RADX Catalyst PXIe-GPU-T1000-8GB & Transform-SA-P PSD-FFT Benchmarks 24MAY2023 V1.1 скачать в хорошем качестве

RADX Catalyst PXIe-GPU-T1000-8GB & Transform-SA-P PSD-FFT Benchmarks 24MAY2023 V1.1

This video, captured at NI Connect 2023 on 24MAY2023, shows actual PSD / FFT performance obtained with various size PSDs (FFTs) when running the RADX Transform-Signal Analyzer-Python (Transform-SA-P) Example on a RADX Catalyst PXIe-GPU-T1000-8GB-1SP in a PXIe-1092 / PXIe-8881 Dev System. With Transform-SA-P, the application reads TDMS files from disk (in this case from a RADX Trifecta PXIe-SSD-8M.2F-16TB PXIe RAID Module) and then processes the sample data on either the GPU and/or the CPU, which enables us understand the gain of the GPU vs. the CPU on compute intensive applications. in this case, we're just running the GPU (for a customer) to show the observed performance of the Catalyst PXIe-GPU-T1000 on a QAM-8 file with PSDs that range from 8k to 4M samples in length. As you can see from the video, the performance on small FFTs is quite constant until you get to 128k or larger FFTs. Below ~128k FFTs the performance is largely throttled by data transfer, since the GPU is largely idle and is simply waiting on data. Once you get to larger FFTs, the GPU performance comes into play. See the table below for a summation. If you have questions on this or on RADX PXIe GPUs in general, please email us at info@radxtech.com. FFT (PSD) OBSERVED GPU OBSERVED TIME WITH CPU-TO-GPU LENGTH PROCESSING TIME DATA TRANSFER TIME AND GPU PROCESSING TIME 8k / 8,192 ~0.50 msec avg. ~0.60 msec avg. 16k / 16,384 ~0.50 msec avg. ~0.60 msec avg. 32k / 32,768 ~0.50 msec avg. ~0.60 msec avg. 64k / 65,536 ~0.50 msec avg. ~0.60 msec avg. 128k / 131,072 ~0.50 msec avg. ~0.70 msec avg. 256k / 262,144 ~0.50 msec avg. ~0.80 msec avg. 512k / 524,288 ~0.60 msec avg. ~1.20 msec avg. 1M / 1,048,576 ~1.00 msec avg. ~2.40 msec avg. 2M / 2,097,152 ~2.00 msec avg. ~5.00 msec avg. 4M / 4,194,304 ~4.00 msec avg. ~10.00 msec avg.