SUMMARY: KT133A boards make FSB overclocking a real option; Iwill’s KK266 outperforms ABIT’s KT7A by a significant margin at reasonable SDRAM voltages.
PC Nut was kind enough to send me a KT7A with RAID to test out, complete with a Global Win FOP38, 128 MB Mushkin High Density SDRAM and a One GHz T-Bird. Nice kit! I earlier reviewed Iwill’s KK266 and thought a head-to-head test would be interesting.
Since the 133A chipset is plug compatible to the 133, many of the boards that feature the 133A are basically the same 133 boards with a 133A chipset. ABIT’s 133A entry, the KT7A, is basically the KT7 with a new chipset. I eyeballed each board and found no major difference between the two (a couple of chips in the lower right hand corner were replaced with larger ones, but that was all I could see). So if you are looking for basic features etc, just read any KT7 review or ABIT’s website and you’ll get the idea. Makes a review easy but not too informative.
The major difference, of course, is the KT133A chipset and its ability to run higher FSBs – the KT133 was crippled and rarely did I see anyone running a KT133 much over 110 FSB reliably. With 133 MHz a spec speed, FSBs greater than 133 MHz become really interesting. So the focus of this article is not a review per se but a test, a test to evaluate what might you reliably expect from high FSB performance between the ABIT KT7A and the Iwill KK266.
To test FSB performance, I relied on a test suite marketed by UXD. More details on this test suite are given at the end of this article.
First I ran some SiSandra, Quake and 3DMark2000 benchmarks to compare the Abit and Iwill at 8.5×133. These were run with 128 MB RAM Mushkin High Density Rev. 3.0 SDRAM, CAS2, with a T-Bird One Gig and Leadtek GTS2:
|3DMark 2000 800x600x16
Something interesting here – there is a significant performance difference between the two with the Iwill clearly jumping ahead. I would expect to see some correlation between Memory Bandwidth scores and the Quake/3DMark 2000 scores, but the Iwill scores better than the ABIT in spite of slightly lower Memory Bandwidth scores.
Next, I used the RAM Stress Test to see how far I could get with FSBs. I started this series at 7×150 and worked up, varying both FSBs and SDRAM voltage; I had to use CAS3 at these speeds as CAS2 would not hack it (Memory Interleave at 4 way), with the following results:
|7 x 150
PASS @ 3.6v
PASS @ 3.46v
|7 x 155
PASS @ 3.7v
PASS @ 3.46v
|7 x 160
FAIL @ 3.9v
PASS @ 3.46v
|7 x 166
FAIL @ 3.9v
FAIL @ 3.46v
Iwill can hit higher FSBs than ABIT, with the usual caveat about single sample tests*. I ran full board stress tests at 133 MHz on both boards without a hiccup, so don’t read these results as indicative of performance problems at that speed. However, there is no doubt that the Iwill KK266 can deliver reliably higher FSBs.
From what I see of SDRAM specs, running more than 10% is out of spec (at least on the sticks I have), so SDRAM voltages in excess of 3.6 volts puts you into some questionable territory. Also note that at these speeds, the PCI bus is starting to run at some hefty speeds; typically, sensitive peripherals like NICs and some hard drives really flake out.
For all practical purposes, FSBs beyond 160 MHz are marketing hype. Running at any FSB which requires SDRAM voltage increases of more than 10% is (IMHO) risky; RAM is cheap, but not expendable. In addition, it is quite likely that SDRAM at speeds much in excess of 133 MHz must run at lower settings, which slows down performance a bit.
SDRAM that is spec’d for 166 MHz has to run at 6ns; 200 MHz spec SDRAM has to run at 5ns. Right now we are seeing CAS2 PC133 @ 7ns from Crucial. I would guess that widely available PC166 may be a while coming.
As an example of CAS2 vs CAS3 differences:
Overall about a 2-3% performance difference; for almost all applications, not really noticeable. Even though CAS3 does exact a performance hit, it is outweighed by the higher memory bandwidth gains. For those of you curious about what to expect at 160 MHZ:
7 x 160 = 1120 MHz
|3DMark 2000 800x600x16
There ARE real performance advantages at higher FSBs.
The Iwill KK266 clearly outperforms the ABIT KT7A. Considering these tests, I think the Iwill KK266 is setting the KT133A performance standard.
We will be testing it against the ASUS A7VA and see how it stacks up. While the KT7A has more BIOS tweaks, for those of you who like more memory tweaks than Iwill has in its BIOS, use ODA’s WPCREDIT. ABIT’s KT7A is an acceptable board up to about 150 FSB, depending on RAM quality and ABIT’s erratic Quality Control. I would not look to the KT7A for much reliably/safely beyond that.
Many thanks again to PC Nut for lending us the KT7A to test out.
*Most reviews I have seen to date have similar experiences with the ABIT board, although one stated “I wasn’t able to achieve any long term stability at FSB speeds greater that 133FSB.”
I used a diagnostic board called “PHD PCI” that stress tests all components on a motherboard; you can see more about it HERE. The company that makes it, Ultra-X, markets a series of hardware and software products for diagnosing PCs, and I also used Quick Tech Pro, a software test suite. What’s neat is I can run it in “Burn-in” mode which repeatedly tests each component on the board. Briefly, I tested the following:
- System Board – Continuous loop testing verifies proper operation of CPU, FPU, DMA channels, interrupts, timers, BIOS, keyboard, and other I/O functions.
- Bus / CPU Benchmarking including signals for: CLK, OSC, BALE, RAM Refresh rate.
- RAM – Tests Critical RAM without using any of the motherboard installed memory. True low memory testing from address 0 to 1024k. Also tests Gate A-20, memory address / data lines, and refresh circuit.
- Base RAM
- Extended RAM – Twelve algorithm patterns and graphic mapping identify the exact failing module or chip.
- Cache RAM – Multiple pattern algorithms are performed for accurate and reliable testing.
- PCI Bus
- I/O Port Monitor – Monitors all I/O ports from 0 to 3FFH.
- LPT and COM Ports – Tests all serial and parallel ports.
- RTC/CMOS RAM
The test suite covers much more, but for this test I concentrated on motherboard components and how it handles RAM at high bus speeds. For the RAM test, I used a product called RAM Stress Test. When you run it, it puts the RAM through a series of test algorithms including
“Stuck High and Low, Snake On and Off, Pseudo Random, Parity (Standard and Inverse), Walk Left (Standard and Inverse), Walk Right (Standard and Inverse), Checker Board (Standard and Inverse), ATS, MATS+, Marching C, Block Move, Leak (the most thorough RAM test available). It also tests the Data Bus, Address lines and Refresh Circuit.”
Now I can’t tell you what all that exactly does, but I can tell you that the RAM chips get fairly warm while all this is going on. I can also tell you that if the RAM fails this test, getting into Windows and running any intensive app is a problem. Where this really came in handy is for testing both boards at high FSBs to see what they can reliably handle.