One of the issues we encountered designing a small virtualization farm was that as you add more DIMMs to the system, the clock speed of the RAM had to go down - addressing more addresses over a limited bandwidth meant ou traded of speed for capacity...
we chose capacity - and lower speed, lower cost RAM - which worked out for our use case (a few large high capacity VM's). If we had been planning one of the bioinformatics grid compute modes, it would have been a different story... On 20 May 2018 at 01:17, Russell Coker via luv-main <[email protected]> wrote: > https://en.wikipedia.org/wiki/DDR4_SDRAM > > At the LUV meeting today someone asked me how servers can have so much > RAM. > Firstly the above page is worth reading, DDR3 (which is in most desktop > PCs in > use now) can have a maximum of 16G per DIMM, some PCs only have 2 slots, 4 > is > very common, and 6 is reasonably common for high end systems (I own a > couple > of those). 6*16G=96G as the theoretical maximum in a high-end DDR3 > desktop > system and 6*64G=384G as the theoretical maximum of a DDR4 desktop, of > course > the BIOS or chipset might not support so much. > > https://en.wikipedia.org/wiki/List_of_Intel_chipsets > > The above page lists Intel chipsets and gives the maximum RAM supported. > A > lot of the Core 2 family were limited to 4G of address space, NB that's > NOT 4G > of RAM, that's 4G including address space for video cards etc - so 3.25G > was a > common amount of usable RAM on such systems. The Wikipedia page doesn't > give > information on the RAM limits of the i3/i5/i7 systems. > > The DDR4 Wikipedia page says that one of the benefits of DDR4 is a maximum > module size of 64G. A quick check of my favorite PC parts store showed > that > they don't sell DDR3 larger than 8G modules or DDR4 larger than 16G > (preumably > I could get larger via mail order). If I got a motherboard that took > 6*DDR4 > DIMMs then I could have 96G of RAM, but that would cost me 6*$295=$1770 so > I'm > not about to do it. > > http://www.dell.com/en-au/work/shop/povw/poweredge-t640 > > The Dell page for the PowerEdge T640 says that it has 24*DDR4 slots for up > to > 3TB of RAM with a caveat that the 128G modules aren't available yet (as an > aside it seems like the DDR4 Wikipedia page needs an update in this > regard). > > https://en.wikipedia.org/wiki/Fully_Buffered_DIMM > > Threre are significant engineering issues related to supporting large > numbers > of DIMM sockets. The above Wikipedia page is a good place to start if you > want to learn about how server RAM is different from (and incompatible to) > desktop RAM because of such issues. > > https://en.wikipedia.org/wiki/ECC_memory > https://en.wikipedia.org/wiki/Hamming_code > > Another difference with server RAM is the use of Error Correcting Codes > (see > the above Wikipedia pages). Server RAM has extra bits in each word to > store > codes that allow single bit errors to be detected and corrected and double > bit > errors to be detected. It's a really good feature to have if you don't > want > your data to be corrupted. It's also supported in high-end workstation > systems and some systems have support for both ECC and non-ECC RAM. ECC > RAM > is usually "buffered" and won't work in systems that take "unbuffered" RAM > (IE > all the desktop systems that don't use ECC RAM). There is unbuffered ECC > RAM > for low-end server systems. Apart from buffered vs unbuffered there's > apparently no reason why ECC RAM shouldn't work in non-ECC systems, > although > when I tried this back in the P4 days (before buffered RAM was invented) > it > didn't work. > > One of the more dedicated members of this list got a free server system > from > LUV and uses it as his personal workstation. It has something like 96G of > RAM > but makes more noise than most people want in the same building they are > in. > There's no reason why you couldn't design a system with 24 DIMM sockets > that > doesn't sound like an aircraft taking off, but most people who want so > much > RAM have a soundproofed server room. > > -- > My Main Blog http://etbe.coker.com.au/ > My Documents Blog http://doc.coker.com.au/ > > > > _______________________________________________ > luv-main mailing list > [email protected] > https://lists.luv.asn.au/cgi-bin/mailman/listinfo/luv-main > -- Dr Paul van den Bergen
_______________________________________________ luv-main mailing list [email protected] https://lists.luv.asn.au/cgi-bin/mailman/listinfo/luv-main
