NVIDIA / NVIDIA/cuCascade

Should we have multiple converters between the same idata_representations

Open
#17 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

feature request improvement
Dominant language
C++
Stars
59
Forks
35
Avg merge
17h 38m
Merged PRs (30d)
3

Description

Say that you have two paths in order to move data from GPU to CPU in a particular system:

We could have one converter that uses compression to move the bytes to cpu. This would reduce pressure on the bus between the cpu and the gpu but increase memory pressure temporarily on the GPU to perform the compression itself.

We could have another that just copies the bytes out of gpu into CPU without needing to make any temporary allocations.

Having a system that can understand which of these approaches is the most frugal with respect to memory consumption or some other resource constraint could be interesting form a performance and resiliency perspective

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

The issue names no files, tests, or entry points, so begin by locating the idata_representations and converter abstractions in cuCascade. Clarify how multiple GPU-to-CPU paths should be selected, which resource constraints matter, and what measurable behavior would define the completed feature.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp
Domain
hpc, performance
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.