Welcome to our site! EDAboard.com is an international Electronics Discussion Forum focused on EDA software, circuits, schematics, books, theory, papers, asic, pld, 8051, DSP, Network, RF, Analog Design, PCB, Service Manuals... and a whole lot more! To participate you need to register. Registration is free. Click here to register now.
Blade 100 is cheap one , but its performace is not good..
maybe good for IC5 , but not good for hspice-like tools, it will take you
long times to finish your spice job.
the same price for x86 , you can get a p4-2.8g or even more power when
choose amd .
when choose p4-2.8, it computing power is about almost 4~5 times of
blade box ..
and blade its computing power is about the same as intel's p3 600 or 700
..
maybe you can wait ic5 for linux ...
well, it depends on how big your design is. However, the best solution, in my opinion, is mixed PCs and Sun boxes. That is also the solution at my company. Thus, the optimun solution is how to define your methodology base on your tools and resources.
i.e
+ A cheap Sun box (ultra 1 200NHz, 512MB, 2x9GB HD about $150 on ebay.com) for slow stuff, i.e standard cell layout (virtuoso), cell characterize (hspice, spectre). You may use the SunBlade100 for synthesize for better performance(since the synopsys tools are not really great in x86 arch)
+ A fast PC (dual OS WinNT and Linux) for simulation (hspice, hsim), synthesis (synopsys, ambit) and auto route engine (SOC).
hi,
I am using blade1000, 1G mem. And I feel it's slow for synthesis and postsim. So I think blade2000 is good choice.
by the way,who has used x86 for IC design ,how about it?
I have done a compare with a x86 linux machine with a SUN blade 100 workstation on DC, primetime and nanosim.
The PC is dual AMD athlon MP 2000 with 2GB DDR266RAM, a SUN blade 1000 with single 950MHz CPU with 2 GBRAM.
On sythesis, the x86 only take 1/3 time to complete the physical synthesis. On primetime and nanosim, x86 as least 4 time fast than SUN blade 1000. They all use same version of software on different OS (linux and Solaris).
However, they hardware price difference. The PC only cost a little less than US$2000, but the SUN cost US$25000 !
I agree with you guy all about the cheap and high performance of PC compare to Sun machine. But remember, EDA companies are very conservative:
-They are reluctant to release WinNT or linux versions so you have very few limited option.
- There are none or few EDA application with 64 bit capability.
There is a big difference between 32bits app and 64 bits app. The limitation of 32 bits app is the output limit to 4GB (or 2GB), the 64 bits app limit is the sky !
So Blade is a good choice. Hey you guy are lucky, I still stuck with my Ultra5 WS with a Ultra1 is the server!
Sun is losing the EDA market because of the poor performace and much higher cost compared with x86 platform. More and more EDA vendor will support linux platform, not only because customers request but the big guys IBM,HP they are migrating their product to support linux. For 64bit OS capability, AMD
K8 will be production this year, and 64bit linux is on the way.
It's true for the future. But now most people have to stuck on Sun. EDA vendor doesn't care much about your cost of hardware. they feel safer (?) when their product installed in a IDed machine (Sun, HPUX, IBM...)
May somebody give ssome comment about signle-CPU vs dual-CPU on x86 machine? My target EDA tools are Hspice
and Nanosim/HSIM. Besides, for the same grade of CPU rating, AMD XP 2.8G vs P4 2.8G, which platform is better in terms of speed ??
Some company stuck with SUN is because of historical reason. The company have long history on software and design, they need some machine to compatible with their old software and library.
Btw, it is true that the big company drive the EDA software vendor to linux. It is because of price and performance.
I don't agree with that 64bit apps much faster than 32bit apps. Because the 64bit apps mainly benefit on database software but not EDA software. It is more important on floating point performance than 64bit addressing space
- almost every major vendor (except Cadence, hopefully in a while) supports Linux
- x86 machines are better and cheaper than their low-cost SUN competitors (e.g., Blade 100). Low cost SUN machines, in fact, are only not-so-reliable PCs with a SPARC processor (by far les powerful than Pentium).
- x86 dual CPU are, indeed, a very good choice but only if your software supports it (e.g. Mentor calibre)
My past experiment result
AMD > Intel >>>> Sun(64bits solaris>=2.7) > Sun(32bits solaris <2.7)
for multi-cpu , it's useless . The load is usually totally un-balance .
It's more suitable for "mutil users" not "mutil thread" .
downgrade speed for new version i found :
hspice 2000.4 > 2002.2
hsim 1.3 > hsim2.0
The most old guys (usually senior or your boss) still live in the 1980 for those past good days. They always think the SUN/HP is the only way to build asic.
what i would like to say , if all eda vendor support x86 , your sons and daughter would know SUN only in museum .
I think every low budgeted engineer should run out and buy a sun blade 100. I got one and love it. I run apollo, dc_shell, primetime, star_rcxt, caliber, and vcs on it and for most part am satisfied with it's performance. I added 2 gig of ram and a 100 gig hard disk to it to beef it up though. One thing I especially like is that I can run the same tools and scripts on it as I use at work. You might want to keep this last point in mind if your work requires you to use solaris operating system for your design work.
I don't agree with that 64bit apps much faster than 32bit apps. Because the 64bit apps mainly benefit on database software but not EDA software. It is more important on floating point performance than 64bit addressing space
The main difference between 32bit and 64bit is not speed but the OUTPUT CAPACITY. For example, if you run a long simulation with a big chip, your result will be chop off at 4GB limit. If you run a P&R tool a big chip, the job will be crash and not finish. 64bit is the solution for big design. application
Not really, you can speed up a lot with multithreading. For example with Hspice 2CPU can speed up 1.67 times and 6 CPU can speed up 3.0
(Hspice Manual page 3-60). A lot of application such as Star RCXT, Hercules, Astro all need multithreading to speed up big job
Nobody said:
downgrade speed for new version i found :
hspice 2000.4 > 2002.2
hsim 1.3 > hsim2.0
.
I recognize that too by running tutorial and compare speed and memory consumption. I haven't tried the big job yet so maybe our conclusion only right with small circuit. Maybe big circuit and more simulstion cycle will favor the later version
I'm running Cadence suite and Hspice under Blade150/650MHz. I feel process a bit run unevenly though a small circuit. I wish high speed x86 with IC5.0 could perform it better.
Anyway, for a limited budget and specific SUN's application, it is one of a good choice.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.
