DIAdem
DIAdem on Ryzen Platform
Since WFH had an opportunity to run a vbs script at home on a Ryzen 9 3950x @ 4GHz versus my work Intel i7-6600u @ 2.6GHz. Work PC is 32GB and home is 64GB. Both are SSD. Home PC in RAID stripped on PCIe 4.0 versus 3 on work don't think it is the drive.
Same script same dataset.
Home version is on evaluation, work is not. Maybe evaluation versions are de-tuned?
The Intel is running 2x faster and is 4 iterations behind. The scripts are single threaded. Seems like activity is using the same number of threads between the two.
The script is mostly executing local data finder searches.
Has anyone seen a performance difference this large between Ryzen and Intel on VBS? Is there a way to optimize for AMD?
Will update if I find something.
Noticed that the 4 core intel is sitting fairly loaded at all cores 75% to 100%. Ryzen core is more bursts and averaging lower.
Check timing on the runs:
Ryzen 3950x so the heavy hitter is confirmed to be mostly search. 730 seconds of search where as the 448 seconds...
Using the same files and index files almost like it is not using indexed parameters. Will check the indexed parameters and re-index and see what happens.
Custom properties were disabled on both platforms so it was a good apple / apple comparison.
Re-indexed the files on both platforms with same optimized properties:
AMD 371 seconds for search, Intel 205 seconds 1.8x slower for AMD with optimized fields. And non-optimized data 760 versus 448 or 1.7x slower on AMD.
So about the same... Checking small file drive access.
1MB file access sequential 128MB in 1MB/blocks.
AMD System: 1Queue and 1T Read 3300 MB/sec Write 6600 MB/sec
Intel System: 1Queue and 1T Read 1100 MB/sec Write 850 MB/sec
4k random data 128MB in 4K blocks
AMD System: 1Queue and 1T Read 47.3MB/sec Write 133MB/sec
Intel System: 1Queue and 1T Read 29.3 MB/sec Write 58.27 MB/sec
So not the drives being slower on the AMD System
So not sure why this would be as the datafinder search seems to be just slower on AMD System even if it is better spec'd.
In the examples provided is there a standard script that is used to compare performance in application areas?
I don't think the cores are throttling... Cores are not throttling. Cinebench runs in the 7000's. Some report 9000s in this processor but I am running 64GB 2666MHz memory so I think that impacts me some and I didn't close down anything.
Since the eval is the same, are there any configurations with the DF, DIAdem, Windows settings that might impact by platform?
Can you recommend a test case / stress case for a relative benchmarks? From the example perhaps that perform repeat searches to allow a standard comparison?
https://github.com/ltr-az/ni-diadem-benchmarks/tree/master
Started a basic benchmarking tool and put it on github.
Basic Search.vbs:
I am running this script on 3950x 31 times in 120 seconds.
Benchmark Changes:
Added "main" script:
Added Re-index test that finds all of the files in the example and re-indexes them. 2.3 files / sec on the re-index.
Fixes:
Fixed a oTimer issue with Halt double counting.
ToDo:
- Add Sort test for VB and DIAdem channel native
- Add in file save and delete from the internal Data Portal.
My home trial has expired so will have to try and get a home license installed or get the trial extended so until then only the work PCs.
Added re-indexing test
Added file io test
Added PC info information
To Do:
- sorting test
- chnSort
- VBS Sort
- Need to push the tdms/result file somewhere. Probably try to push back to github for starting host.
Added sort for channels via the ChnMultipleSortExt().
Added commands to commit the result TDM file via git back to github. Just doing it manually via command line or git tool of choice.
Probably will leave it like it is for a while. Test takes about 10 minutes to run. Results are in the properties for each test channelGroup.
Thanks!
LR
https://github.com/ltr-az/ni-diadem-benchmarks
