The Abit KT7-RAID motherboard has been out for a couple months, and it has been reviewed extensively on the web, but I thought it would be worth taking a quick look at nonetheless.  And with all the mystery surrounding when the new DDR-capable AMD systems will actually be available, it's nice to know that very fast, very affordable systems can be had right now.

Introduction: Abit is known for making motherboards with excellent overclocking features, and the KT7-RAID is no exception.  As you can see in the picture above, the Abit KT7-RAID is a socket A motherboard with six PCI slots, one ISA slot, one 4X AGP slot, and three SDRAM sockets. It also has two ATA/66 mode 4 IDE controllers, and 2 ATA/100 RAID controllers on board.  I am very happy with Abit's decision to go for the 6PCI/1ISA configuration, since it is the best option for most people.

Click Here For the KT7 RAID Spec Sheet 

Setup: The test system included an Abit KT7-RAID motherboard, 128MB of PC-133 SDRAM, a Hercules 3D Prophet GeForce-1 DDR card, 1 Fugitsu 4GB ATA/66 mode-4 IDE drive, and 2 IBM 75GXP 20GB Deskstar ATA/100 mode-5 drives.  The ATA/66 drive was the boot device, and had Win 98SE+ installed with DX 7.0a and NVidia's Detonator-3 6.31 drivers. For most testing, I used the ultra-affordable AMD Duron 600 processor.

Installing the KT7 was fairly straightforward.  The one thing that needs improving on all socket-A motherboards is the mechanism for clamping the heat sink on the processor. The KT7 is no exception, with critical components located dangerously close to the catches for the clamp (see picture below).  However, the primary blame goes to AMD for their heat sink clamping design.  It needs to be reworked to make it easier to latch and unlatch the clamp without endangering components on the motherboard.

Before you install the CPU in the socket, you will need to bend the thermal sensor for the CPU down slightly.  It sticks straight up out of the center of the CPU socket (the blue dot in the picture below), preventing the CPU from seating properly. However, this was intentional on Abit's part.  You need to bend the sensor down just enough so that the CPU goes into the socket, and then the spring action of the heavy leads that connect the thermal sensor to the motherboard will hold the sensor tight against the bottom of the CPU.  This ensures an accurate temperature reading from the sensor.

The KT7 has a fan and heat sink on the Northbridge of the VIA chipset, which should help keep things cool if you decide to overclock the front side bus.

However, now that multiplier overclocking is possible with socket A processor's from AMD, bus overclocking is not necessary. Bus overclocking with Intel processors can offer substantial speed boosts. However, bus overclocking with AMD processors is often less successful due to the stringent timing requirements of the EV6 bus.

As of some time in October, AMD has been locking the multiplier on their new Athlon and Duron processors in the socket-A format. This is accomplished at the factory by using a laser to cut several contacts on the surface of these processors. By reconnecting these contact lines, you make it possible to change the CPU multiplier in Abit's Softmenu III BIOS setup menu.  If you want to know more about T-Bird and Duron overclocking, and how to do it, click Here.

Overclocking the Duron 600: The Abit KT7-RAID is the premier motherboard for overclocking Duron and Athlon processors in the socket-A format. The default settings for the Duron 600 is 6 x 100MHz with a core voltage of 1.5v, and an I/O voltage of 3.3v. The results of overclocking attempts are shown below.

Boot success:

7 x 100MHz      1.5/3.3v = OK
7.5 x 100MHz   1.6/3.3v = No
7.5 x 100MHz   1.675/3.4v = OK
8 x 100MHz      1.7/3.4v = OK
8.5 x 100MHz   1.7/3.4v = OK
9 x 100MHz      1.7/3.4v = No

By boosting the core voltage by 0.2 volts and the I/O voltage by 0.1 volts, I was able to achieve a speed increase of 250MHz on a $65 processor. That's about a 40% increase. At this speed, the processor temperature after a round of benchmarking was 42 degrees Centigrade (107 Fahrenheit). 

Stability:

The KT7 motherboard was exceptionally stable using multiplier overclocking. Over the course of 4 days of testing, the system hung only once at 850MHz. This is mainly due to the fact that Abit includes excellent control over parameters that help stabilize overclocked processors. The ability to adjust the CPU core voltage, and I/O voltage independently, and to control the multiplier directly from Softmenu III in the BIOS makes the KT7 an overclocker's dream come true.

