BIOS Settings:Megatrends FX83/PPro200/512 vs. Megatrends HX83/P200MMX

1024X768X16@85

This site has been accessed times since July 11th, 1997

JUNE 30th, 1997: These are optimal settings. They are not predicted by the AMI BIOS optimal settings nor are they predicted by the optimal settings published for the SKE/PII. Surprisingly, the PCI Frame Buffer USWC must be enabled for the FX83/Pro200/512 to run stable. Differences from these settings and the BIOS settings posted by Megatrends for the SKE/PII configuration indicate that the Pro200/512 has unique BIOS requirements. It took over 100 hours of testing to arrive at these optimal settings. My previous experience in tuning BIOS settings centered on graphics stability. With the FX83, the overriding goal was to achieve stability of the SCSI chain. Enabling any of the disabled settings shown in the above table results in instability. Again, enabling PCI Frame Buffer USWC is absolutely neccessary for stable operation. I am still in disbelief, but I have tested this over and over. Key WinMarks with these settings are: Bus Graph = 80.3, High Disk = 4950, Bus Disk = 1680, and CPU32 = 625. I expect Bus Graphics to approach or break 90 with the Millennium. BTW, these scores question the notion that a Pentium MMX beats a Pentium Pro when running OSR2 at the same clock speeds. More about this later in the week.

JULY 01, 1997: The verdict is still out on these settings. The graphics driver appears to interact with the AMI bios settings to further complicate things. The above settings were arrived at using the 128-2 VRAM with the 3.41 drivers. Now there is a Millennium installed. I am testing 3.62, 3.61, 3.51, 3.41. Each of these driver versions has their own quirks that further complicate BIOS settings. More on this later. It is too early to be certain, but I might have a stable configuration using the 3.62 drivers (Business Graphics 87.1, High End Graphics 39.1). If stable, this score challenges a P200MMX at 3X66.

JULY 02, 1997: Although I find it hard to believe, I have the FX83/Pro200/512 running stable using 128MPEDO with 50nS timing and the Matrox 3.62 drivers. If you have followed that saga of the 3.62 drivers you know that they are tricky but fast. I completed "all tests" in Winbench97 at both 60 and 50nS. The BIOS settings under chipset setup are all disabled except for PCI Frame Buffer USWC. Interestingly, disabling these settings does not appear to hurt performance when using the 3.62 driver. Enabling Read Around Write and CPU to PCI posting increases performance with other drivers, but these settings make the 3.62 configuration unstable. Having only the PCI frame buffer enabled, allows using 3.62. The 3.62 driver gives more performance than what is lost by disabling BIOS settings. That is, I can enable more settings using the 3.41 Matrox driver or the 3.41 Nine driver. Enabling these BIOS settings does not make up for the performance inherent in 3.62 if your able to run it. Matrox appears to have released a temperamental race car. They retracted the driver. If you configure for the 3.62 driver, there is considerable gain in performance.

I am using the 3.62 configuration for a few days in application computing and continue to run benchmarks. If the configuration is truely stable, I will begin publishing charted benchmarks for the FX83/Pro200/512. Arriving at this optimal configuration has been a long road. There were many counter intuitive paths. Perhaps AMI could produce a BIOS more friendly toward the Pro CPU or two BIOS versions depending on the CPU in use. To be fair, the MMX is more amenable to over clocking. Although the Pro holds its own at the same clock values, the MMX can move up with faster bus speeds. If the Pro holds its own against the Pentium MMX in OSR2 (actually, it dusts the MMX in all but graphics), I suspect the FX83/Pro will pull far ahead in NT4.0. I will be comparing OSs' on the FX83 down the road.

For now, expect to see HX83 vs FX83 comparisons. Considerations of price and performance undoubtedly casts value on the differences. In the end, let me say that this FX83 kicks serious $%@^.

JULY 03, 1997: Another rather unpredictable quirk surfaced in the BIOS settings. Despite, that the FX83 runs at 50nS with the Matrox 3.62 driver (described above), the performance appears noticeably improved when running at 60nS. Oh well, better performance at what appears to be a less aggressive setting, I take it. Something about looking a gift horse in the mouth comes to mind. Nevertheless, apparent contradictions of this nature questions the maturity in the BIOS.

