A Hardware Dilemma: |  | by Yoshinori H. T. Himel |
You may have seen the ads in Sacra Blue for the "New Technology" 486 computer that's "faster than a Pentium." The Davis family includes his wife Karen and their son Nathan, 19.
The Davis family's products can be found in some SLUG (Sacramento Lawyer Computer Users' Group) members' offices. Former SLUG chair Bob Harris tells the story of buying the "new technology' computer and then finding a colleague's papers on his desk. The colleague, who has his own computer, so craved the "new technology" speed that, according to Harris, "he sneaked over to use mine!"
So, what's the "new technology"? And how can a 486 be faster than a Pentium?
The answer begins with memory caching. Today's desktop computers need lots of main memory, perhaps 4 to 32 megabytes. But buying that much memory fast enough to keep up with the CPU costs too much. So computers manage their memory using small blocks of fast memory and large blocks of slow memory. They use the faster memory to hold a copy of information from the slower memory that the CPU is likely to need (for example, information that it used recently). The faster memory is called a cache.
As Tracy Kidder's Soul of a New Machineexplained in the minicomputer context, "Maybe the Sys Cache has that next block of instructions. If so, it doesn't have to send away for them, but can pass them right on [to the CPU], and time is saved. If, however, the Sys Cache doesn't find those instructions in its own memory, it retrieves them from Main Memory and passes them on ..."
Main memory in PCs has ranged in speed from 200 (in the original IBM PC) down to about 15 nanoseconds, corresponding in principle to 5 to 66 million memory cycles per second. Costs dictate that most of today's PC main memory is rated at 60 or 70 nanoseconds. Also because of cost, main memory is dynamic RAM. Dynamic RAM costs less than static RAM, but "refreshing" (meaning electrically recharging the memory circuits that hold the data) takes dynamic RAM out of service for multi-nanosecond periods.
Caching PC memory became popular along with 80386 CPUs, because of the disparity between the CPU speeds (16 to 40 megahertz, corresponding to 25 to 60 nanoseconds per processor cycle) and the 80 to 100 dynamic memory in use at that time. The 386 computers relied on cache controller chips and static RAM cache memories 64K to 256K in size, mounted on the motherboard outside the CPU chip, to cache their 1 to 8M of main memory.
But 486 processors made memo caches and cache control logic a part of the CPU itself. These several nanosecond on-chip static RAM caches were limited by scarce chip physical area to 8K or 16K. Pentium CPUs similarly maintain small on-chip caches.
Computer makers, recognizing the wide speed gap between the very fast caches built into the 486 CPUs and the 60 or 70 nanosecond main memories, continued to build caches on their 486 motherboards. So now there are two levels of memory caching: first the 8K on-chip cache, and second a 256K, 15 to 20 ns static cache on the motherboard. This second level of memory caching had been almost universal in PCs produced before 1995.
So, what's "2L"? A sequel to Scott Turow's account of the first Harvard Law year, "1L"? And was "2L, or no 2L" Mr. Turow's soliloquy on whether to write the sequel?
This 2L stands for second-level cache. "Not 2L" is Davis' "new technology": the use of very fast dynamic RAM, 20 ns or below, as main memory instead of cache memory. While with a slow main memory a 2L cache speeds the system, a main memory running as fast as a 2L cache speeds it even more.
Thus, with an Intel 486DX4-100 and fast main memory on a suitable non-2L motherboard, Davis records benchmark speeds slightly greater than a Pentium 90 with the conventional 60 or 70 ns main memory and a 2L cache. This is despite the 486's inherent handicaps of a less efficient internal architecture and a memory bus speed of 33 MHz versus the Pentium's 60 MHz. Of course, accepting benchmark comparisons may need a leap of faith. SPCUG's Rick Hellewell says, "There are lies, damned lies, and benchmarks." Faith assumed, Davis substantiates his claim of a 486 that's "faster then a Pentium."
You may have heard the term "EDO RAM." EDO (no allusion to Japan's feudal capital of Edo) stands for "extended data output." Davis' fast main memory suffers one potential problem: when conventional dynamic RAM is that fast, it stops signaling its contents sooner than the CPU can read it. EDO keeps signaling longer; the technique makes dynamic RAM in the 15 to 20 ns range feasible for main memory.