Benchmarks:  For benchmarking I used common NVidia graphics cards, including the 32MB Asus V7100 GeForce2 MX card and the Hercules 3D Prophet GeForce-1 32MB DDR card. I tested Direct 3-D performance using 3-D Mark 2000 (ver. 1.1), and NVidia's detonator-3 driver's (ver. 6.31). The chart below shows Direct 3-D performance with the Duron 600 processor overclocked to 850MHz. In addition, the GeForce-1 DDR card was overclocked to 125MHz on the core, and 333 MHz on the DDR memory. The GeForce MX card was overclocked to 210MHz on the SDRAM.

Direct 3-D Performance with the two video cards at their overclocked settings was very similar. The DDR GeForce card pulled ahead at higher resolutions and color depths. Overall, Direct 3-D performance was excellent on the overclocked Duron system with KT7 motherboard.

OpenGL performance was tested using the 32MB Asus V7100 GeForce2 MX card and Quake III Arena version 1.17. Settings included maximum geometric and texture detail, lightmap, trilinear filtering, and V-sync off.

As you can see, even with a low-cost GeForce MX card, OpenGL performance on the overclocked Duron system is exceptional. The system was able to deliver over 60 frames per second at 1024 x 768 in 32-bit color.

ATA/100 RAID IDE: Setting up a RAID array on the KT7 is quite easy.  Click Here for more details. 

The chart below compares the performance of several different hard drive types using Scisoft's Sandra hard drive benchmark.  This benchmark seemed to give the most realistic view of hard drive performance from among the various programs I tested.

The graph shows that ATA/66 drives perform better than Ultra Wide SCSI drives, and that ATA/100 drives perform slightly better than ATA/66 drives.  Setting up a RAID array with the ATA/100 drives provided about an 18 percent performance boost. A lone Ultra-160 SCSI drive (IBM 10K rpm) offered approximately 26% better performance than the IDE RAID solution.  But considering the large price differential, most people will still opt for ATA/100 RAID over U160 SCSI.

U160 SCSI vs. ATA/100 RAID IDE: Many web sites have reviewed ATA/100 controllers and drives, and I have heard them claim that there is no need for SCSI anymore.  While ATA/100 drives have improved immensely, and often have better "logical" (cached) read performance than U160 SCSI drives, the physical performance is not nearly as good in "real world" comparisons.

The chart below shows the difference in the time to copy a 54MB file between different drives on the KT7-RAID, vs. U160 SCSI drives running off the Adaptec 29160N SCSI controller. Keep in mind that the comparison is between a striped ATA/100 RAID array, and single Ultra-160 SCSI drives.  The SCSI drives would of course have delivered even better performance if they were set up in a SCSI RAID array.

The letters in the chart indicate the source and destinations drives (C-D means copying 54MB from C to D). The C drive used in conjunction with the RAID array was an ATA/66 drive on the standard Master IDE controller of the KT7.  So C to D copying is from the ATA/66 to the RAID array, while D to D is from one directory on the RAID array to another.

The RAID C-C number in the chart above is for copying a file across directories on the ATA/66 drive.  The RAID D-D number is for copying across directories on the RAID array.  It took twice as long to copy the 54MB file on the RAID array as it did on the lone ATA/66 drive.  That's not a good result. And when comparing the IDE RAID performance to U160 SCSI, it's obvious that U160 SCSI will get the job done in much less time.

The RAID performance on the KT7 could be improved with optimized HighPoint drivers. In addition, other benchmark tests, such as the drive benchmark in Norton Utilities 2001, indicated that the ATA/100 RAID array had better cached read and write performance than the U160 SCSI drives. So depending on what the different benchmarks are stressing, the ATA RAID array can offer very competitive performance under the right conditions.

Summary:  As far as I'm concerned, this is one of the best motherboards Abit has ever manufactured.  Softmenu III is fantastic, giving you the ability to change the multiplier of AMD processors after a small, simple modification to the CPU. This is the coolest thing to happen to overclocking in years. Intel can't touch the overclockability of inexpensive, widely available AMD chips.

The RAID support on the KT7 is robust, and well implemented.  RAID is a major step forward for IDE drives, and it comes with the KT7 RAID at no extra cost. The only thing that disappointed me was the performance of RAID striping with the HighPoint ATA/100 controller.  In some benchmarks it's performance was admirable, but in real-world disk to disk or directory to directory copy tests, the performance did not even equal that of lone ATA66 drives running on the standard onboard IDE controllers. New drivers may improve future performance.

In conclusion, if you want a T-bird/Duron socket-A motherboard, this is the one to get.

© Copyright, November 3rd, 2000