Вы находитесь на странице: 1из 4

Back to Basics - Memory, part 3

Toronto Star Fast Forward column for May 21/98 Back to Computer Wares index Back to White Pages main article index Copyright, Myles White, 1998 Tip of the Week: Just say, "No" to screen savers.

Over the past two weeks, we've looked at where memory can be found in your
computer (just about everywhere), at the two basic types (RAM and ROM), and the basic considerations to keep in mind when you want to get more (make it the physically the same as what you have now). This week, we'll try to wrap up the memory primer, then take a week or two away from memory for a such things as a feature on shareware. Sometime early in April, we'll come back to some questions about memory that have been flooding into my mailbag. When you hear someone say, "I've got 32 megs of RAM in my PC," they've only told you part of the story about the main system memory the computer has. The next question is, what kind of RAM? There may be one of three main types: * DRAM -- dynamic random access memory comprises the main system memory on most older (486 and early Pentium) PCs. Also known as FPM (fast page mode) DRAM, it comes in 80, 70 or 60 nanosecond (ns) speeds (the lower the number, the faster the memory). Called 'dynamic' because it gets zapped with a small electrical current each time it changes state (i.e., to read from it or to write to it) and the charging time (latency) reduces its speed. FPM DRAM takes two processor wait-states (cycles where the CPU does nothing except wait) each time a read or write takes place. Advantages: it's compatible with just about everything. Disadvantates: it's the slowest of the memory types currently available and because it isn't in demand, it's now more expensive than faster memory alternatives such as EDO and SDRAM. DRAM may also be found on lower-priced video graphics controllers and some wave table audio controllers, but it will almost invariably come in a proprietary form that is controller-specific. You may also discover that some models of printer use DRAM -- either in proprietary form or on industry standard 30- or 72-pin Single Inline Memory Modules (SIMMs). * EDO RAM -- Extended data out random access memory (EDO RAM) and its cousin, Burst EDO RAM (BEDO RAM), are anywhere from 4 to 15 percent faster than FPM-DRAM. EDO RAM gets its name by extending the time during which data can be read from memory, because the available read time doesn't become invalid until an additional signal is sent to the chip. By contrast, conventional DRAM normally discharges its contents after each read and must be refreshed before another read can occur. EDO RAM takes one processor wait state, which led some publications to claim it was fifty percent faster than FPM-DRAM. It isn't. In order to use EDO RAM, your motherboard's chipset must explicitly support it.

The memory market fluctuates constantly, EDO RAM is now usually more expensive than FPM-DRAM, but less expensive than SDRAM, EDO RAM comes in 72-pin SIMMs and, occasionally on 168-pin DIMMs (dual inline memory modules). * SDRAM -- synchronous DRAM. Its timing is synchronized to the system clock. By running in sync to an external clock signal, SDRAM can run at the same speed as the CPU/memory bus. This results in a memory system without wait-states to slow every thing down At the same speed, SDRAM is about 10 percent faster than EDO RAM, but newer versions of it are significantly faster, running at 10 ns instead of the 60 ns common for EDO RAM. SDRAM now also comes in two distinct types: for 66 MHz motherboards and for 100 MHz boards using Intel's new 440BX chipset and faster (350 and 400 MHz) Pentium II processors. At any speed, your system's chipset must support it explicitly before you can use it. 100 MHz SDRAM is the most expensive of the memory types currently available, but 66 MHz SDRAM is often the least expensive of the three mainstream varieties. SDRAM comes in 72-pin SIMMs and 168-pin DIMMs.

Cache memory
* SRAM -- static RAM. Unlike dynamic RAM, SRAM is powered once and doesn't need to be continually refreshed. Typically SRAM is used in external (also known as L2) cache memory at speeds of 15ns or less. Pentium-class CPUs directly tied to 60ns or 70ns DRAM waste wait states. The traditional solution: place 15ns or 20ns static RAM (SRAM) as a buffer between the two. However, a Pentium running its memory bus at 66 MHz requires only 15.2ns for each cycle--somewhat of a close shave--which brings us to synchronous SRAM. Older, asynchronous SRAM performs reads and writes somewhat sequentially, with one operation completed before the other begins. Pipelined, synchronous SRAM lets a second data access begin before the first is completed, permitting burst transfer rates as high as 100MB per second. Synchronous burst SRAM can cut cycle time to roughly 10ns, an easy match for a 66MHz bus. Will faster cache memory be useful on a 100 MHz motherboard? You betcha!

