[MAHOUT-2079] investigat Cuda as a backend for in core matricies.

• Type: Wish
• Status: Resolved
• Priority: Major
• Resolution: Won't Fix
• Affects Version/s: 0.14.1
• Fix Version/s: classic-15.0
• Assignee: Andrew Palumbo
• Reporter: Andrew Palumbo
• Created: 05/Dec/19 11:17
• Updated: 31/Jan/24 21:42
• Resolved: 31/Jan/24 21:42

Description

with nvidia card  memories now exceeding what used to be max for some machines.. 16 G easily it would be good to be able to dumb into Cuda memory directly from a statment or assignment.  look into backing in-core matrices with CIDA memory or CUDA Shared memory.