VOGONS


First post, by Moogle!

User metadata
Rank Member
Rank
Member

Brought it inside and went about cleaning it up. It had an Asus P5S-VM with k6-2 380 (95 x 4) and 64MB ram. I have since updated the BIOS from the limited HP supplied Phoenix to the Asus supplied Award, v1005, which gave me more options and settings. (Getting the new BIOS written turned out to be an issue itself). The Motherboard has the (in)famous Sis 530, with 'support' for the 133MHz FSB.

I have been experimenting with various configurations using the K6-III 450 and memtest.

133 x 3.5 (466) with motherboard cache: Errors
133 x 3 (400) with motherboard cache: Errors
124 x 4 (496) With Motherboard cache: No errors
133 x 3.5 (466 without motherboard cache: No errors

I played with this for awhile, trying a few things like changing voltages, changing ram modules (all used were 7/7.5 ns (133mhz) modules), putting a better heatsink on the chipset, and installing an external video card instead of using the onboard video. None of these helped. So I started looking at the board itself. I noticed that while the motherboard cache is rated at 5ns (200 mhz), the cache chip's tag ram is only rated for 8ns (124Mhz), and I believe this is why the board isn't stable with the cache turned on at 133Mhz. (Incidentally, I have seen older style 286/386 simms with tag chips that are slower than the main ram chips, so if you have problems, see if that is the cause.)

Any thoughts or questions, Vogons?

Reply 1 of 12, by lazibayer

User metadata
Rank Oldbie
Rank
Oldbie

I had two P5S-VMs with different revisions, and one was giving me wrong voltages at certain settings. You can verify actual voltage in the BIOS if it supports voltage monitoring, and cross reference the jumper setting with the VRM chip's datasheet.

Reply 2 of 12, by gdjacobs

User metadata
Rank l33t++
Rank
l33t++

Nice snag! What's the performance differential between 4x124 and 5x100, caches on?

All hail the Great Capacitor Brand Finder

Reply 3 of 12, by Moogle!

User metadata
Rank Member
Rank
Member

Don't really do a lot of benchmarking. Did you have a program in mind?

Reply 4 of 12, by kanecvr

User metadata
Rank Oldbie
Rank
Oldbie

The on-motherboard cache does not support 133Mhz FSB on most super 7 motherboards. If you have the gear and experience, you could try to replace the tag sram with one that supports 133Mhz, but it's not worth the bother in my opinion. If you're using a K6-III, the machine should perform faster w/o the slow on-motherboard cache anyway (depending on the amount of ram you're using).

Benchmarks at 133Mhz fsb vs 100Mhz fsb would be interesting (speedsys, PCP Bench, dos quake 320x200 and 640x480 as well as GL_Quake).

Reply 6 of 12, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

You may also have PCI timing errors, like the 83Mhz problem. If it gives a 1/3 divider for 100Mhz, at 133 it's 44.3 Mhz.

I have seen a lot of 8ns chips that will actually do 133, usually from before they split hairs and binned them at 7.5ns because the next bin was 7ns, so a lot of 8ns chips might have actually been close to 7ns, like 7.1, 7,3, etc, but after 7.5ns binning was introduced they got binned as that.

Edit: actually a lot of the early 7.5ns was remarked 8ns.

Edit2: Actually, personally, I'd have doubts is was the 8ns tag or RAM if it was happening soon after cold, cold boots, before you'd even stress tested it, since I've observed that any 8ns will do 133 until it gets warm. So if it's failing in first 5 mins after cold cold boot, or not even booting to OS cold, then I'd blame the PCI being out of spec.... IF you can get further with no PCI cards, then onboard stuff is just about coping, so selective swapping of PCI cards might get you one that does 44.3Mhz... if you can't, then it's onboard stuff fritzing.

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 7 of 12, by lazibayer

User metadata
Rank Oldbie
Rank
Oldbie
BitWrangler wrote:

You may also have PCI timing errors, like the 83Mhz problem. If it gives a 1/3 divider for 100Mhz, at 133 it's 44.3 Mhz.

P5S-VM supports 1/4 PCI divider at 133MHz.

Reply 8 of 12, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

This might be worth a read....

"Super 7" SiS 530 chipset.

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 9 of 12, by Moogle!

User metadata
Rank Member
Rank
Member

Ok. Four sets of SpeedSys benchmarks. All benchmarks were performed with an AMD K6-III 450, running at 2.3v instead of the usual 2.2v.
(Didn't want to try a 133Mhz at 4x multiplier, I don't want to risk screwing up this CPU)

100Mhz x 4.5 for 450 mhz with cache
vmyESKx.gifqXjlBNL.gif
124mhz x 3.5 for 434 mhz with cache
ttIy8zE.gif39p70HX.gif
124mhz x 4.0 for 496 mhz with cache
ksaHQVb.gifTEDzCXy.gif
133mhz x 3.5 for 466 mhz No cache
qBQeW6A.gifdbaaWPQ.gif

Reply 10 of 12, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

Looks to be performing well.

Moogle! wrote:

Ok. Four sets of SpeedSys benchmarks. All benchmarks were performed with an AMD K6-III 450, running at 2.3v instead of the usual 2.2v.
(Didn't want to try a 133Mhz at 4x multiplier, I don't want to risk screwing up this CPU)

I'm not sure how delicate the L2 was on these. K6-2s were really hard to kill, people would have them at 2.7 volts to hit the high 500s or 600, but I wouldn't recommend running them a long time over 2.5.

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 11 of 12, by kanecvr

User metadata
Rank Oldbie
Rank
Oldbie

Note how memory performance increases when not using on-motherboard cache - it's a slight improvement, but it's there. I'd personally use the machine with on motherboard (L3 cache) disabled. Great little rig you've got there - wish I had a SiS super 7 board to play with myslef 😜

Could you do some Quake or PCP bench tests at 133Mhz FSB as well?

Reply 12 of 12, by Moogle!

User metadata
Rank Member
Rank
Member

I just found Phil's collection of Benchmarking programs, so I may post some of those soon. At some point I'd like to dig out my windows 98 HD and see what kind of framerate I can get with Kega Fusion and Knuckles Chatoix.