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: 
Visitor sturmel2007
Visitor
1,233 Views
Registered: ‎07-06-2017

Vivado linux vs windows benchmark

Jump to solution

Some one as bechmark that show the difference of performance between Linux and Windows when synthesing with Vivado?

0 Kudos
1 Solution

Accepted Solutions
Scholar drjohnsmith
Scholar
759 Views
Registered: ‎07-09-2009

Re: Vivado linux vs windows benchmark

Jump to solution

Thank you.

 

be careful about drawing a conclusion from a corelation,

Th eexample I use is that "bad breath causes Lung cancer is a clear and defined correlation".

Which is is ,

If you dont relaise that the bad breath is for those that smoke,

    and its the smoking thats causing the bad breath and cancer.

 

But yep, most machines nowdays ( apart from the companies I seem to work at )  have plenty of memory , and linux does run faster than Windows for vivado.

 

 

<== If this was helpful, please feel free to give Kudos, and close if it answers your question ==>

View solution in original post

34 Replies
Scholar drjohnsmith
Scholar
1,221 Views
Registered: ‎07-09-2009

Re: Vivado linux vs windows benchmark

Jump to solution

Lest just say ,

all the places I know of, use linux for speed,

windows for playing around / experimenting.

I did some testing a good while back ,  and found that running linux in a VM under windows was still faster than streight on windows.

 

Other things to look at for speed,

    raw CPU grunt is more important than No of cores.

     Disc speed is very important, use either a SSD, or best by 10 to 1 , a M2 PCIe disk.

     memory, dont have video that shares the PC memory, Have a seperate Video card with its own PCIe bus.

If your doing a lot of work, look at even a simple UPS,

    there is nothign more likely to cause pain than a run thats lasted a day, being killed at night by a power glitch.

 

 

<== If this was helpful, please feel free to give Kudos, and close if it answers your question ==>
Visitor sturmel2007
Visitor
1,209 Views
Registered: ‎07-06-2017

Re: Vivado linux vs windows benchmark

Jump to solution

Thanks drjohnsmith,

I'm seeking informations more formal about the difference between Linux and Windows.

As I readed, It seem that Linux is able to use more CPU in parallelle than Windows.  But in your case, You said that it was faster even if you run Linux in a VM under windows?!?!

I guess that Xilinx should have benchmark...

0 Kudos
Scholar dpaul24
Scholar
1,201 Views
Registered: ‎08-07-2014

Re: Vivado linux vs windows benchmark

Jump to solution

@sturmel2007,

I guess that Xilinx should have benchmark...

I guess they definitely do have......the question is whether they are willing to provide such an info in an open forum!

--------------------------------------------------------------------------------------------------------
FPGA enthusiast!
All PMs will be ignored
--------------------------------------------------------------------------------------------------------
0 Kudos
Scholar drjohnsmith
Scholar
1,182 Views
Registered: ‎07-09-2009

Re: Vivado linux vs windows benchmark

Jump to solution

I dont think Xilinx will have formal benchmarks of different performances,

 

The biggest question woudl be , what does one test with ?

    Your design will require very different resources to mine, so what is standard.

 

Xilinx do publish minimum requirments for the RAM and bits for different FPGA's

i.e. https://www.xilinx.com/support/documentation/sw_manuals/xilinx2019_1/ug973-vivado-release-notes-install-license.pdf

https://www.xilinx.com/products/design-tools/vivado/memory.html

 

 

<== If this was helpful, please feel free to give Kudos, and close if it answers your question ==>
Highlighted
Xilinx Employee
Xilinx Employee
1,169 Views
Registered: ‎03-29-2013

Re: Vivado linux vs windows benchmark

Jump to solution

There is no such compile time benchmark available, due to the diversity of systems. Even for Linux servers, compile time can vary depending on various aspects (CPU model, cache architecture, RAM latency/capacity, NFS performance). Windows machine usually have less memory or CPU resources, and run more unrelated jobs.

One important thing to control if/when comparing compile time between Linux and Windows is the default number of threads used by Vivado: it is always lower for Windows. For example, on Windows, Vivado 2018.3 shows the following message in the log file:

INFO: [Place 30-611] Multithreading enabled for place_design using a maximum of 2 CPUs

For Linux, the default is always 8, unless fewer threads (hyperthreads included) are available on the target machine, in which case Vivado reduces its default.

