Date: Thu, 12 Aug 1999 03:51:20 -0700 (PDT) To: clash@tasam.com (Joe Gleason) Subject: Re: On freezes in 3.2-Stable In-Reply-To: <000e01bee4a7$d5d94550$0286860a@tasam.com> from Joe Gleason at "Aug 12, 1999 05:48:21 am"
| previous in thread | raw e-mail | index | archive | help
[Charset iso-8859-1 unsupported, filtering to ASCII...] > Other than price, is the a good way to tell the diffrence between 16 and 36 > chip modules when shopping? > > What manufacture/dealer do you deal with? I won't disclose that as they wouldn't due business with you at these quantities anyway so it would do you little good to know. I can say that the chips are always from major manufactures, Micron, Seimens, Hitachi, Fujistu, NEC, Samsung etc. > I am currently looking to upgrade my server that has 2x 256mb modules (the > 36 chip kind). > > Also, I am not on location so I can't check myself, but am I correct in > assuming that 36 chips would be for parity/ECC and 32 for non-parity? And > the same with 18/16? Yes, that would be the correct assumption. Note that in lower capacity parts you get into 8 and 9 chip or the oddball 3 and 12 chip versions of these things (on DIMMS that would only occur on newer parts that are 4 or 16MB if I did the math in my head right). > Do you have any idea what the "safe" threshold is for number of chips? The typical memory bus designer uses the standard current best practicies when determining the bus load that he is anticipating for the design. This leads to the conclusion that 16/18 chips per module as the limit. There is a reason that 36 chip modules will work if you stick 2 of them in a board that has 4 dimm slots, you only putting a total of 36 chips on a bus that was designed for 36 chips: (4*18==2*36, magic ain't it :-)). > Can you mix 36 chip and 18 chip modules in the same server? Generally not, the big thing to remeber is to keep the total chip count under the design load of the memory drivers, and today that is typically calculated as number_of_dimm_slots * standard_dimm_chipcount, nominal numbers are 4 * 18 == 72 chips total. If you have in EE background you'll note that this is getting upwards of a 100pF load, something pretty hard to drive at a 100MHZ without huge driver transitors and a lot of current! > Sorry for all the questions, but you seem to know what your talking about. > ;-) Having been in memory system design back in the days of the 16K to 256K DRAM chip doing 2+MB boards that where 12x18" gives me a pretty good clue as to what goes on with memory designs... and I keep up on all the latest in technology. Just wait till RAMbus hits in volume, boy are we going to have some ``flakey'' boxes caused by some of the poor design work done by the off shore groups. > > > > > 1. Supermicro can't handle a full 1GB, despite their specs > > > > > > $100 says that's it. > > > > $200 says that you using low density 256MB DIMMS that are overloading > > the BUS specification of the chipset on the motherboard. > > > > Current price ratio (well, okay 7/30/1999) on 18 chip vs 36 chip > > 32Mx72b ECC PC100 8nS is still 1:2, or about $300.00 vs $600.00. > > > > Alls I can say to those folks trying to do more than 512MB in a box > > with any current chip set is ``You have to spend the money for 18 chip > > parts if you want to have any type of reliable operation, no current > > chipset was designed to drive 144 chips'' > > > > Between David Greenman and my self we probably went through $25,000 > > of parts from various manufactures who claimed that they would work > > before we finally decided they are all wrong, and that we would just > > have to wait for the next generation of chips that would allow 18 chip > > 256MB DIMMS. > > > > If your memory vendor doesn't know what 18 chip 256MB DIMMS are drop > > me an email, they are in ready supply from either manufacture we deal > > with. All parts are lifetime warrantied. > > > > -- > > Rod Grimes - KD7CAX - (RWG25) rgrimes@gndrsh.dnsmgr.net -- Rod Grimes - KD7CAX - (RWG25) rgrimes@gndrsh.dnsmgr.net ----- End of forwarded message (env-from rgrimes) ----- -- Rod Grimes - KD7CAX - (RWG25) rgrimes@gndrsh.dnsmgr.net To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?>