LOSTCIRCUITS |
|
DIMM'S and DIMM's
FAQ's and such
|
|
(Review by MS)
|
(Special Thanks to Lensman1 for reading and commenting on this section)
| Enter some of your own DIMM experiences
| DIMM
Performance Comparison Table |
General section:
What kinds of memory are utilized by my computer?
Basically all modern computers use more than
one type of memory. The still most common type of layout which is utilizing
the Intel Pentium Classic, the Pentium Pro, the Pentium MMX, the
AMD K5, K6 and K6-2 (formerly known or rather anticipated as K6 3D), the
IBM/Cyrix processor family, as well as the IDT and the upcoming Rice CPUs
is characterized by the socket that holds the CPU on the mainboard. This
socket is called "Socket 7" to distinguish it from older types commonly
found on 387 and 486 mainboards.
Socket 7 based systems use 3 types of memory.
The first memory type which is the fastest is integrated into the CPU itself,
in order to keep transfer time between the processing unit and the memory
at a minimum. This type of memory is capable to transfer data at clock
speed, that is, in a 200 MHz CPU, it can transfer 200 millions of
data bits per second, in a 300 MHz system, it runs at 300 MHz. This type
of memory is called the level 1 (L1) cache.
The second type of memory which serves as a buffer
between the main memory and the L1 cache is called the L2 or "level
2" cache. This memory is located on the mainboard, usually in multiples
of 256 kB (per chip) and runs at bus speed, which is the frequency at which
the mainboard operates. Typically, this is 66 MHz, however, in more recent
mainboards, the possible bus speed has been increased to 75, 83, 92, 100, 112 or 124 MHz. the fastest bus speed is found on some ALi Aladdin boards that offer up to 140 MHz options. In case the AMD K6-III is used that has an on chip L2 cache, the nomenclature changes in that the L2 will now be the L3 cache
The third type of memory is the so called main
memory or RAM (random access memory). RAM comes either in the form
of "single inline memory modules" (SIMM) or as "dual inline memory modules"
(DIMM). SIMMs have a restricted data transfer bandwidth (32 bits) which
is less than the data width required by Pentium and above processors because
those require a data width of 64 bits. This is the reason why, in Pentium
(586) and higher systems, SIMMs always have to be installed in pairs, otherwise
the additional memory will be ignored or the system will not boot at all.
This is also the reason why, every once in awhile, some clever vendors
offer SIMMs at outrageous prices with a mail-in-rebate (limit one per customer)
at equal the price of the SIMM. One SIMM is as useless as a hole in the
head unless you buy a second one, the price of which often exceeds current
market prices for two SIMMs at other places.
DIMMs, on the other hand have a data bus width
of 64 bits which is the same width that is required by all current CPUs
and , therefore, can be installed as single modules.
Can you mix DIMMs and SIMMs?
It is not recommended for various reasons, e.g.
difference in timing (when one combines SDRAM with EDO) and differences
in voltage (DIMMs run at 3.3 V whereas SIMMs require 5V). However, in some
cases, the results are positive. One caveat of course is that the slowest
memory is the one that, in the end, will determine the overall performance
of the memory, on the other hand, an increase of the total amount of RAM
may more than compensate for a "slower" transfer rate.
How much RAM can my motherboard cache?
That depends on the chipset and on the amount
of L2 (L3 with Slot1 boards) cache.
A typical example are the current Intel TX and
VX chipsets which use 512 kB of L2 cache and cannot cache more than 64
MB of RAM. Other examples are the VIA VP2 chipset which, with 1 MB L2 cache
can cache 128 MB in write back and 256 MB in write through mode. However,
this does not depend so much on the size of the L2 cache than on the 10
bit tag ram (see below) used. A similar situation is found in the VIA MVP3
chipset (one notable exception being the upcoming 2 MB version in the
Photon 100 HC which supposedly can cache 768 MB of RAM). In contrast, the
ALiAladdin chipsets only utilize 512 kB of L2 to cache 128 MB RAM, again
that is mainly caused by the tag RAM used in the L2 cache (in some cases
like the current versions of the ASUS P5A , the mainboard manufacturer
has disabled the tag RAM of the chipset and added the necessary equivalent
(in better quality and also because the original tag RAM is faulty in the
E revision of this particular chipset) as a separate unit to the mainboard.
How does exceeding the cacheable amount of RAM affect my system's performance?
That depends on the operating system. UNIX or
LINUX access memory from bottom to top (that means starting with the first
bytes / memory addresses available) and therefore, a performance hit will
occur only if the amount of used memory exceeds the amount of cacheable
memory. Windows 95/98 as well as the older 3.xx versions use the RAM
the other way around. That is, they start at the highest memory addresses
and therefore, if those are beyond the cacheable limit, the system will
suffer a noticeable performance hit. The reason is that, no matter, how
much memory is occupied, a portion will always be in the uncached area
and result in slower memory access of the system also commonly called "performance
hit".
What is the difference between L1 and L2 Cache?
The L1 or level 1 cache is the memory integrated
into the CPU. Different CPUs have different amounts of L1 cache but the
one thing in common is that the L1 cache runs at CPU clock speed. Level
2 Cache is, in the case of Socket 7 motherboards, integrated into the the
motherboard and is usually 256, 512 or 1024 kb. The main function is to
cache the main memory (DIMMs or SIMMs), that is, to keep the addresses
of the individual files within the RAM that are loaded while the computer
is operating. The L2 Cache runs at bus speed, so does the main memory.
