h-a-l-9000 wrote:

PCICLK-to-ISA SYSCLK Divisor: PCICLK/3 | PCICLK/4 | PCICLK/2

Is it me, or does it seem that the absence of a larger then PCICLK/5 divider will make it difficult to get 486 PCI boards to run at fsb's past, say, 50Mhz?

Is the PCI bus directly connected to the fsb? (on a 486 board)

Tetrium wrote:

Is it me, or does it seem that the absence of a larger then PCICLK/5 divider will make it difficult to get 486 PCI boards to run at fsb's past, say, 50Mhz?

It rather is the absence of a divider between FSB and PCI bus. 40 MHz on PCI is already rather high, there is no chance it'll work at 50 MHz. 16-bit ISA cards rarely have trouble at 10 MHz.

Tetrium wrote:

Is the PCI bus directly connected to the fsb? (on a 486 board)

On most of them, yes. I vaguely remember reading about one board (ALI chipset?) with a FSB/PCI clock divider. It should be somewhere here on Vogons.

5u3 wrote:

It rather is the absence of a divider between FSB and PCI bus. 40 MHz on PCI is already rather high, there is no chance it'll wo […]
Show full quote

Tetrium wrote:

Is it me, or does it seem that the absence of a larger then PCICLK/5 divider will make it difficult to get 486 PCI boards to run at fsb's past, say, 50Mhz?

It rather is the absence of a divider between FSB and PCI bus. 40 MHz on PCI is already rather high, there is no chance it'll work at 50 MHz. 16-bit ISA cards rarely have trouble at 10 MHz.

Tetrium wrote:

Is the PCI bus directly connected to the fsb? (on a 486 board)

On most of them, yes. I vaguely remember reading about one board (ALI chipset?) with a FSB/PCI clock divider. It should be somewhere here on Vogons.

5u3, in this Vogons thread, you say (half way down page 1) "... apparently it has a configurable PCI bus divider (Jumper W20), which greatly increases the chances to run a 50 MHz FSB! 😉" I appreciate that this board is not ALI chipset based, but is this the board you were referring to?

^^^ Yes, it is the board. 🤣 OK, let's start again: I vaguely remember writing about one board....

Seriously, this one has good chances to hit 50 MHz, and the buses will stay within specs up to 66 MHz. The biggest problem on that board will be the L2 cache and the TAG RAM. How fast are the chips?

The fact that retro's board would even boot at 50mhz when he tested the SX/33 tells me that there's something going on to keep the PCI speeds in check... if it were actually running the PCI bus at 50mhz, you probably wouldn't have even been able to see the POST display (given that it's a PCI video card).

I dunno if it might be automatically setting a PCI divider, or if it's using a separate fixed-frequency clock to run it...

5u3 wrote:

^^^ Yes, it is the board. 🤣 OK, let's start again: I vaguely remember writing about one board....

Seriously, this one has good chances to hit 50 MHz, and the buses will stay within specs up to 66 MHz. The biggest problem on that board will be the L2 cache and the TAG RAM. How fast are the chips?

Most 'modern' cahce chips will be rated at 15ns. How does that translate to Mhz?
Another thing that could be worth a shot is to try a VLB board. Many should be able to run at 50Mhz (given you find a graphics card that will work reliably at that speed) which in theory could be clocked to 200Mhz.

I'll have to find the list of my motherboards to see which ones have that magic divider 🤣!

Old Thrashbarg wrote:

The fact that retro's board would even boot at 50mhz when he tested the SX/33 tells me that there's something going on to keep the PCI speeds in check... if it were actually running the PCI bus at 50mhz, you probably wouldn't have even been able to see the POST display (given that it's a PCI video card).

I dunno if it might be automatically setting a PCI divider, or if it's using a separate fixed-frequency clock to run it...

OT's right. This could be worth investigating

Edit: Almost all the 486 PCI boards have 'Factory configured - do not alter' jumpers...there might be something in it?

15ns would be 66,6Mhz, but I think it's more important to know the speed of the TAG-RAM. Maybe it's 15ns also.

unmei220 wrote:

15ns would be 66,6Mhz, but I think it's more important to know the speed of the TAG-RAM. Maybe it's 15ns also.

The tag ram is just the 9th sram chip, right? Or is it one of those very tiny ones often soldered on the motherboard? If it's the 9th one, those often are of the same speed as the other 8 so shouldn't be a problem then

Edit: Yay! Found out another board that has this jumper, the Chaintech 486SPM and I have it! 😁
Dunno if it works or not though...
Link: http://stason.org/TULARC/pc/motherboards/C/CH … 486-486SPM.html

5u3 wrote:

Seriously, this one has good chances to hit 50 MHz, and the buses will stay within specs up to 66 MHz. The biggest problem on that board will be the L2 cache and the TAG RAM. How fast are the chips?

OK, this sounds really exciting! 😀 How fast are the chips? - hopefully, I will be able to determine this on Sunday. Do I examine the labelled identification on the cache chips themselves, or do I run a cache analysis utility in DOS? If so, which one please? Thanks a lot.

