UPGRADE YOUR BROWSER

We have detected your current browser version is not the latest one. Xilinx.com uses the latest web technologies to bring you the best online experience possible. Please upgrade to a Xilinx.com supported browser:Chrome, Firefox, Internet Explorer 11, Safari. Thank you!

cancel
Showing results for 
Search instead for 
Did you mean: 
Highlighted
Explorer
Explorer
967 Views
Registered: ‎10-09-2014

Longer implementation time in 2017.4 and 2018.1

Hi,

 

I have two slightly different designs targeting KU115 device. In Vivado 2017.2, the implementation time is about 4 hours. However, when I used 2017.4 and 2018.1 with same option (PostRoutePhyOpt), it is taking me about 5 hours. I tried with three runs per version, and the one hour difference is quite clear to me. I have excluded the only runs in 2017.4 and 2018.1 in which PostRoutePhyOpt actually does something. All of the runs meet timing eventually. Is this expected?

 

Thanks,

Jimmy

0 Kudos
4 Replies
Scholar jmcclusk
Scholar
950 Views
Registered: ‎02-24-2014

Re: Longer implementation time in 2017.4 and 2018.1

First, thank you for running these benchmarks!

 

Secondly, these results are not surprising, given that new (and more complex) algorithms are put into Vivado to help close timing.  If it's taking longer, then I can only conclude that the Vivado is able to close timing on more difficult designs, given the increased routing time.   I kinda wish that Xilinx would do some research into GPU acceleration for place & route.

 

I just downloaded 2018.1 this morning, and I will be testing it tonight and tomorrow.  Now it's time to read the release notes!

Don't forget to close a thread when possible by accepting a post as a solution.
0 Kudos
Explorer
Explorer
946 Views
Registered: ‎10-09-2014

Re: Longer implementation time in 2017.4 and 2018.1

Hi @jmcclusk,

Thanks for the reply. Unfortunately I cannot speak for the improved ability to close timing as I don't have a design that doesn't meet timing. However, one more hour or 25% longer run time is quite significant to me. The release note is showing an route option of ultrathread that can improve timing, but it doesn't seem available in project mode. GPU may be a big one, but I think even allowing more than 8 threads will be nice if the algorithm is that scalable.

Jimmy
0 Kudos
Moderator
Moderator
879 Views
Registered: ‎02-07-2008

Re: Longer implementation time in 2017.4 and 2018.1

@linzhongduo, agree, 25% is a significant increase. If I were to send you a PM, would you be willing to submit the 2017.2 design so I can run some local tests with it?

-------------------------------------------------------------------------
Don’t forget to reply, kudo, and accept as solution.
-------------------------------------------------------------------------
0 Kudos
Explorer
Explorer
849 Views
Registered: ‎10-09-2014

Re: Longer implementation time in 2017.4 and 2018.1

Hi @peadard,

 

Sure. I can send you a linked design check point. 

 

Thanks,

Jimmy

0 Kudos