rone: (Default)
[personal profile] rone

We're using a particular Tyan board on many of the workstations at work.  It maxes out at 4 RAM sticks of 1 GB each, for a total of 4 GB.  However, upon booting, the OS (and memtest86) would only find 3296 MB (despite the BIOS claiming "4096 MB OK").  I updated the BIOS and that let us have 128 MB more, for a total of 3424 MB.  A friendly Google search yielded information that suggested we enable the "software memory hole" in the BIOS setup; apparently, these things reserve space between 3.5 GB and 4 GB for god knows what, and enabling the setting frees that up.  Upon enabling the SMH, the RAM test during bootup shows it happily counting up to 3072 MB, freezing for a second, then counting up from 4096 MB to 5120 MB, and upon reaching that, going back to display "4096 MB OK".  The OS now admits to seeing all the RAM.

(Oh, and i didn't even go into the near-impossibility of making a bootable CD that included the BIOS Flash file and utility that worked.  We ended up finding a floppy drive and disk rather than suffer and keep dumping coasters.)

It's a good thing that my livelihood doesn't depend on understanding computers, because god damn i hate the fuckin' things.

Date: 2006-10-18 12:49 am (UTC)
From: [identity profile] dpk.livejournal.com
That's PAE. The upper part of the first 4GB of memory space is used by some PCI devices or something like that, and the actual RAM within that space is remapped above the 4GB barrier. That's why you're seeing it count to 3072MB and then 5120MB.

Some OSs will report that you have 5GB of RAM, too. Here's what a FreeBSD 5.4 box (manually patched to get around a severe bug the FreeBSD team didn't fix til 5-STABLE or something) says it has on a SuperMicro server with 4GB of RAM:

real memory = 4831838208 (4608 MB)
avail memory = 4194848768 (4000 MB)

All in all it's pretty obnoxious behavior, but I guess it's necessary to address the additional RAM. I don't know how it works exactly, but I think I'd prefer all real RAM be remapped above 4GB so it's consistent.

Date: 2006-10-18 12:52 am (UTC)
ext_8707: Taken in front of Carnegie Hall (thanks)
From: [identity profile] ronebofh.livejournal.com
As ever, you validate the existence of alt.fan.dpk.

Date: 2006-10-18 12:55 am (UTC)
From: [identity profile] dpk.livejournal.com
Nooooo. How embarassing. Heh.

Date: 2006-10-18 01:28 am (UTC)
From: [identity profile] mskala.livejournal.com
It occurs to me that that sounds pretty much exactly like how the 640K limit worked on 8086 PCs.

Date: 2006-10-21 04:17 pm (UTC)
From: [identity profile] cks.livejournal.com

The PCI issue is that basic PCI only has 32-bit address lines, and PC custom is to map the PCI address space straight into normal system memory space. This means that various sorts of memory and pseudo-memory resources on PCI cards has to show up somewhere below 4GB.

(RAM gets remapped above 4GB instead of PCI cards because generally remapping RAM above 4GB is much, much simpler than remapping PCI cards.)

Date: 2006-10-18 01:06 am (UTC)
From: [identity profile] pennyhill.livejournal.com
It's a good thing that my livelihood doesn't depend on understanding computers, because god damn i hate the fuckin' things.

It's a good thing that my livelihood doesn't depend on understanding what you wrote, because god damn at least you're cute.

xoxoxo.

Date: 2006-10-18 01:18 am (UTC)
From: [identity profile] tskirvin.livejournal.com
At least you got access to that memory. The Sun Wx100z workstations have the same memory hole, with no way around it; the rack-mount equivalent released a firmware patch to fix the problem about two months ago, about a year after I first encounted one of the systems.

Date: 2006-10-18 04:08 am (UTC)
ext_8707: Taken in front of Carnegie Hall (grumpy)
From: [identity profile] ronebofh.livejournal.com
Ugh. Thanks for the tip.

Date: 2006-10-18 05:22 am (UTC)
ext_243: (fanny)
From: [identity profile] xlerb.livejournal.com
But look on the bright side: at least you have the vast quantities of RAM that make this a problem in the first place.

I've been living with this in virtual storage

Date: 2006-10-18 08:08 pm (UTC)
From: [identity profile] vardissakheli.livejournal.com
since before I ever had a PC. Way back before VM/XA, the CMS nucleus had to sit below the 16 MB line. Larger virtual machines had a hole in them from 12 to 16 MB. I still occasionally deal with a few old programs that need to use shared segments below the line. Actually, I had thought this was finally gone, but I still seem to have a hole from 9 to 13 MB where our SYSPROF loads four dummy segments that used to be needed to keep the space available for those programs (primarily OfficeVision, now long gone) to use.

Profile

rone: (Default)
entombed in the shrine of zeroes and ones

December 2022

S M T W T F S
    123
45678910
11121314151617
18192021222324
252627282930 31

Most Popular Tags

Style Credit

Expand Cut Tags

No cut tags
Page generated Jul. 12th, 2025 09:05 am
Powered by Dreamwidth Studios