VOGONS


Reply 100 of 111, by FullYes

User metadata
Rank Member
Rank
Member

Ok I can feel I’m pushing the limits of this machine now as I’m starting to have more and more issues

The single sided 128MB stick still won’t work on its own. Perhaps the machine needs 2 ranks to boot properly, or the stick has become faulty after me removing and refitting it so often.

I decided to try and see if I could get the machine to boot at 83Mhz FSB. I tried 83x6. Nothing. 83x5.5. Nothing. 83x5. Still nothing. It should work at 83x5 even with this poor man’s k6-3.

I tried 4.5x. Now we are in business. It still reports the CPU frequency as 450Mhz in BIOS. It’s probably a mistake. It should be about 375MHz

The system isn’t very stable but I’m able to boot to the command prompt and do a couple of benchmarks

Chkcpu reports the system as 450MHz. running 100MHz FSB. WTF?

Ran speedsys which seems to confirm it. I will have to get a screenshot tomorrow when I have more time

I was able to do a couple of benmarks, but I was not able to boot into windows. It’s too unstable

Quake 320 x 200 - 89.7fps (hard locks on quit)
Quake 640 x 480 - 33.9fps (hard locks on quit)
Speedsys
CPU score 510, L1: 2134, L2: 1351, Ram 203.82MB/s
Compare to 6x75:
CPU score 510, L1: 2129, L2: 1352, Ram 152.43MB/s
This would seem to indicate that it is indeed running at 450MHz???

I had to take a look at the clock generator on the board and the jumper settings I’d chosen. There are 2 different clock generators mentioned in the manual, and I didn’t realise it but BOTH configurations are printed on the motherboard, and I was using the wrong configuration for 83MHz for the clock generator my motherboard actually has fitted - when I was testing the Pentium at 83MHz, I can only assume that I was using the correct setting for 83MHz when I was testing that CPU

My clockgen is an IMI SC643AYB. I had a look online for a datasheet incase there are some undocmented settings that allow for 100MHz.

I found this datasheet on the retroweb. But the settings don’t all line up with what’s in the motherboard manual. The tables in the manual and on the motherboard itself match up ok.

https://theretroweb.com/chip/documentation/im … 02159427507.pdf

There is no mention of 100MHz FSB on the datasheet and as I mentioned already there are a couple of settings that don’t match up. Time for another spreadsheet!

What’s interesting is there seems to be a provision for a 5/2 clock divider for the PCI bus with this generator. Which would explain why 100MHz *could* work as this would give a PCI clock of 40MHz which is somewhat attainable.

I have my oscilloscope now so I can measure bus frequencies. I was having a hard time picking up the signal on the Athlon system and I have tried on the socket 7 yet. It may be easier with the motherboard out

Reply 101 of 111, by FullYes

User metadata
Rank Member
Rank
Member

If I flip the bits mentioned in the datasheet and swap the 3rd and 1st column around, then I end up with 4 sets of matching data, and 4 sets of mismatching. I have tried to find other patterns but this is the only one that comes close to making any sense.

The attachment table2.png is no longer available

I'm planning to look at some 440LX / BX motherboards on the retroweb to see if any also use this clockgen. The datasheet says it's a pentium II class generator

Reply 102 of 111, by FullYes

User metadata
Rank Member
Rank
Member

I read the datasheet again. It is only designed for Pentium, Pentium pro and Cyrix CPUs in mind (no AMD?) I didn’t find it on any other MSI boards of the era. I guess I was hoping to see it on a SS7 board, or a slot one or socket 370 with 100MHz as one of the options

Anyhow. I will test all of the iterations with an oscilloscope and report my findings

Reply 103 of 111, by FullYes

User metadata
Rank Member
Rank
Member

I managed to get the oscilloscope to give me some numbers. It’s a pain to measure the PCI clock but it’s not so difficult to measure the ISA clock. And since I know the divider I can work out the PCI and FSB frequencies from that

I forgot to take a picture at 66MHz but I confirmed the ISA frequency of 8.33MHz. I was more interested in 75, 83 and the undocumented setting!!

Below is the ISA frequency at 75MHz. This would suggest 37.5MHz PCI clock (as expected)

The attachment IMG_8606.jpeg is no longer available

83MHz FSB / 41.6MHz PCI / 10.4MHz ISA

The attachment IMG_8607.jpeg is no longer available

100MHz FSB. Oh dear, no wonder it’s not very stable. PCI bus is running at 50MHz!

The attachment IMG_8608.jpeg is no longer available

What a shame it’s not running at that 5/2 divider mentioned in the datasheet!

Reply 104 of 111, by FullYes

User metadata
Rank Member
Rank
Member

Had a bit of a play around again today.

