Stories By Fabian Pascal

32-Bit Computing

Performance Tweaks
for the Rest of Us

Part I: BIOS, Operating System and Processor Subsystem Optimization

The Communications Subsystem
DSL vs. Dialup

Best of Breed System -- 2nd Generation:
Processor, Graphics and Disk Subsystems

Intel "Compatibility" Under Windows NT

Hard to Keep Your Cool

Disk Subsystem

PC Advisor Column (www.dsi2000.com)

The Graphics Subsystem

The Processor Subsystem

Selecting a Slot 1 System board

On Chipsets and Buses

 

Also of Interest:

Transferring Files With Flipdisk

By Ed Kahn

Wireless PC

By Woody Liswood

Linux:
Today's Curiosity Or Tomorrow's Workhorse?

By Kevin Savetz and Neil Randall

Open Source Software: More Than An Emerging Alternative

By Michael Robin

Feedback To Author | Feedback To Webmaster

Copyright © 1999 by Fabian Pascal. All Rights Reserved

32Bit Computing

By Fabian Pascal

Performance Tweaks for the Rest of Us
Part I: BIOS, Operating System and Processor Subsystem Optimization

Going by what is being published by the media on hardware and performance, you would think that aside from corporate networks, the computer market has only one other segment: gaming. Vendors and online media, in particular, are obsessed with 3D performance -- a purely gaming issue -- to the point of absurdity. Most product development and reviews are focused on 3D performance and employ game tests, or synthetic benchmarks that only correlate with gaming loads. There is a continuous proliferation of web sites that go to ridiculous effort and expense to overclock processors and video cards just so Quake or Unreal run faster. Frankly, observing this mania, I don't know whether to laugh or to cry.

Aside from the perceived profitability of such endeavor -- gamers seem willing to spend their dough -- one reason for this fixation is due to the fact that 3D games consume considerable hardware resources, which the PC platform -- its ever increasing speeds notwithstanding -- has not yet been able to satisfy. The high-end components that could do the job are very expensive, so squeezing every MHz out of common CPUs or graphics cards -- whose price decline rapidly -- is the only alternative (although some of the techniques employed often are not much less demanding in terms of expense and effort). Another reason for the gaming bias is simply the fact that business applications are less graphics- (and, thus, memory- and CPU-) intensive and more dependent on disk access, which is considerably slower than CPU and memory. Consequently, business application performance is much less sensitive to video card and CPU speed improvements, particularly since disk performance has not progressed significantly. Vendors have little choice, therefore, but to target faster video cards and CPUs on 3D games.

This suggests that in the absence of non-incremental improvements in operating systems, applications and hardware technology, not much can be reasonably done by the moderately knowledgeable user to improve business performance, beyond matching best-of-breed system components, such as those recommended in this series (by reasonable I mean (a) non-technical (b) simple and fast (c) free or inexpensive (d) no disassembling or modifications of hardware (e) little or no damage risk.

Note: Even the most promising hardware improvements, such as AMD's forthcoming new K7 processor, are already previewed in a gaming context:

"slightly improved 3DNow! units and three indepedent FP units make the K7 floating point performance superior to the floating point performance of any x86 CPU in existance today. We estimate that the K7 will be, on average, 40% faster than the PIII in 3D games."

Various fine-tuning and streamlining techniques have been proposed that are claimed to speed up performance noticeably. The question is, do they have any effect on business applications? Since the components of the recommended system were already selected as best performers based on business application benchmarks (see previous articles), I decided to test the claims for some of the most accessible suggestions, using the very same Ziff-Davis business Winstone benchmarks: the 98 version for individual common applications and the 99 version for multitasking.

In this article I discuss BIOS, operating system and processor subsystem optimization. Subsequent articles will discuss the graphics and disk subsystems and CPU overclocking.

BIOS Settings

Most popular system boards come with the Award BIOS, which the manufacturers modify and expose to users in CMOS in different ways. Some of the settings tend to be poorly documented, if at all; for some of the more obscure settings, the effect of changing the default value on both performance and stability is unpredictable and can go either way. The only way to find out is to benchmark the possible values, but since combinations of values for multiple settings may make a difference, the effort would not be reasonable.

Figure 1 shows the CMOS settings for the Abit BH6 board that can possibly affect performance: the first column shows the default values. After experimenting with various value combinations, the effective changes proved to be those shown in the second column of Figure 1 for the highlighted settings.

=========================================
   Default Tweak                                                      
---------------------------------+-------+-------
 BIOS Features Setup
  CPU Level 1 Cache              + 
  CPU Level 2 Cache              +
  CPU L2 Cache ECC Checking      +      -
 Chipset Features Setup
  SDRAM CAS Latency Time         3      2
  SDRAM Precharge Control        -
  System BIOS Cacheable          -      +
  Video BIOS Cacheable           +
  Video RAM Cacheable            - 
  8bit I/O Recovery Time         1 
  16bit I/O Recovery Time        1
  Memory Hole At 15M-16M         -
  Passive Release                +
  Delayed Transaction            -
  Spread Spectrum Modulation     -      
=========================================
 * Total system memory

Figure 1: BIOS Settings

The benchmark results for the default values are shown in the first column in Figure 2 and the second column shows the results after the tweak. The largest gain is 10% for spreadsheet/database applications, followed by 7% for multitasking, with the remaining gains insignificant (3-6%).