JULY 04, 1997: The HX83/200MMX is running smooth as silk these days. I am using it at 3x75 as a work horse for my "real" job. Testing continues on the FX83/PPro200/512. Unable to resist, I temporarily over clocked the FX83 to 233. It cranks out a cool CPU32 of 725. There is an occasional blue screen declaring a failed VxD. Nevertheless the system continues without locking -- even when running Winbench97. I suspect that, at the 233 clock speed, the 3.62 will have to be replaced by 3.41. The OS has taken quite a beating. I am still on the second install of OSR2. Scandisk has activated dozens of times as I pushed the limits of performance. Repartitioning the Cheetah and reloading the OS might stabilize the 233 configuration. Soon, I will publish HX83 Vs. FX83 benchmarks.

JULY 04, 1997: I tested "real world" comparisons between the HX83 and FX83 systems this morning. As part of my "real" job, I work with digitized histological data. For this, I use a Sony DKC5000 digital camera. This camera has outboard memory and transfers image data to the PC over the SCSI bus. The image size is fixed at 5M (TIFF format). Often, I work with 10 (and more) images in memory at a time. Thus, it is not unusual for me to select 10 thumbnails in Picture Publisher representing 50 Meg of image data stored on the Cheetah. The FX83/Pro200/512 system, running at 200MHz, loads and displays 10 of these images in 13 seconds. That is, the FX83 loads and displays 10 images of 5M each for a total of 50M of data in 13 Secs. The HX83/200MMX configured with the same hardware and running at 225 Mhz (3X75) loads these same images in 30 Secs. The magnitude of this difference is not predicted from standard benchmarks. I only wish I had a PII 233 to make this same comparison. I spent a fortune on the Pro200/512. I feel better to know that, for what I do, the Pro has an edge over the much less expensive 200MMX. To be fair the Pro has double the installed RAM at this time. This might account for the difference.

JULY 05, 1997: Oops, as it turns out, I had a bad SCSI cable. On a few lazy occassions, I tested the FX83 with the cover off and laying on it's side. The blue screens at 233MHz went away. I doubted that heat had anything to do with it. If anything, the system runs hotter with the case off. I modified the case with high CFM fans, all of which are hardwired to run continuously. The Cheetah and 3.5" frame remain comfortably warm -- not what I would consider hot. I suspected that there was a cabling problem. Close inspection of the SCSI III cable to the Cheetah revealed that one side of the cable retainer was not seated. I removed the connector entirely, cut back the ribbon, and reinstalled the connector with a bench vise. Susequent benchmarks showed higher Disk Marks. Nevertheless, all of what I have reported to date regarding BIOS settings appears to hold. I am considering buying a set of Teflon insulated silver SCSI cables. I know that the engineers balk at this sort of thing. I use home brew teflon/silver inconnects and power cables in my hi fidelity system. Engineering types have all but sent me letter bombs. Oh well, another story.

The FX83/PPro200/512 is running rock solid at 200MHz with 128M PEDO and Matrox 3.62 drivers. For practical purposes, it runs fine at 233. However, I have this stubborn criterion that I will not compare benchmarks unless the configuration successfully completes all tests of Winbench97. Although each individual test is completed, when running "all tests" the system locks on high end disk marks (no more blue since repairing the cable). Winbench97 scores for individual tests at 233MHz (1024X768X16@85) are: High End Disk = 5300, Business Disk = 1750, CPU32 = 725, Business Graphics = 99, and High End Graphics = 43.6. I suspect that stable 233MHz might be possible with ultra high quality EDO RAM rather than the PEDO that I am using. I should have 128M of Advanced Megatrends Memory (EDO) in a few days. This memory is for another system, but I will intercept it long enough to test in the FX83.

On the other hand, I am not sure that different memory will matter. Unlike the PentiumMMX and the 430HX chipset, the memory timing on the FX83/Pro200/512 has only minor effect on performance. In fact, going from 50nS to 70nS causes only a minor drop in performance. Running at 70nS has a permissive effect on enabling other BIOS settings. Thus, using 70nS timing actually increases performance by allowing stable operation with additional BIOS features enabled.