Many make the assumption that the amount
of L2 cache determines the amount of cached RAM on the motherboard and
this is not exactly true. This fallacy was probably caused by the popular
practice of adding cache on a stick (COAST) modules to the motherboard to increase the amount of L2 cache and this allowed a larger amount of system RAM to be cached
as well. This increase in system supported memory is often attibuted to
the increase in L2, but, in fact, the deciding factor is the addition of
tag RAM also contained on the COAST modules.
In other words, L2 cache is actually made up
of two components, i.e., the cache itself and the tag RAM. The L2 cache
is used to store the most recently accessed instructions or data to make
this info available quickly to the CPU. A mobo with 256KB or 512KB L2 cache
is refering to the size of this data storage area. The tag RAM is slightly
faster than the cache because it is accessed first to determine if the
info that the memory controller is looking for is stored in the cache.
The tag RAM is used to store the main memory address of the data stored
in cache and thus the size of the tag RAM, ie its width in bits, determines
how many addresses can be stored in the tag RAM. The size of the tag RAM
is the determining factor in how much system RAM can be cached on the motherboard
and of course it is also necessary for the mobo chipset to support it as
well.
What is WriteThrough (3T Write sustained) and WriteBack mode ?
With WriteThrough, a memory address data change
is written to the address in L2 and to the main system memory which
requires more time.
With WB the update is made primarily to
L2 and is only written back to system memory when the data cache line in
L2 is needed by a more recent memory request. By avoiding the slower system
memory (main RAM) update until is actually needed, the system will
gain some performance. On the other hand, in WB mode the L2 cache can only
handle half of the amount of RAM that the WT mode is capable of.
What is L3 Cache?
In the case of PII processors, the situation
is somewhat different, since the CPU itself contains an L2 cache which
runs at half of the CPU clockspeed, that is, it is substantially faster
than the L2 cache of Socket 7 boards. Depending on the board and chipset,
there may be a separate on-board cache which functions like the L2 cache
of Socket 7 boards but is termed L3 cache. Of course, if the L2-cacheless
Celeron is substituted for the PII, the nomenclature will be changing the
L3 cache again into L2.
A similar situation is found when an AMD K6-III is used. This CPU contains both L1 and L2
"onchip" and the difference to the PII will be that the L2 as well as the
L1 will both run at clockspeed. This is expected to boost the performance
of the K6-3 way above the PII but since both levels of cache are already
implemented into the CPU, the L2 cache of the mainboard will become L3.
How much memory can be accessed in W95?
Basically as much memory as is physically present. However there are some system settings that influence the configuration of the memory as it is used in W9x. These configurations can be selected in the control panel in the performance sub menu and are termed "Mobile or Docking System, Desktop, or Network Server" profile.
Each disk performance profile adjusts the values of the following
file-system settings in the Registry:
PathCache, which specifies the size of the cache that VFAT can use
to save the locations of the most recently accessed directory paths.
This cache improves performance by reducing the number of times the
file system must seek paths by seraching the file allocation table.
The number of paths is 32 for the Desktop computer profile, 16 for
Mobile Or Docking System, and 64 for Network Server.
NameCache, which stores the locations of the most recently accessed
filenames. The combined use of PathCache and NameCache means that
VFAT never searches the disk for the location of cached filenames.
Both PathCache and NameCache use memory out of the general system
heap. The number of filenames is about 677 names (8KB) for the Desktop
computer profile, 337 names (4KB) for Mobile Or Docking System, and
2729 names (16KB) for Network Server.
BufferIdleTimeout, BufferAgeTimeout, and VolumeIdleTimeout, which
control the time between when changes are placed in the buffer to
when they are written to the hard disk.
The values to be assinged to each disk performance profile are stored
in the following Registry key:
Hkey_Local_Machine\Software\Microsoft\Windows\CurrentVersion\FS Templates
The following subkey contains the actual settings for the profile
currently used:
Hkey_Local_Machine\System\CurrentControlSet\Control\FileSystem
An additional performance setting in the FileSystem subkey,
ContigFileAllocSize, can be used to change the size of the contiguous
space that VFAT searches for when allocating disk space. Under MS-DOS,
the file system began allocating the first available space found on
the disk, which ensured a great deal of disk fragmentation and
related performance problems. By default under Windows 95, VFAT
tries to allocate space in the first contiguous 0.5MB of free space,
then returns to the MS-DOS method if it can't find at least this much
contiguous free space. This optimizes performance for both swap file
and multimedia applications.
In some cases, you might choose to set a smaller value in the Registry,
such as if you are not running demanding applications on the computer.
A smaller value for ContigFileAllocSize, however, can lead to more
fragmentation on the disk and, consequently, more disk access for the
swap file or applications that require larger amounts of disk space.
Special thanks to Russ here for digging out all the information.
next page: => Continue with the Specific DIMM FAQs =>
If you enjoyed reading this article and found it useful, please consider making a small donation to LostCircuits.
Thank you!
General disclaimer: This page only reflects the author's personal
opinion and assumes no responsibility whatsoever regarding any of the contents
or any damages that may occur explicitly or implicitly from reading the
contents of this site. All names and trademarks mentioned in this review are the exclusive property of the respective parent companies.
All contents of this site are protected by international copyright laws. Reproduction of the contents
even in parts is not allowed except after written permission by the author and referral to this site.
Copyright 1998 - 2007 LostCircuits