Since fast EDO RAM can make a 486 faster than a Pentium, the next question is whether such a system is economically competitive. Davis says "yes!" He quotes prices a bit over $400 (maybe under $400 soon) for 8M of EDO RAM at 20 nanoseconds, a speed that would appear to support zero wait states at the 486DX4-100's 33 Mhz memory bus speed. This compares with street prices of $250-300 for 8M of conventional dynamic RAM. But DX4 systems cost a few hundred dollars less than Pentium 90s, saving more than the extra memory cost. (Doing the same comparison for 16M of main memory would at least narrow the cost advantage.)
A story in the July 10 Electronic News gives another perspective on the fast main memory technique. Author Anthony Cataldo quotes a major memory supplier as saying that the static RAM used in 2L caches is in very short supply until the first half of 1996, and attributes the interest in non-2L systems to this temporary shortage.
Davis told the SLUG attendees that he eagerly awaits the next fast EDO RAM motherboard, which will take Pentium CPUs. Slow main memory must hamper Pentiums, with their faster processing and memory access speeds, even more than 486s, so a non-2L, Pentium system should show a dramatic speed advantage over a conventional system with the same Pentium CPU.
But there may be a ceiling on how fast a Pentium CPU this technology can work with. Cataldo's piece quotes a cache memory supplier (perhaps a biased source) that "these cache-less systems are limited to PCs with 75MHz Pentiums on board, which match the 5OMHz bus speeds of EDO DRAMs and allow zero wait state bursts. There is less interest in circumventing cache for systems using MPUs operating at 9OMHz and above because EDO cannot handle the 60- and 66MHz bus speeds."
Davis disputes that conclusion by pointing to newer 8 to 10 ns, EDO RAM. These speeds are consistent in principle with memory bus speeds approaching 100 Mhz. That's faster than existing Pentium memory bus speeds, which top out around 166 MHz. Thus, as Pentiums continue to take over the CPU market, "2L, or not 2L" will still be a question.
Hardware does nothing without software; and today's processor-intensive, storage-intensive, and graphics-intensive games give the hardware its most severe test. So Nathan Davis demonstrated his hardware's prowess using Wing Commander III: Heart of the Tiger, the newest incarnation of such a game.
Brian Taugher won a copy in a drawing, and exclaimed: "My kids will love it!" (Brian - you mean you don't?) Don Heckman won another game donated by the Davises: X_ Wing Collectors' Edition.
Thanks to Richard G. Hellewell, one of SPCUG's mentors and a some time presenter at SLUG meetings, for commenting on this piece. Of COW, all errors are solely the author's.
Windows 95
Kenneth J. Hopkins' card for his business, Hopkins Computing, is based on a GUI metaphor: a screen shot of a Windows display. To the capacity crowd at the Sacramento Lawyer Users' Group meeting on November 16, 1995, he spoke of the newest Windows from Bill Gates: Windows 95. Hopkins, a Win 95 beta tester for two years, asked how many people had Win95. Of the 40 in attendance, 3 had it installed already.
Win95 is the newest 32-bit PC operating system on the market, replacing operating systems (like DOS) that could run 16-bit processors such as the original IBM PC's 8088 and the IBM AT's 80286. Existing 32-bit operating systems include IBM's OS/2 (including Warp) and Microsoft's Windows NT.
Win95 is a GUI, or graphical user interface, designed to let the user "point and click" on icons instead of typing commands. It joins other GUIs including Microsoft's Windows, Windows for Workgroups, and Windows NT; IBM's OS/2 and Win-OS/2 (OS/2's Windows emulator); the Macintosh operating system; and X-windows (running on UNIX systems).
Hopkins said Microsoft has a short-term and a long-term strategy involving Win95. To be successful immediately, he said, Microsoft intends Win95 to be compatible with any program written to run under Windows 3.1 and most DOS programs.
Bill Gates' new book, The Road Ahead, confirms the Microsoft leader's view that compatibility is a key computer industry lesson:
"Before the [IBM] 360, computer designs were intentionally incompatible with those from other companies because the manufacturer's goal was to make it discouragingly difficult and expensive for customers heavily invested in one company's computer to switch to a different brand. Once a customer committed to a machine, he or she was stuck with offerings from the computer's manufacturer because changing the software could be done but was too difficult. Amdahl and the others ended that. Market-driven compatibility is an important lesson for the future personal-computer industry. Customers choose systems that give them a choice of hardware suppliers and the widest variety of software applications."
