I was considering purchasing an AMD64 quad-processor system, but have no idea whether any of the FPGA development tool software would take advantage of multi-processors (Synplicity for Lattice, Actel, Xilinx, etc). Any ideas or recommendations?
There are no Xilinx tools that would take advantage of such hardware. I suggest you spend your money on a faster chip with more cache rather than more CPUs.
the primary cpu bound process for Xilinx, par, can be run in a multiprocessor and cluster configuration for converging the best routing solution sooner. Do a serch for MPPR in your par documentation or the Xilinx answer base. At least on older ISE 6.x releases under linux this is VERY useful.
I suppose it depends on your Xilinx tool usage. If you develop in the hardware itself as some people do, the best routing is rarely needed. Hence, this feature of Xilinx is hardly useful in that case. If you develop large projects using software simulation, or if you make a small change to a large project that barely passes timespec on a regular basis, perhaps this is useful, but I remain unconvinced.
I've been using dual/quad/octal systems for UNIX/FreeBSD/Linux desktops for about 15 years. Learning to use threads, MPI, PVM is an important programming resource for many classes of difficult problems which can be parallelized, and can be a valuable resume builder.
I've made part of my living over the years writing device drivers, where having SMP machines for testing multiprocessor locks is a requirement. As a hardware developer of system boards, it's very useful to learn to write your own device drivers and diagnostics for system bringup, rather than waste a lot of time resources educating software only guys how to program your board (besides learning where to make hardware/software tradeoffs for performance or cost).
SMP machines as a UNIX/Linux desktop are much more responsive when doing background compiles, PCB routing, FPGA routing, etc as your interactive desktop session doesn't need to fight for processor access.
Justification of the increased cost depends greatly on your use. As more machines are hyperthreaded, have multiple cores and dual/quad systems are getting cheaper, SMP aware programs are getting more common and expected. This should also show in EDA/CAD software over time as well. I have several friends that do high end gaming on dual booted (linux/xp) systems, and feel the SMP gives them that FPS gaming edge.
So ... good luck with whatever new machine you get.
PAR is multithreaded but none of the simulators are so it's not really cost effective to get an Opteron 8xx system. A couple of Athlon X2 4400+ systems are a much better choice. For simulator performance the cache size is the most important thing. NCVerilog is twice as fast on a 1M cache Athlon 64 as it is on a 1/2M cache system. I haven't updated this page in a while so it doesn't have the 4400+ on it, but these benchmarks are still valid,
The 4400+ is about 10% faster on single threaded jobs then the 3400+ (both have 1M caches and run at 2.2GHz, but the 4400+ has two DDR400 memory buses while the 3400+ has a single DDR333 memory bus). The 4400+ is able to use both cores efficiently, I frequently run NCVerilog and PAR at the same time.
YIKES, that thing must be a beast!!
I feel so left behind with my XP 2500 @ 2.2Ghz. It is all I can afford for now. It seems fine for me though. But then again, I am not making the huge designs you guys are.
-Isaac
Look at the -m option for par. The 7.x documentation states that it is not available under windows, so it may be of use when running under Sun Solaris and Linux(? not sure about Linux; check with your local Xilinx FAE and/or their hotline support).
-bob
p.s. as side note: there are two versi> R> > I was considering purchasing an AMD64 quad-processor system, but have no
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.