Also, Vivado synthesis default is set to 4 threads as it is able to launch parallel jobs under the hood instead of just relying on threads.

These max threads settings can be reported and changed:

% report_param *thread*

% set_param general.maxThreads 4

% set_param synth.maxThreads 4

0 Kudos
Visitor sturmel2007
Visitor
1,141 Views
Registered: ‎07-06-2017

Re: Vivado linux vs windows benchmark

Jump to solution

Thanks freferi to confirm that Windows uses less core by default.

 

So Xilinx do not have any clue about how faster Linux can be in % against Windows on the same system with the same design?

Usually, the first thing Linux guy's do is trying to find out this information to prove that Linux is better...

Often, they are right :-)

0 Kudos
Visitor sturmel2007
Visitor
1,122 Views
Registered: ‎07-06-2017

Re: Vivado linux vs windows benchmark

Jump to solution

I did a quick test since there is no informations about that anywhere on the net.

I have installed Linux along Windows on my Computer to test Vivado on these two OS.

Same Machine, Same design Start from scratch to Bit file generation.

Windows 3h30

Linux 1h30

Amazing...

I would like to have explanation from Xilinx about this huge difference!!!!!!

Best regards,

Scholar drjohnsmith
Scholar
1,114 Views
Registered: ‎07-09-2009

Re: Vivado linux vs windows benchmark

Jump to solution

Its all down to the OS overhead,

different anti virus has an amazing effect.

 

<== If this was helpful, please feel free to give Kudos, and close if it answers your question ==>
0 Kudos
Xilinx Employee
Xilinx Employee
1,093 Views
Registered: ‎03-29-2013

Re: Vivado linux vs windows benchmark

Jump to solution

Can you check how many threads were used in both cases? Is it the same? Antivirus can definitely slow things down on Windows.

0 Kudos
Visitor sturmel2007
Visitor
1,027 Views
Registered: ‎07-06-2017

Re: Vivado linux vs windows benchmark

Jump to solution

I tried to route with a RAM DRIVE but I didn't see any improvement.

So Hard drive access is not a problem.

0 Kudos
Visitor sturmel2007
Visitor
1,015 Views
Registered: ‎07-06-2017

Re: Vivado linux vs windows benchmark

Jump to solution

Conclusion:

I have installed Linux along Windows on my Computer to test Vivado on these two OS.

Same Computer, Same design Started from scratch to Bit file generation.

Windows 3h30 (With Antivirus)

Windows 2h15 (Without Antivirus)

Linux 1h30

Note: Using RAM DISK for the project files do not increase the processing speed.

0 Kudos
Xilinx Employee
Xilinx Employee
1,003 Views
Registered: ‎03-29-2013

Re: Vivado linux vs windows benchmark

Jump to solution

What is the number of threads used on Windows vs Linux?

0 Kudos
Visitor sturmel2007
Visitor
983 Views
Registered: ‎07-06-2017

Re: Vivado linux vs windows benchmark

Jump to solution

This information is not easy to find...

Task manager on Windows and System monitor on Ubuntu.

Some tasks takes only 1 core on both OS,

Some tasks takes multiple cores on both OS but it hard to see how much...

0 Kudos
Xilinx Employee
Xilinx Employee
954 Views
Registered: ‎03-29-2013

Re: Vivado linux vs windows benchmark

Jump to solution

Look for messages like the ones below to figure out the maximum threads used for a given flow step. It should cover key algorithms which take most of the compile time:

.../project.runs/synth_1/runme.log:INFO: [Synth 8-5580] Multithreading enabled for synth_design using a maximum of 4 processes.

.../project.runs/impl_1/runme.log:INFO: [DRC 23-27] Running DRC with 8 threads

.../project.runs/impl_1/runme.log:INFO: [Place 30-611] Multithreading enabled for place_design using a maximum of 8 CPUs
.../project.runs/impl_1/runme.log:INFO: [Physopt 32-721] Multithreading enabled for phys_opt_design using a maximum of 8 CPUs

.../project.runs/impl_1/runme.log:INFO: [Route 35-254] Multithreading enabled for route_design using a maximum of 8 CPUs

0 Kudos
Explorer
Explorer
949 Views
Registered: ‎03-17-2011

Re: Vivado linux vs windows benchmark

Jump to solution

are machines on linux and windows both the same?