========================================================
                       Default  BIOS     NT    PwrTwk   All
----------------------+-------+-------+-------+-------+-------
 Business Winstone 98   30.8     32.5   31.4    31.0    32.3  
  Netscape               3.73     3.95   3.82    3.71    3.82
  WordPerfect            3.18     3.29   3.19    3.16    3.28
  QuattroPro/Access      2.42     2.66   2.54    2.51    2.59 
  PhotoShop              3.7      3.81   3.71    3.63    3.82
----------------------+-------+-------+-------+--------+------
 Business Winstone 99   22.8     24.4   23.9    22.8    24.6   
==============================================================
 PII/300 @66 FSB
 1280x1024x32bit @75Hz refresh

Figure 2: Performance gains -- BIOS, operating system and chipset gains

Operating System Tweaks

Over a period of time I've collected various NT registry changes claimed to improve performance. The two most promising of those are shown in Figure 3.

==============================================================
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\FileSystem]
 NtfsDisableLastAccessUpdate = 1* 
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Session Manager\MemoryManagement]
 IoPageLockLimit = 16384000 (default=512)**
 SecondLevelDataCache = 512 (default=256)
==============================================================
 * Create new key value
 ** Function of available memory:
     32MB :  4096000
     64MB :  8192000
    128MB : 16384000
    256MB+: 65536000

Figure 3: NT Registry Settings

As the third column in Figure 2 shows, there is practically no effect on performance.

Powertweak

Powertweak (www.powertweak.com) is an optimizing utility described as follows:

"It configures your hardware components, in particular the processor and the motherboard chipset, thus allowing them to work at their highest capabilities. What is optimized in fact is the speed and consistency of the transfers between the different components. For instance, the accesses to the video memory can be optimized by gathering multiple transfers into one single, bigger transfer, so that the processor doesn't have to wait for the rest of the system to accept the data before going on. This is just one example of the many features of Powertweak. Each processor, each chipset has a particular set of options, which is sometimes similar, and sometimes completely different."

The benchmark results after running Powertweak are shown in the fourth column in Figure 3. Like the NT registry edits, the optimization has no effect on Winstone scores.

To test for the possibility of interaction between the three set of changes, the last column in Figure 3 shows the results after applying them all. They are the same as those for just the BIOS changes, leading to the conclusion that the registry changes and Powertweak do not have any effect on Winstone scores.

Conclusion

This article demonstrates that, as expected, suggested changes in BIOS and registry settings and processor subsystem optimization with Powertweak have little to no effect on business application performance (as measured by Winstone!).

As always, when it comes to benchmarking, caveat emptor: your mileage will vary with the details of your specific hardware and software configuration, the combination of applications you run and the ways in which you run them. Since they cost nothing, you should try these tweaks on your system, but it is reasonable to assume that a difference of 10-15% or less in benchmark results is not likely to be noticeable.

RECOMMENDED SLOT 1 SYSTEM

SYSTEM BOARD

BH6
Abit Computer (USA) Corp.
www.abit-usa.com

PROCESSOR SSUBSYSTEM

Pentium II/300/333Mhz (Deschutes core)
Intel Corp.
www.intel.com

VEK-12 CPU Cooler [1]
Acadia Technology Inc.
www.globalwin.com.tw

PC100 SDRAM CAS2
Mushkin Components
www.mushkin.com

GRAPHICS SUBSYSTEM

FlexScan FX-E8 21" Monitor
Eizo Corp.
www.eizo.com

Millenium G200 Graphics Adapter
Matrox Graphics Inc.
www.matrox.com

SCSI DISK SUBSYSTEM

AHA-2940U2W 64bit PCI Ultra2Wide Dual Channel Controller [2]
Adaptec Inc.
www.adaptec.com

Cheetah 9LP Ultra2Wide 9GB 10,000rpm Drive [3]
Seagate Technology Inc.
www.seagate.com

UltraPleX 32X Ultra CD-ROM Drive
Plextor Corp.
www.plextor.com

GlobalWIN King Kong Drive Cooler [4]
Acadia Technology Inc.
www.globalwin.com.tw

FLOPPY SUBSYSTEM

SuperDisk LS-120 IDE Floppy Drive
Imation Corp.
www.imation.com

COMMUNICATION SUBSYSTEM

FastTrak ADSL
Pacific Bell
www.pacbell.com

56K Dialup (backup)[5]
MessageSaver Data/Fax/Speakerphone/Answering Machine Modem
Multi-Tech Systems Inc.
www.multitech.com

INPUT SUBSYSTEM

Cordless Desktop Keyboard and Mouse Combo
Logitech Inc.
www.logitech.com

SYSTEM CASE

EN-8900 ATX Tower Case
Enlight Corp.
www.proqc.com.tw/~enlight/

ALTERNATIVES AND OPTIONS

[1] If not overclocking
VIVA P2 CPU Cooler
TennMax United
www.tennmax.com

[2a] Dual Channel
FirePort 40 UltraWide Controller
Diamond Multimedia
www.diamondmm.com

[2b] Single Channel
AHA-2940UW UltraWide Controller
Adaptec Inc.
www.adaptec.com

[3] 7,200rpm LVD
Viking II 9GB Ultra2Wide
Quantum Corp.
www.quantum.com

[4]
GlobalWIN I-Storm Drive Cooler
Acadia Technology Inc.
www.globalwin.com.tw

[5] For 56K dialup
HotCall 2000
Command Communications Inc.
www.command-comm.com