09-10-2019 08:59 AM
@markcurry: parallel synthesis does not kick in when the design is too small (no compile time gain) or when complex timing constraints are applied to the design. This last condition should be addressed in the coming Vivado release (2019.2).
09-10-2019 09:02 AM
09-10-2019 09:03 AM
Hi @sturmel2007 : it looks like you have misinterpreted the content of https://www.xilinx.com/products/design-tools/vivado/memory.html. This page shows that system memory guidelines are the same for Windows and Linux. For each device, there is a typical memory utilization (average design with average timing constraints complexity), and a peak memory utilization (maximum observed by Xilinx on a large set of designs, due to high utilization, congestion, complex timing constraints, etc...).
09-10-2019 09:15 AM
@sturmel2007: thanks for your faith :)! Unfortunately, this is not something Xilinx spends time on. I'll restate the reasoning behind it:
1) There are too many different systems to benchmark. Xilinx focuses on release to release compile time comparison for a same system & OS, and whenever needed, competitive analysis as well.
2) Windows vs Linux performance differences are not unique to Vivado or any Xilinx tools. General information on this topic should be available on the web. Linux is generally more efficient as it is a leaner OS with fewer extra services/processes burning compute resources (firewall, disk encryption, web browsers, etc...).
I hope this closes this thread. Thanks for sharing your thoughts and findings!
09-11-2019 05:01 AM
06-21-2020 02:14 PM
Hi, still a beginner, but i have tested recently Vivado2019.2 on a very good machine, a Dell T7600, two Xeon E2687w, 3.1Ghz, 128Gb RAM, so i have 16 cores, 32 threads. I had a project for the Nexys Video that took 8:40 minutes an a Lenovo i5 2520M machine (T420).
5:40 on WIndows 10
5:20 on Windows 7
5:00 on CentOS 7.
The strange thing, i can choose 32 jobs even in Windows, but the hard and longest job is done in only one core, 100%. I never had more than 6% processor usage and more than 16 Gb RAM.
Sorry for my crude benchmark, but i gave up, i think the only improvement is due to the clock difference, 3.1Ghz, vs. 2.5Ghz.
06-23-2020 06:13 AM - edited 06-23-2020 06:15 AM
Just a quick one,
The problem is the same as all software, trying to split an algorithm across multiple cores.
n one has "yet" made an algorithm that can be split into separate individual parts to optimise and fit a chip.
If you crack that you will make a fortune,
Historical , its been faster to run a Virtual Machine on a computer,
and on that run an OS that you run Vivado on.