SimTK: MemtestG80 and MemtestCL: Memory Testers for CUDA- and OpenCL-enabled GPUs: [#801] Show available memory for testing

MemtestG80 and MemtestCL: Memory Testers for CUDA- and OpenCL-enabled GPUs

Share
Follow	Project Follow publicly Follow privately Unfollow See followers (2) List Map More info

Trackers

Date:
2009-05-07 14:09

Priority:
3

State:
Open

Submitted by:
Nobody

Assigned to:
Nobody (None)

Summary:
Show available memory for testing

Detailed description
It may be prudent to display how much memory on the gpu is not being used by the OS and thus is available for testing. Obviously testing all the memory that is not currently in use will generate errors. So it may also be wise to recommend how much memory to test based on how much is currently available. I understand an exact value is very difficult to determine, as a "pre-test" will probably have to be run to find out this information. So, an approximation would be adequate. For example, I have a 9800GT with 512 MB of memory. 400MB of this memory is available and not in use by the operating system. MemtestG80 may recommend that 80% of the available memory be tested, and would output: "Available memory: Approximately ~400MB" "Recommend testing 320MB"
It may be prudent to display how much memory on the gpu is not being used by the OS and thus is available for testing. Obviously testing all the memory that is not currently in use will generate errors. So it may also be wise to recommend how much memory to test based on how much is currently available. I understand an exact value is very difficult to determine, as a "pre-test" will probably have to be run to find out this information. So, an approximation would be adequate. For example, I have a 9800GT with 512 MB of memory. 400MB of this memory is available and not in use by the operating system. MemtestG80 may recommend that 80% of the available memory be tested, and would output: "Available memory: Approximately ~400MB" "Recommend testing 320MB"

Comments (1)
Attachments
Changes

Add A Comment:

Message
Date: 2009-05-12 03:42 Sender: Ben Barsdell I recently had a go at writing a CUDA-based GPU memory test, and in it I used the extremely crude brute-force method to determine the allocable memory (i.e., beginning from total mem, keep trying to alloc decreasing amount until success), and it seemed to work ok. Using the allocable memory determined this way as the input to MemtestG80 also seems to work fine. Would implementing this be an option?

No Changes Have Been Made to This Item

SimTK is maintained through Grant R01GM124443 01A1 from the National Institutes of Health (NIH). It was initially developed as part of the Simbios project funded by the NIH as part of the NIH Roadmap for Medical Research, Grant U54 GM072970.

Version 4.2.2. Website design by Viewfarm. Icons created by SimTK team using art by GraphBerry from www.flaticon.com under a CC BY 3.0 license. Forked from FusionForge 6.0.5.

Feedback