Currently we use Windows and 32 bit Linux versions of ISE. We are looking for a way how to increase speed of design implementation. Has someone done any real-world comparison of implementeation (synthesis + map + p&r) run times between 32-bit and 64-bit ISE for Linux? Will upgrading to 64 bit Linux machines bring any shortening of design implementation run time? And if so, how much? At least some rough estimation, please.
I would welcome something like... design implementation runtime on Athlon XP 3000+, 2GB RAM was 30 minutes, after upgrading to Athlon A64
3000+, 2GB RAM the same design took just 20 minutes to implement;)
The thing that makes the most difference, at least for simulations, is cache. The 3000+ has a 512K cache, get an A64 with a 1M cache and you'll see a huge difference. NCVerilog runs twice as fast on my 3400+ (1M cache,754 pin) as it does on my 3800+ (1/2M cache, 939 pin).
It took me a very long time to install ISE 64 bit. I did not have RHEL, and it would not install on Fedora Core 3 x86_64. I was only able to install the 32 bit version.
Finally I have downloaded CentOS, which is equivalent to RHEL, but built by some group from the sources and than freely distributed.
I went a bough a separate drive for this process, as I really did not want to screw up a perfectly working setup. I installed CentOS on the new drive, than ISE 64 bit, and the latest patch
7.1.02i ... (btw, centos made my dual opteron system slower than my 750 MHZ lab notebook - I winder if RHEL is the same way).
Anyway after the installation, I rebooted in Fedora Core 3, mounted the new drive, and copied the newly installed ISE
7.1 64 bit to my usual tools location. After installation, ISE 64 seems to works just fine under Fedora Core 3.
So, was it all worth the effort ? Here are two test results. I have scripts where I can switch the actual version of ISE and EDK that I use. I did two trial runs:
ISE 7.1 32 bit ============== Virtex 4 lx60 (15% full) Elapsed time: 13:27 virtex 2-1000 (66% full) Elapsed time: 10:36
ISE 7.1 32 bit ============== Virtex 4 lx60 (15% full) Elapsed time: 13:39 Virtex 2-1000 (66% full) Elapsed time: 11:24
So speed wise, there was virtually no difference at all. Since I only have 2G of memory, using the 64 bit version did not add any other advantages as well. I did not have the time to see if I can now use chip scope and program devices directly from the compute server. That might be the only advantage, to have native 64 bit modules which are required if you want to use you Parallel Cable under linux.
Now I hope Altera can send me a copy of Quartus for Linux x86_64, so I can try that out as well !
Hope this helps, rudi ============================================================= Rudolf Usselmann, ASICS World Services,
Your Partner for IP Cores, Design, Verification and Synthesis