--Sebastien
0 Kudos
Scholar dpaul24
Scholar
940 Views
Registered: ‎08-07-2014

Re: Vivado linux vs windows benchmark

Jump to solution

@sturmel2007,

Thanks for bringing up this thread.

I would be interested with the thread count info along with any other interesting data.

I also stick to my earlier comment that there are multiple example designs of varying complexity which are tested internally at Xilinx on Linux and Windows OS (machines with same h/w config). This is a kind of last level testing any good EDA tool should go through for internal use and records. As Xilinx is a huge company, so where this info on OS-testing resides might be unknown (or most probably hidden/meant to be hidden) even to forum moderators.

--------------------------------------------------------------------------------------------------------
FPGA enthusiast!
All PMs will be ignored
--------------------------------------------------------------------------------------------------------
Visitor sturmel2007
Visitor
892 Views
Registered: ‎07-06-2017

Re: Vivado linux vs windows benchmark

Jump to solution

Frederi, you the man to find this info !
We rely on you :-)

 

 

0 Kudos
882 Views
Registered: ‎07-23-2019

Re: Vivado linux vs windows benchmark

Jump to solution

As they have said, it depends on so many things that such a "benchmark" would have so many dimensions (cores, graphic cards, antivirus options, etc) chances are there will be lots of political make-ups in the figures, you know. I believe a dedicated Linux machine is faster than the clumsy fancy windows. For short synthesis times, I wouldn't mind if it takes 8 or 10 minutes. When it takes longer than a coffee cup, then I would consider incremental compile techniques.

0 Kudos
Visitor sturmel2007
Visitor
870 Views
Registered: ‎07-06-2017

Re: Vivado linux vs windows benchmark

Jump to solution

Hi archangel-lightworks,


Saying (I believe a dedicated Linux machine is faster than the clumsy fancy windows) do not help.

Engineers needs facts.  With real stats, we can take good decisions.

Stats from Xilinx.  It's what I'm seeking for since the beginning of this post and I'm amaze that this information is so hard to find....

Few people wants to migrate to Linux for many reasons so to migrate they need very good reason.  In this case a substantial increase of processing power.

On the other hand, one of your point was right, incremental compile techniques helps for any OS.

thanks :-)

 

0 Kudos
Scholar drjohnsmith
Scholar
861 Views
Registered: ‎07-09-2009

Re: Vivado linux vs windows benchmark

Jump to solution

well have to disagree me thinks

saying "With real stats, we can take good decisions"

is our point,

you only get real stats if you define the paramiters of the test,

   Just look at machines, there are effectivly an infinate number of paramiters.

Your choice of machines and enviroments is likely very differen to mine,

     so your numbers will be of little use to me over a generalisatoin,

whihc is what we have alreayd shared with you

If you want real engineering numbers,

    I look forward to hearing your results,

 

<== If this was helpful, please feel free to give Kudos, and close if it answers your question ==>
0 Kudos
Visitor sturmel2007
Visitor
850 Views
Registered: ‎07-06-2017

Re: Vivado linux vs windows benchmark

Jump to solution

I dont see the problem of the parameters since the only thing that changes is the OS.

Same Computer, same Vivado version, same design started from scratch to Bit file generation.

There is few things I can do more since I'm not a Vivado developper...

As a customer, I already did more than many does.
I asked to have benchmark or statistics from Xilinx about the processing power on both OS with Vivado.  Without answer, I did my own LINUX installation and did a place and route  of one of my project on both OS (Windows and LINUX) on the same machine.

I gave my results that shows a difference between both OS.

Normally, Xilinx or at least, the developpers shouls have these informations...
Anyway, on my side, I never asked to my customers to give me stats on my own products...

Now I'm waiting for Frederi to seek for more information's about this subject.

I saw with his questions that he really want's to understand and I'm greatful for that.

Since he works for Xilinx, he should be able to have contact with Vivado developpers :-)

0 Kudos
842 Views
Registered: ‎07-23-2019

Re: Vivado linux vs windows benchmark

Jump to solution

Well, you know, Xilinx is a big company, Microsoft as well, they both sell things and CEOs don't go shooting their neighbour's feet without a good reason... so you will probably never hear from Xilinx that Vivado, as a general rule, runs faster in Linux. That's something you try, you find it, they will say "oh, well, congratulations, you have a very optimized environment that in some cases beats Windows, well done, enjoy it" but they are not going to generalize in that direction. 

