Prev: Xilinx project failed timing constraints
Next: Xilinx FFT core -- Is varying precision through the core possible?
From: Jason Thibodeau on 6 May 2010 11:47
On 05/06/2010 10:44 AM, Petter Gustad wrote:

> In my experience the Linux and Windows versions run at pretty much the
> same pace.
>
> Petter

My completely unscientific, 'seat of the pants' comparisons have found
the same result.
--
Jason Thibodeau
From: General Schvantzkoph on 6 May 2010 14:31
On Thu, 06 May 2010 15:18:46 +0100, Nial Stewart wrote:

>> The parallel processing modes only work in Linux, at least up to ISE
>> 11. I don't know if they've added parallel support for Windows in ISE
>> 12. Quartus 9.1 has parallel processing support on Linux, I've never
>> used it on Windows so I don't know if it also has parallel support on
>> Windows.
>
> 'General',
>
> Do the parallel processing modes make much difference in your
> experience?
>
>
>
> Nial.

I'm using the full parallel mode on Quartus right now and I'm getting
pretty good times on an iCore7 machine. However I haven't taken the time
to benchmark it so I don't actually know if it helps. I don't use Windows
except in a VM to run Word and Quickbooks so I don't have any numbers
that I've generated myself. Xilinx FAEs have told me that the Linux
versions of ISE are faster, Altera FAEs have told be that the Windows
version of Quartus is a little faster. However I don't know how true
either statement is with the latest versions. The code base for the core
applications is identical, it's only the GUI code that's different, so
you wouldn't expect a giant difference. Linux had a huge advantage until
recently because it has been 64 bits for years, until Win7 very few
people were using a 64 bit version of Windows. Large Xilinx FPGAs require
a lot of RAM, I've found that the V5 300s require about 10G, so 64 bits
is an absolute requirement for the big parts which made Linux an absolute
requirement.

From: whygee on 7 May 2010 00:43
hi !

General Schvantzkoph wrote:
> Linux had a huge advantage until
> recently because it has been 64 bits for years, until Win7 very few
> people were using a 64 bit version of Windows. Large Xilinx FPGAs require
> a lot of RAM, I've found that the V5 300s require about 10G, so 64 bits
> is an absolute requirement for the big parts which made Linux an absolute
> requirement.

Does it mean that... Win7 has killed Linux ? :-D

yg
--
http://ygdes.com / http://yasep.org
From: Chris Maryan on 7 May 2010 10:19
On May 5, 11:49 pm, Eric <delage.e...(a)gmail.com> wrote:
> Hi,
>
> Does someone have some benchmarks comparing the compilation time
> between the Windows 64b and Linux 64b editions of the Xilinx ISE
> Design Suite? I need some arguments to invest in the right development
> platform.
>
> Many thanks.
>
> Eric

This is somewhat anecdotal, as I almost always use 32bit ISE on a 32
bit windows XP machine, but I have tried out some other combos and
this is what I've got:
- This applies only to ngdbuild/map/par/bitgen, I don't use XST.
- The last time I tried this was circa ISE 11.1 I think.
- 64bit ISE on a 64bit windows XP machine is 2-3 times slower in map/
par than 32 bit ISE on the same 64 bit windows XP machine
- 32 bit ISE is essentially identical in performance on both 32bit
winXP and 64bit winXP
- 64 bit ISE inherently uses much more memory than 32 bit ISE
- 64 bit ISE allows you to use more memory (for very large designs)
- Supposedly, 64 bit linux ISE is more on par with 32bit win ISE in
terms of performance than 64 bit win ISE

Chris
From: General Schvantzkoph on 7 May 2010 12:54
On Fri, 07 May 2010 06:43:54 +0200, whygee wrote:

> hi !
>
> General Schvantzkoph wrote:
> > Linux had a huge advantage until
>> recently because it has been 64 bits for years, until Win7 very few
>> people were using a 64 bit version of Windows. Large Xilinx FPGAs
>> require a lot of RAM, I've found that the V5 300s require about 10G, so
>> 64 bits is an absolute requirement for the big parts which made Linux
>> an absolute requirement.
>
> Does it mean that... Win7 has killed Linux ? :-D
>
> yg

Of course not, Linux is a vastly more productive environment. The high
performance simulators, NC and VCS, are Linux only. It's vastly easier to
use multiple machines, I'm currently sshed into a 4 core i7 and two Core2
boxes that are running Verilog simulations and Quartus builds. I'm also
sshed into a remote system that has SignalTap running. I'm doing the
Quartus builds on the i7 and I'm rsyncing the results over to the remote
system. Try doing any of that with Windows. When I run Xilinx tools I do
that with scripts which is easier to do on Linux although the scripts can
run on Windows if you have Cygwin installed.

First  |  Prev  |  Next  |  Last
Pages: 1 2 3
Prev: Xilinx project failed timing constraints
Next: Xilinx FFT core -- Is varying precision through the core possible?