I managed to finally measure the PCI clock at one of the slots. Confirmed it’s 50MHz with the ‘scope when I set it to the claimed 100Mhz FSB

Here is the table of results i promised

The attachment IMG_8615.jpeg is no longer available

I’ve finished the tests I could do at the speeds achievable from this k6-3 chip. It won’t boot at 500MHz (83.3x6) the maximum voltage I have availble is 2.5V (it’s a 2.4V CPU) the next step after 2.5 is 2.8V

Some screenshots of DOS testing at 100MHz FSB. All tests completed fine at 2.5V

There is no chance of booting into windows unless I can find a PCI ide/sata controller with bios, which will be ok running at 50MHz PCI speed…

I guess if I had a better k6-3 or 3+, I could get some very nice benchmark results. Is it possible to get 100FPS in Quake bench in a socket 7 platform???

Reply 105 of 111, by RetroPCCupboard

User metadata
Rank Oldbie
Rank
Oldbie

Must have taken ages to do all those benchmarks. If your 3DMark results are any indication, the K6-3 has a much bigger lead over the 233MMX in Windows than it does in DOS. I guess that makes sense as DOS games aren't really designed to take advantage of it. Writing to the graphics card is probably bogging the CPUs down.

Reply 106 of 111, by FullYes

User metadata
Rank Member
Rank
Member

That’s not even the full set!! I’ve omitted several combinations so it’s easier to see the progression

I started doing the benchmarking about a month ago. I did some before i got the voodoo3, but all of the ones in the attachment are from afterwards, so yeah there’s a few hours lost there to the hobby!

And yeah the Pentium holds up pretty well in DOS by the looks of it, but I think the scailing at 320x200 of the k6-3 is hindered by the platform as much as the CPU. I should’ve tried a few more 640x480 benchmarks in DOS with the Pentium. I think there will be a bigger difference between the two if I do that.

I may rerun some tests on the Pentium setup. Now I know a bit more I might be able to improve the results. With the S3 I was running univbe, but the voodoo wasn’t supported (and was still faster than the s3 without any driver) I’ve not tried fastvid on the Pentium to see if that boosts anything

The windows results might be skewed due to the 3dnow! Optimisations in 3dmark99. I should probably try another benchmark like GLQuake. It’ll be interesting to collect more data

Reply 107 of 111, by RetroPCCupboard

User metadata
Rank Oldbie
Rank
Oldbie

Would be interesting to compare with what speed you get in DOS benchmarks with the socket A system. You'd think it would blow the Socket 7 system out of the water. But, I think, for most DOS benchmarks, it probably won't be as big of a lead as you'd expect from clock speed difference. Quake would probably increase the most.

Reply 108 of 111, by FullYes

User metadata
Rank Member
Rank
Member

I’ll look into that but yeah I think everything will just be over in a flash on that 🤣

I haven’t touched the socket a for a while. I’m in a bit of a quandry with it:

Since I have the socket 7, Do I need the socket a to have DOS compatibility? (probably not TBH)

If I don’t need full dos compatibility, (ie the isa slot) I could run a later socket A board. I already have the A7V333 - it was in the system I bought for the 5900XT. This would give me less trouble with the Barton and and give me more performance thanks to its DDR memory. I already have an audigy card as well, which should give me better sound in early 2000 games with eax

Since pre-2000 games will now run alright on the socket 7, do I need another windows 98 PC at all? I guess I need to try some more games before deciding what I do. As the gap I need to bridge now between the socket 7 and the windows XP system I have is pretty small

Something else interesting (to me) I was in another thread about the ultimate XP setup. One of the last cards that will work in XP is the 980 TI. I bought one new, but it’s been in my son’s PC for the last few years. It’s just become available again. They don’t fetch a great deal on eBay these days but I am planning to sell it. It might be worth one last hurrah before I do

Reply 109 of 111, by RetroPCCupboard

User metadata
Rank Oldbie
Rank
Oldbie

There are some DOS games like screamer 2 (in hi-colour, high res mode) and System Shock (at 640 x 480) that would most likely run poorly even on your K6-3. But, I think they are an exceptional case. Most DOS games will run fine, and these will too at lower settings

Reply 110 of 111, by FullYes

User metadata
Rank Member
Rank
Member

Have you ever played Big Red Racing? This has a hires mode and it’s one of the games that has been transformed going to the k6-2/k6-3. It’s finally playable in hi res. K6init makes a massive difference as well.

I will have a look at those games. Never played system shock!

Just tried GLquake. 125fps at 640x480 running at 75x6.

I’ve decided this is as far as I want to go / need to go with this setup. I have visions of it running a k6-3+ at 600MHz using a PCI add-in IDE card. I’ve found a suitable card for this. But I’d still need a CPU. And then there’s no guarantee it’ll actually be ok running 24/7 at 50MHz PCI frequency.

Reply 111 of 111, by RetroPCCupboard

User metadata
Rank Oldbie
Rank
Oldbie

I have heard of it, maybe played a demo, but never had the full game. I will have to check it out