Date: Wed, 30 Aug 1995 11:33:14 -0700 (PDT) From: "Rodney W. Grimes" <rgrimes@gndrsh.aac.dev.com> To: marino.ladavac@aut.alcatel.at Cc: hardware@freebsd.org Subject: Re: Upgrade to my machine Message-ID: <199508301833.LAA08959@gndrsh.aac.dev.com> In-Reply-To: <9508301429.AA11372@atuhc16.aut.alcatel.at> from "marino.ladavac@aut.alcatel.at" at Aug 30, 95 04:29:02 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> > > > > > > > -Vince- stands accused of saying: > > > > Hmmm, what about machines in terms like SUN's, HP's will the P90 > > > > compare to since the Alpha is a fast machine. > > > > > > Depends lots on what you're doing with them; in a straight line, the P90 > > > is pretty quick, but what you put around it largely determines how it will > > > perform in an applications context. (Especially memory/cache/disk) > > > > > > >> Anyone who does big models of any sort uses huge amounts of memory, > > > >> as Rod already observed. > > > > > > > > That's true but who would actually need a gig of ram? > > > THINK for a minute about large applications. An Intel Pentium 90/100 CPU > > chip as 3.3 billon transistors on it. Each cmos transitor takes at least > ^^^^^^ > Minor nit. > > Million. Otherwise, we would have been having at least > 500 megabyte RAM chips for several years now. Cheaply. > > I wish we did :) I will publically admit I may very well be in error here, 2 other folks have sited references that state the number is 3.3 million, and I went and checked 4 references here on this and they all say millon, but I am really having a hard time fitting the Pentium design into 3.3 million transistors from a few simple calculations. 16k of static sram takes >768K transistors assuming a 6 transistor cell, now perhaps the cache is done in quasi-static 4 cell cmos sram, but then how do they do stop clock ??? Perhaps the cache is invalidated when coming out of stop clock mode, or maybe that is what happens during the cycles it takes to shut down. Also the claims that the RISC core of the 486 is in excess of a million transistors in the 1994 data book sets makes me wonder how one could do 2 of these for the dual issue super scaler pipe in anything less than 2 million transistors. Now we need to add in the MMU, TLB, and BIU to my already almost 2.7 Million transistors, okay, 600K transistors, 200K gates, yea, I suppose I might squeeze it into that... oh, and the microcode, that must be atleast 80 bits by 4k deep, but thats not much given ROM is 1 transistor per bit, plus decoders (and 4k decoderes aren't much when you use TGMX's :-)) Perhaps the 3.3 million transistor count is the RISC super scaler core excluding the cache, MMU, TLB and BIU _and_ microcode. I don't directly consult for the processor group at Intel so I don't have ready access to the right data. Published data like this is unclear as to exactly what it is. Anyway, the real point I was trying to make is there are things that get _HUGE_ in the CAD applications with respect to memory needs. I have seen single chip logic synthises runs go well into the 1.2G of data segment size on a day in day out basis, and so many other similiar sized things in my 15 years as a design automation specialist that it would make the average PC unix user's head spin when you look at a 1G main memory and 8G of local disk plus 1TB of back room disk, and a room for of engineers screaming that `we need more!!!''. I've watched an MIS group run and hide 2 weeks after we installed HSPICE on a 3084 and the engineers fired up some rather large test jobs to see if MIS's claim that the 3090 could do what we where doing on DN10000's just as well. Needless to say the 3090 buckled under once the job went to allocate the 400MB needed for state storage of the simulation since it was an MIS configured machine sporting all of 128MB of main memory. Needless to say, the purchase request for 14 additional 768MB DN10000's went past MIS after this without blinking an eye :-). > /Alby > > > 6 rectangles to represent the minimal transitor data and 3 contacts to hook > > it up, now thats 19.8G assumming I can stuff a rectange into a byte :-). > > > We haven't even started to talk about interconnecting these 3.3 billon > > transistors... > > > Can you say that a gigabyte in this world is actually a very small amount > > of data! I will hold by that statement, been there too many times. -- Rod Grimes rgrimes@gndrsh.aac.dev.com Accurate Automation Company Reliable computers for FreeBSD
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199508301833.LAA08959>