Old Thrashbarg wrote:

The fact that retro's board would even boot at 50mhz when he tested the SX/33 tells me that there's something going on to keep the PCI speeds in check... if it were actually running the PCI bus at 50mhz, you probably wouldn't have even been able to see the POST display (given that it's a PCI video card).

I dunno if it might be automatically setting a PCI divider, or if it's using a separate fixed-frequency clock to run it...

Actually, when I tested the SX/33 on the SYL8884PCI-EIO mobo, it was at 60 MHz! What I need to do is put the SX/33 back in to the mobo again, and run some more tests, to check system stability. I reckon a fairly good test is to see if Windows 98 boots and runs normally. I could run SuperPi.exe, for example. Also, I've got a DX/33 around here somewhere. I could also use that, for some further tests. I can do all of this on Sunday or Monday, and I'll post back with the results...

retro games 100 wrote:

Do I examine the labelled identification on the cache chips themselves, or do I run a cache analysis utility in DOS?

Usually the access time (in ns) is printed onto the cache chips, at the end of the chip number (e.g. -15 or -12).
Here is an article about cache on old machines.

I don't think there are many 486 boards with a PCI divider or an asynchronous bus speed. After all, only a few CPUs were intended to run at a higher FSB than 33 MHz. By the time those boards came out, they had to be very cheap, because power users were switching to Pentiums already, so it wouldn't have been economical to implement bus dividers.
Besides, running the PCI bus at half the clock rate probably ruins the advantage of the fast FSB.

Thanks for the info! I can check the cache chips immediately. They are, perhaps unfortunately, all labelled -15. There are 5 chips in total. 1 chip looks smaller than the other 4. Something about the info on the chips intrigues me. Here is the information:

UM61 M 256k-15 (smaller chip)
9602D RB0105

HM62H512BK-15 (same identification for all 4 chips)
9637B L69B1 (same id as chip 3)

HM62H512BK-15
9638B L69I2 (same id as chip 4)

HM62H512BK-15
9637B L69B1

HM62H512BK-15
9638B L69I2

Firstly, sorry for my ignorance, but is this 512K cache? If it is, then I need to set the mobo's cache jumpers from 256K to 512K! Also, do I add another stick of RAM, to increase the RAM size from 32MB to 64MB?

Secondly, I wonder why cache chip numbers 1 and 3 are the same, and cache chips numbers 2 and 4 are the same? Thanks a lot for any additional info.

retro games 100 wrote:

Thanks for the info! I can check the cache chips immediately. They are, perhaps unfortunately, all labelled -15. There are 5 […]
Show full quote

Thanks for the info! I can check the cache chips immediately. They are, perhaps unfortunately, all labelled -15. There are 5 chips in total. 1 chip looks smaller than the other 4. Something about the info on the chips intrigues me. Here is the information:

UM61 M 256k-15 (smaller chip)
9602D RB0105

HM62H512BK-15 (same identification for all 4 chips)
9637B L69B1 (same id as chip 3)

HM62H512BK-15
9638B L69I2 (same id as chip 4)

HM62H512BK-15
9637B L69B1

HM62H512BK-15
9638B L69I2

Firstly, sorry for my ignorance, but is this 512K cache? If it is, then I need to set the mobo's cache jumpers from 256K to 512K! Also, do I add another stick of RAM, to increase the RAM size from 32MB to 64MB?

Secondly, I wonder why cache chip numbers 1 and 3 are the same, and cache chips numbers 2 and 4 are the same? Thanks a lot for any additional info.

I've downloaded almost everything here:
Link: http://www.chipmunk.nl/DRAM/DRAM.htm
It may help in deciphering what those numbers are!

Also, theres a second (and VERY easy) way to check if your cache chips are 32k or bigger (64k / 128k).
On boards with 256k total I noticed all 9 chips are of the same size. If you, for instance, have 256k in 5 chips (1 as big as the 32k should be, and 4 which have 4 pins more, 2 on each side) then you have 4 64k chips.

My point is, cache chips larger then 32k each are larger. You'll have to check the exact numbers to differentiate 64k chips from 128k chips though.

For example;if you look at my pic with the 9 mobo's again:
Link: http://i941.photobucket.com/albums/ad254/inte … um/DSC00201.jpg
, the middle one has 5 chips in total, of which 4 are larger (have 4 more pins, 32 instead of 28 iirc) and is jumpered for 256k total memory.
Also many of the boards have 9 cache chips in total with all sockets populated, but many of those sockets are larger then the chips that are placed there.

Just by looking at your pic I can tell it has 1 28 pin chip and 4 32 pin, so you have either 256k or 512k total L2 😉
Mine has the smaller ones, alas 😜

Edit: Your numbers aren't on the link I posted but judging from the numbers myself I'd say your mobo is jumpered correctly

1st chip:
UM61 M 256k-15
other chips:
HM62H512BK-15
so twice as large in memory size.
So your total amount of cache should be 64k x 4 = 256k 😉

Thanks very much for the info! I took a photo of the cache, and all 5 of the chips occur the entire area of the cache sockets space.

