI. It keeps performance, affordability and scalability statements framed as company claims, and it notes the lack of benchmark methodology and cost figures.

Written by Joshua Lopez

summary — uncertain: Several central terms are left unexplained for a general reader. The reading says code "works on HPUs" but omits the source's own gloss (Intel Gaudi processor units). CUDA, interconnect, 8 bit precision and "almost linear scaling" are not explained, and the definitions cover only GPU, AI accelerator and benchmark. Smallest repair: say HPUs are Intel Gaudi processor units, and add brief plain-language explanations or tooltip terms for CUDA, interconnect, 8 bit precision and linear scaling. The structure, the argument and the stated limits are otherwise faithful.

HPUs (acronym for Intel Gaudi processor units)

date — pass: The reading's December 7, 2023 matches the UTC day of originalPublished and the source's "Dec 7" dateline. The source is a publisher page, so no archive disclosure is needed. The page was modified in October 2024, but the cited body text reads as the original announcement, and the reading uses no later sidebar material and no hindsight framing.

Company Dec 7 Written by Joshua Lopez

support — pass: Each claim matches the source text and is labeled as source-stated. Rendering "Accessibility" as availability reflects the source's lead-time reasoning. The stated limitations about the missing benchmarks, missing cost figures and unquantified effect are accurate gaps. The question follows from those gaps and is not leading. The October 2024 edit does not appear to affect the cited body passages.

enabling large model training at almost linear scaling

media — pass: No media was selected, and no verified media candidates were supplied, so null is appropriate.

Challenge or correct this reading

Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.