Talk:Spatial architecture

Review for Spatial architecture

This article is well-established and supported by many solid references. It offers a clear, structured overview and effectively enriches readers' knowledge across multiple aspects. Thank you. Garden Yuanlin (talk) 13:20, 11 July 2025 (UTC)[reply]

General Comments

The article is well-structured. Just few comments - not sure the concept of "kernel" is well known even in computer science...I would define what a kernel is. For now, I just added a sentence about it Here and there, some concepts may need some prior knowledge. But overall, there are several strong references. Well done! --P3pp399 (talk) 13:25, 11 July 2025 (UTC).[reply]

Review comments

The structure of the article is good. But the beginning is a little unclear. There are many types of computer architectures. It would be more helpful to understand if the spatial architectures was pointed out which type of computer architecture it belong to. 去冰 (talk) 13:36, 11 July 2025 (UTC)[reply]

Peer review

Very good article, well structured and with many meaningful references. I only added an example in the "Specific implementations" paragraph mentioning another kind of spatial architecture, so that it could give another prospective and use case apart from the already mentioned AI architecture accelerators. FrancioT (talk) 14:36, 11 July 2025 (UTC)[reply]

Article name and links from other articles

A few months ago, I wrote a note Talk:Dataflow architecture#Creating a new article for the dataflow used in NPUs because I had noticed a growing use of the phrase "spatial dataflow". That did not attract any attention, but now User:EMJzero has created this article which I can learn from. Thanks.

I am not sure if "Spatial architecture" is the best title. Google AI Overview says:

"Spatial architecture" refers to the design and organization of physical spaces, encompassing both interior and exterior environments. It involves the arrangement of elements within a space to create a functional, aesthetically pleasing, and meaningful environment for users. Spatial architecture also considers how people experience and interact with the environment, taking into account factors like flow, circulation, lighting, and accessibility.

"Spatial dataflow" is more clearly about computing.

Anyway, the dataflow architecture article probably should have a link to this article. Also, I wrote a section Spatial computing#Other use of the term which could be merged here and replaced there by a "Not to be confused with" hat note. JonH (talk) 13:01, 12 July 2025 (UTC)[reply]

Thank you JonH for the feedback!

I apologize for not noticing your talk, thank you for pointing it out!

About the dataflow architecture, as you correctly observed, that is surely a precursor to modern "spatial architectures" or NPUs. In particular, the concept of dataflow, as introduced in dataflow architectures, carries on to spatial architectures well, and could justify a cross-reference between the two articles. I will work on that. Same goes for the spatial computing disclaimer. Similarly, I think that I should also add a link to this article in the Dataflow page.

About the choice of title. It arose from the term "spatial architecture" being used prominently in Eyeriss’s article and all subsequent works, that became the foundation behind this wikipedia article. I agree that Google often suggests it as a term used for physical spaces, but I think that is the case simply because not much material is available, outside of scientific literature, using this term in computer science. I originally though of using the title "Spatial architecture (computer science)" to disambiguate, but since no other article existed with just the "spatial architecture" name, I opted to use it instead, and add "In computer science, …" in the leading section. That clarification was removed in a previous edit, but I brought it back for the reason I just stated. I am sure that, given enough time, Google and LLMS will digest this article and update their answer accordingly ^-^.

Nonetheless, "spatial dataflow" is an interesting alternative, that I would consider adding as an alias. However, it is not common place in scientific literature, or worse, it has been occasionally used to refer to the “parallelization” part of a mapping (see the "GAMMA" reference). Therefore, I am unsure about using it as the main title. EMJzero (talk) 18:32, 12 July 2025 (UTC)[reply]

Thanks for responding. I am happy, and probably will not comment here again. Perhaps in a couple of years, computer science will evolve new terminology and distinctions. JonH (talk) 20:32, 12 July 2025 (UTC)[reply]