I've just had time to take a photo of the "magic" configurable PCI bus divider (Jumper W20) motherboard. I can't operate on it for a day or two however. Must get some other work done, besides retro fun and games!

retro games 100 wrote:

OK, this sounds really exciting! 😀 How fast are the chips? - hopefully, I will be able to determine this on Sunday. Do I examine the labelled identification on the cache chips themselves, or do I run a cache analysis utility in DOS? If so, which one please? Thanks a lot.

As I already said, 15ns chips are 66 Mhz rated speed. Are my posts being ignored ? That doesn't necessarily mean they can't run faster.
Also, you have 256K of cache, don't pay attention to the 512 part written on the chips, I doesn't necessarily mean how much cache they are. The smaller chip is the TAG-RAM.
If you change the jumpers to specify more cache than the one you have, then probably the machine will hang somewhere, but no harm will be done, so you can try experimenting if you like.

OK, thanks a lot for the explanation. I adjusted the mobo's cache configuration jumper from 256 KB to 512 KB, and you were right, the mobo simply hung during BIOS POST boot up time. I then switched back to the original 256 KB mobo jumper config, and all was well again.

unmei220 wrote:

As I already said, 15ns chips are 66 Mhz rated speed. Are my posts being ignored ? That doesn't necessarily mean they can't run faster.
Also, you have 256K of cache, don't pay attention to the 512 part written on the chips, I doesn't necessarily mean how much cache they are. The smaller chip is the TAG-RAM.
If you change the jumpers to specify more cache than the one you have, then probably the machine will hang somewhere, but no harm will be done, so you can try experimenting if you like.

I did read it, unmei220 😉
And I also mentioned that if the cache can run at 66Mhz, we should have no problem overclocking this cache 😉

The 512 part usually means it's a 512kilobit chip which will translate to 512/8=64kilobytes.
Also, looking at the 256 part on the smaller chip and the fact RG100's mobo is already jumpered to 256kb I can conclude his chips are in fact 64kb chips, making a total amount of 256kb of L2 cache on his board.

And it wouldn't surprise me if I was in fact correct on the matter 😉

And besides, why would someone put 512kb of cache on a motherboard, jumper it to 256kb knowing that back in the day L2 cache was the single most expensive part of a motherboard? Wouldn't make any sense.

And about the higher fsb meaning the pci bus will have to get lowered, it will depend on if we can find a high-clocked 486 chip with a low enough multiplier so we can still get an improvement in performance.
One example: Put an AMD DX4-120 and set the multi to x2, set the PCI divider to 4 so we get a 33Mhz pci bus. Next set the 'magic' jumper to double the fsb and we got 33Mhz x 2(doubled fsb bus) x 2(because of the cpu multiplier) = 133Mhz cpu!
In theory we could get an AMD 5x86 to 33 x 2 x 3 = 200Mhz this way with a pci bus running at 33Mhz and the L2 cache running at 66Mhz.
Only problem would be to find a cpu that can run at this, crazy, speed in a stable way.

Edit: wowzers, I took over 15 minutes just to write this post?

Tetrium wrote:

Edit: wowzers, I took over 15 minutes just to write this post?

Time well worth spent! 😁 Thanks!

I've just discovered 2 interesting things. Firstly, I am using the mobo in the O.P. - the one where I was trying to get the AMD P75 to work @ 60 FSB. Well, I am now using an SX/33 CPU. I set the FSB to 60 FSB, and what's this? I get no POST. Oh dear, something wrong? Well, yes and no. The problem is my Dell S3 Nitro GX card. It doesn't seem to be able to "tolerate" 60 FSB. The solution? A Diamond Viper V330 Riva 128 video card. This can tolerate this overclocked environment. Onwards...

The second interesting thing is that when the FSB is run at 60 FSB, I see this new message on the BIOS POST screen: WRITE BACK CACHING ACCELERATED. Now that I have the board working at 60 FSB, I am going to see how it behaves in Windows. I'll post back later.

I have finished testing, and it was successful. Quick recap: SX/33, 60 FSB, Windows 98. Win98 works fine. I ran the win98 "hardware wizard" on a compact flash drive which was used for a slot 1 system, and it identified appropriate hardware for the 486 mobo, removed the slot 1 stuff, and installed everything OK. At the desktop, I decided to increase the colour depth of the Riva 128 card from 16 colors to 256. The PC rebooted, and then after the Win98 splash screen appeared, the monitor just showed a blank screen.

Because of the slowness of using an SX chip with Win98, I didn't spend any time trying to fix the problem with the Riva 128. Instead, I removed it, and replaced it with a Virge 325 card, manufactured by Number 9. Yay, that works. It POSTs, and it loads Windows 98, and at a useable color depth. The system appears normal. Apart from the Riva 128 issue above, it's been running for 45 minutes or so, and nothing weird has happened.

At this stage, I could look for the undocumented 66 FSB setting, but I can't help but wish I could get either the AMD P75 to work at 180 MHz (3x @ 60 FSB), or the Intel DX-4 at 150 MHz (3x @ 50 FSB).