-
Notifications
You must be signed in to change notification settings - Fork 104
Qtc
adanalis edited this page Mar 6, 2012
·
7 revisions
Description: Measures performance of the Quality Threshold Clustering algorithm.
Problem Sizes (number of points, threshold)
- s=1 ( 8*1024, 1)
- s=2 (16*1024, 1)
- s=3 (16*1024, 4)
- s=4 (26*1024, 1)
With these sizes, a run on a Tesla C2070, will take approximately: 3.5, 16, 90 and 550 seconds respectively
Precision: float
Includes PCIe Transfer Time: Yes, in [testName]_PCIe measurements
Parameters
- --s, preselected problem sizes (overrides all other parameters)
- --PointCount, point count
- --DataFile, BLAST data input file name
- --Threshold, cluster diameter threshold
- --SaveOutput, save the resulting clusters in different files
- --Verbose, print cluster cardinalities
- --TextureMem, use Texture memory for distance matrix
- --Dense, use the compact storage distance matrix regardless of problem size
Specific Tests
- TBD - achieved GFLOPS?
- TBD - achieved GFLOPS