Weird Memory
The following are uncommon memory types, but I've listed them here just in case they ever start showing up in consumer-level systems * CDRAM -- Cached RAM, invented by Mitsubishi Electronics, combines an SRAM (static RAM--see below) cache with 4 or 16 MB of DRAM within a single chip. This onboard SRAM can be used as both a cache or a buffer and gives the RAM an approximate 15ns access time. Because the RAM does not have to send its information to an external SRAM cache, overall system performance is improved considerably. * EDRAM -- Enhanced DRAM. Like CDRAM, it also incorporates an on-chip SRAM cache. Developed by Ramtron International Corp. (Colorado Springs, CO). By improving the DRAM's performance to 35ns, and combining it with a 2KB, 15ns SRAM cache, a DRAM chip with a 15ns access time is created. In contrast, CDRAM uses a much larger 15KB, 15ns cache and DRAM with a much slower 70ns access time. * RDRAM -- Rambus DRAM, after the company which developed it. Currently being toyed with by Toshiba and Samsung, it's similar to SDRAM (see above), but faster says Rambus. You're more likely to find it appearing on accelerated graphics cards and the Nintendo N64, but I wouldn't be surprised to see it show up on somebody's PC, eventually.

Video memory types

Aside from variations of FPM-DRAM and EDO RAM, there are three other memory types you may find on a graphics controller. * SGRAM -- synchronous graphics RAM, a form of DRAM for graphics controllers and printers that is synchronized to the system clock and, according to one of its manufacturers, Fugitsu, produces data bandwidth up to five times that of standard DRAM. * VRAM -- Video RAM, the oldest of the so-called 'dual port' memory types that allow the graphics processor to read from memory and redraw the screen simultaneously, eliminating the problem plaguing single-ported memory types (i.e., DRAM), where the memory can only be read from or written to at one time (and the graphics engine must wait each time the screen is updated). With single-port memory, the higher the refresh rate, the resolution or the colour depth used, the more times per second the RAMDAC (the device that converts the graphics card memory's digital contents to an analog signal) will have to read from the frame buffer, and the slower the system gets. * WRAM -- Window RAM, developed by Samsung Electronics and most likely to show up on a Matrox. graphics controller and others. According to Advanced Imaging (June 1996), "when considering the benefits of WRAM over VRAM memory, one must bear in mind that WRAM memory is both faster, offering a 50 percent performance increase, and less expensive, 20 percent lower cost per bit, than VRAM memory." When I started this series, I said I'd discuss a product that claimed it could counteract system lockups resulting from memory errors. Symantec's Norton Crash Guard Deluxe is supposed to do that. However, I've now been running it for nearly a month and I haven't noticed any decrease in unexplained freezes, nor has it's Anti-Freeze module been able to unlock the problems when they have occurred. I'm even convinced that I'm seeing more crashes than I did before installing it. Back to the drawing board.

Tip of the Week


Just say "No" to screen savers
When PCs were new, some version of character-based DOS was the only operating system around, and monitors were either monochrome or very primitive colour, screen savers were necessary. If you left the computer running with anything on screen for a long time, you ran the risk of burning the image into the tube - and its ghost would never go away. Screen savers kept the image moving around and burn-in never took place. That was then and this is now. Not only has monitor technology improved (new ones can be set to go to sleep after a short period of inactivity), but the way Windows is painted on screen makes burn-in a thing of the past. Screen savers are not required to prevent anything. That hasn't stopped them from appearing everywhere and turning into minientertainment or security devices. But they require memory; they slow the system down, and; in some cases, you can't get rid of them. If you constantly hit a memory barrier or something else goes on in your system that's a little odd, you might want to kill that screen saver as a good first step to solving the problem. Right-click anywhere on an open portion of the Windows 95 desktop, select the Screen Saver tab, then select "none," from the list box.

Copyright Notice

This document, and all other articles found within the White Pages Web site, are protected by international copyright. All rights are reserved. You may download items for your personal use. You may place a link to the White Pages at your own web site. However, you may not post the articles at your web site, make paper or digital copies for your friends or your class, or quote any part of these documents in any medium whatsoever for any reason unless you ask me first. Okay?