0 Kudos
Visitor sturmel2007
Visitor
838 Views
Registered: ‎07-06-2017

Re: Vivado linux vs windows benchmark

Jump to solution

We will see but I have faith on Frederi.
He may be able to give us info about that
:-)

 

0 Kudos
Scholar drjohnsmith
Scholar
833 Views
Registered: ‎07-09-2009

Re: Vivado linux vs windows benchmark

Jump to solution

@sturmel2007 

yo right , you have done a lot  more than others on this , and I congratulat you on that

But you have to note, even those small test , got a few questoins re options / configuratoin,

   what I call the "did you try X catagory"

Can yo imagine how big that chorous would be if you had publisehd say 3 different machine configuraitions,

e.g.

 SSD / Hard drive

More / less RAM

Which anti virus / how set up

 

Its just not posisble,

 

Imagine if you did some IP, that ran on many different chips,

   how would you specify the resources and speed ?

   and no matter what you bench marked against, the question woudl always be

 

a) what about chip X

b) I can' t get that speed in chip X, why not.

 

 

 

 

 

 

 

<== If this was helpful, please feel free to give Kudos, and close if it answers your question ==>
0 Kudos
Visitor sturmel2007
Visitor
819 Views
Registered: ‎07-06-2017

Re: Vivado linux vs windows benchmark

Jump to solution

Thanks, drjohnsmith,

I understand and agree for the IP with different devices.

 

On my side, this case is closed.

I won't work more on that.

I'll stick to my conclusion and I'll work with Linux version until now.

 

But I would be pleased if any additionnal information's could be added on this topic.

 

best regards,

0 Kudos
Scholar markcurry
Scholar
798 Views
Registered: ‎09-16-2009

Re: Vivado linux vs windows benchmark

Jump to solution

A seperate, but related query - even though I set things up (according to UG904) to run multi-threaded (on a linux server), almost ALL of my synthesis jobs include the following on the transcript:

Warning: Parallel synthesis criteria is not met 

I've never been able to find any infomation as to why this is, and why my designs don't qualify for "Parallel synthesis".  Can anyone from Xilinx comment as to what is neccesary within a design to qualify for "Parallel synthesis"?

Thanks,

Mark

 

0 Kudos
Visitor sturmel2007
Visitor
762 Views
Registered: ‎07-06-2017

Re: Vivado linux vs windows benchmark

Jump to solution

@drjohnsmith,

Yesterday I watched, again, your links about memory used by Vivado.
https://www.xilinx.com/products/design-tools/vivado/memory.html

Thank you very much!
I didn't saw the importance of it but this information at this moment but this is the most interesting on this post.

In this case, Xilinx did a comparison between Windows vs Linux.
I see a direct link between Memory uses and processing speed.

In this document, we can see that Linux uses roughly twice more memory.  In the mean time, my tests showed that the processing time seem to be roughly a bit less than half.

So we can clearly see the inversely proportional relation between memory usage and processing speed.

It's too bad that Xilinx didn't put the processing time into this document since they did exactly what I wanted to see....

Now I'm happy with this conclusion since I really understand why Vivado under Linus is way faster!!!

Thanks to all :-)

0 Kudos
Scholar drjohnsmith
Scholar
760 Views
Registered: ‎07-09-2009

Re: Vivado linux vs windows benchmark

Jump to solution

Thank you.

 

be careful about drawing a conclusion from a corelation,

Th eexample I use is that "bad breath causes Lung cancer is a clear and defined correlation".

Which is is ,

If you dont relaise that the bad breath is for those that smoke,

    and its the smoking thats causing the bad breath and cancer.

 

But yep, most machines nowdays ( apart from the companies I seem to work at )  have plenty of memory , and linux does run faster than Windows for vivado.

 

 

<== If this was helpful, please feel free to give Kudos, and close if it answers your question ==>

View solution in original post

Visitor sturmel2007
Visitor
727 Views
Registered: ‎07-06-2017

Re: Vivado linux vs windows benchmark

Jump to solution

@drjohnsmith,


Oups...

I did a mistake putting your last comment as the solution...

Is it possible to change the solution to my conclusion so the people will be able to see the real conclusion not a comment...

Rookie mistake :-)

0 Kudos