(Ironically, the CD-ROM version of the same Gates book appears incompatible with many operating systems. The Road Ahead's CD-ROM specifies Win3.1 or Win 95, not OS/2, Macintosh, or even Microsoft's own Windows NT. Although the CD-ROM book installed under Win OS/2 on your writer's machine, it wouldn't run.)
Because of the many steps that Win95's designers took for compatibility with existing Windows and DOS programs, Win95 has a reputation of a melange. Gordon Nelder-Adams, writing in Sacra Blue, says:
"One of Microsoft's most important design goals with Windows 95 was to maintain compatibility with existing DOS and Windows 3.x programs and device drivers, at all cost. That cost has turned out to be outrageous delays in shipping, and a new operating system which is inherently less stable than NT or OS/2. The payoff, however, is an operating system which can run both older 16-bit applications and new 32-bit applications."
Which 16-bit software won't run in Win 95? Hopkins listed virus scanners, disk defraggers, and shells. SLUG member Jim Mize reported that he had trouble installing and running the following programs under Win95: PCAnywhere, Procomm Plus, and Matthew Bender's Search Master for DOS. Mize also reported numerous crashes under Win95, especially in DOS games. Hopkins said that the technically inclined user can adjust many DOS parameters in Win95. But in the long run, he said, Microsoft hopes that Win95 or its successors will eliminate all 16-bit applications from the market.
What will Win95's successors be? Viewing Win95 as temporary, Hopkins calls it an "intermediate operating system to get you to Windows NT. " Gates views almost everything in the PC world as temporary, and claims only that Windows "has the lead for the moment."
What's Windows NT? Oddly, Gates' book doesn't mention NT. Nelder-Adams says that "Windows NT...was originally planned to be OS/2 3.0 (not to be confused with Warp, which is also really OS/2 3.0)."
Patty Reeves asked Hopkins why a user should upgrade from Win3.x to Win95. Of course, Hopkins said, Win95 is more stable, prettier, and faster. But he said the real reason is that we won't have a choice. Many new applications are being written for 32-bit systems only. Hopkins said that at this fall's Comdex computer industry exposition, every program on display claimed Win95 compatibility, even outside the Microsoft Partners tent.
But Nelder-Adams says there may be a choice with the new 32-bit programs. Because Microsoft required that any software vendor seeking a "Windows 95 compatible" logo make the software run under NT as well, Nelder-Adams wrote, the programs with the Win95 logo can run not only on "the less-than bulletproof Windows 95" but also on "the rock-solid Windows NT."
Supposedly, it's quite easy for a software developer to adapt a Win95 program to OS/2 Warp, another operating system with less software written for it than what Nelder-Adams calls the "demonstrably less powerful or stable" Win95. Those using Warp, including your writer, are feeling some 32-bit osmosis from Win95.
References:
Cataldo, Anthony "EDO spots cache gap," 7/10/95 Electronic News (Vol. 41, No. 2073) page 1.
Computers by KD&C, David Davis, Manager, 7412 Hollyhock ourt, Citrus Heights 95621, 722-3325.
Glass, Brett,"Help Desk," InfoWorld, December 13, 1995.
Kidder, Tracy, Soul of a New Machine (New York: Avon Books, 1981) at 124 Wing Commander III : Heart of the Tiger, Origin, An Electronic Arts Company, P. O. Box 161750, Austin, TX 78716.
Gates, William H. III, The Road Ahead (New York: Penguin, 1995), pages 39, 63 Nelder-Adams, Gordon, "Microsoft's New Clothes," June 1995 Sacra Blue, pages 56, 58.
Nelder-Adams, Gordon, "A New Face," November 1995 Sacra Blue, pages 38, 40, 41.
Copyright 1995 by Yoshinori H. T Himel. Reprint from Sacre Blue, The Magazine of the Sacramento PC Users Group.
E-mail me at webmaster@hal-pc.org with any comments you have and tell me what you want to see here.
Back to the User Journal Home Page