Can a BIOS affect RAM Bandwidth Performance?

[KachiWachi]

Hi all -

Can this happen?

As you know, I have two i430VX boards...my Amptron PM-7900 (PCChips M520), and my DFI 596IPVG. I have recently found that the RAM Write Speed is MUCH slower on the DFI than its Read Speed, and VERY SLOW as compared to the Amptron.

The only hardware difference is that the Amptron is using the Rev. 1 (SU085) Northbridge, while the DFI has the Rev. 2 (SU116) version. There doesn't seem to be any changes between revisions (that I could find in the datasheets and errata) that would affect this, though I can't say for sure.

Please see the last entry in my DFI post (sig) for more specific info.

Thanks.

[Rainbow]

Looks like the chipset registers are configured for lower performance on the DFI board.

[KachiWachi]

Rainbow -

Both are configured the same. I used Craig Hart's PCI program to dump both machines (Northbridge) registers, and they match with respect to the RAM timings.

Anything else that could affect this?

[atang1]

Are you using the same cpu? Same cpu stepping?

Even though chipsets are the same; 586 is slower than 686, due to extra instructions in 686? Later rev. chipset may not be used to its fullest?

Then, EDO and sdram sometimes were mixed on those motherboards, wait state assignment might be different? Are there any undocumented burst features in the bios? On board cache might be different?

Can you straighten me out on these issues?

Will you try the K6-3 on the DFI to see if it runs faster or writes faster?

[KachiWachi]

1) Different CPU's. The DFI uses a K6-2/+ 450 ACZ (@6x66 for 400MHz) while the Amptron uses an i200 non-MMX CPU (3x66MHz).

2) Not sure what you are asking/stating here.

3) EDO is a matched set, but different sets between the two boards. Both are 60nS. The Amtron has OKI chips, where the DFI's are TI (If I remember correctly).

4) Can't try the K6 on the Amptron until I get a VRM for it, and the BIOS patched to accept it.

I was recently able to try the Unicore BIOS in the Amptron/M520, but it did not change the results much...still investigating that though.

See this post for the data taken so far.
-> http://www.stud.fernuni-hagen.de/q39981 ... topic=3323

[KachiWachi]

Just re-visititing this post...

In reviewing the data in spreadsheet format, it becomes obvious that the Main RAM Read Speeds are about the same...it is only the Write Speeds that are slower.

At 66MHz (15 nS) and with no caches enabled, the difference is 151 - 90 = 61 nS...or about 4 clocks.

What could introduce a 4 clock "delay" into the Write path only? BIOS wait states? Why would a mainboard designer want to do this?

[KachiWachi]

Here in the data in an easy-to-read format -