Talk:Blackwell (microarchitecture)

	This article is within the scope of WikiProject Computer graphics, a collaborative effort to improve the coverage of computer graphics on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks.Computer graphicsWikipedia:WikiProject Computer graphicsTemplate:WikiProject Computer graphicscomputer graphics articles
Low	This article has been rated as Low-importance on the project's importance scale.

Computing Mid‑importance

	This article is within the scope of WikiProject Computing, a collaborative effort to improve the coverage of computers, computing, and information technology on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks.ComputingWikipedia:WikiProject ComputingTemplate:WikiProject ComputingComputing articles
Mid	This article has been rated as Mid-importance on the project's importance scale.
	This article is supported by Computer hardware task force (assessed as Mid-importance).

Merge with Draft:Blackwell_(microarchitecture) edit

Latest comment: 1 month ago1 comment1 person in discussion

This article is a duplicate with relatively little content (some of which inaccurate) and it may be easier to merge with Draft:Blackwell (microarchitecture) before publishing to mainspace, thoughts? ObjectOrientedCat (talk) 18:12, 19 March 2024 (UTC)Reply

Compute capability 10.X edit

Latest comment: 1 month ago1 comment1 person in discussion

There are no references that indicate NVIDIA announced compute capability 10.X with Blackwell. Compute capability merely indicates the architecture generation of an NVIDIA GPU, and nothing more. A GeForce 2080 Ti and Titan T4 are very different cards, yet they're both compute capability 7.5 (or SM75) to indicate they are both the Turing architecture.

One architecture can also have multiple compute capability tags associated with it. The most recent example of that is the Ampere architecture: 8.0 and 8.6.

Ada Lovelace and Hopper are two different architectures introduced at the same time, and yet the former was 8.9 (SM89), and the latter was 9.0 (SM90).

Another counterexample of two different architectures having the same major version is Volta (7.0) vs Turing (7.5).

While it is likely that Blackwell will be tagged with 10.0 (SM100), there is still no evidence to suggest that definitely will be tagged 10.0 until either NVIDIA releases the tech specs in detail, or the next version of CUDA supporting Blackwell (likely CUDA 12.5) comes out.

NVIDIA has already indicated next generation Tensor Cores (because of FP4 math), but not much more about the underlying GPU itself, unlike previous years.

Reference on CUDA architectures and compute capability: https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/

References already in the page (none of which make any mention of the Blackwell arch tag / compute capability):

https://www.nvidia.com/en-us/data-center/technologies/blackwell-architecture/

https://www.nvidia.com/en-us/data-center/tensor-cores/#blackwell

https://nvidianews.nvidia.com/news/nvidia-blackwell-platform-arrives-to-power-a-new-era-of-computing

https://investor.nvidia.com/events-and-presentations/presentations/presentation-details/2023/NVIDIA-Investor-Presentation-October-2023/default.aspx

128.61.160.106 (talk) 17:05, 25 March 2024 (UTC)Reply

Add topic