• Please review our updated Terms and Rules here

The real unicorn: a working 40 MHz VLB bus

AriesMu

Member
Joined
Jun 23, 2016
Messages
26
Hello fellow retrocomputerers!

With 40 MHz FSB 486 CPUs (such as Am486 DX2-80 and Am486 DX4-120), what VLB and PCI motherboards can surely have the actual VLB or PCI slots actually running at actual 40 MHz without needing to add crappy wait states or down-clocking the slots at 27 MHz?

Do you know if a motherboard with the UMC chipsets UM8498F and UM8496F can be the above unicorn?

Thanks all!
 
VLB always(*) works at the FSB frequency, so with a 40/80/120 MHz CPU it's 40 MHz, a pretty common configuration back in the era, nothing like a unicorn.
Now, running VLB at 50 MHz would be something, I heard rumors of somebody making it, but never any details.

The original PCI works at 33 MHz, in case of a 40 MHz FSB they work asynchronously.

(*) there must be exception with those rare cases of Pentium boards with VLB. Early Pentriums have 60 or 66 MHz FSB, so I guess VLB is run at 30 or 33 MHz.
 
I have run a 486DX50 before on VLB. I don't recall right offhand what chipset it was, but it was a Biostar board with two VLB slots. As I recall I had an Ultrastor 34F SCSI controller, a 5.25 full-height Seagate 11GB drive, and a Number 9 VLB video card. It worked, but it was finicky!
 
40Mhz is not the problem, 50mhz is on VLB. You do understand that the VLB bus sits on the CPU bus and has to deal with traffic from cache and RAM at the same time as card transfers. Everybody dropped VLB because of the bottlenecks and went to PCI which was its own BUS not interfering with FSB of the CPU.
 
wow guys, thanks so much, I'm discovering so many things I didn't know!!! Interesting.
And, for the first time in my life, I hear the word "bottleneck" associated with VLB! Can you believe that!??? I've been blaming myself for having just purchased my first personal computer (a 486) as a ISA-only (1993) as I didn't know yet about what a bus was, to then find out about that thing, and drooling over dreams about having a VLB computer, which I should have purchased as a whole because mine was a Compaq, so I couldn't even simply replaced just the motherboard.... and all that just to avoid "THAT": the ISA "bottleneck". And now, after more than 25 years, I hear that VLB was bottlenecked LOL!
 
I always ran my Biostar VLB board with a 486DX2-66, so I guess the bus was at 33 mHz. I might dig it out and configure it to run my DX-50, using VLB disk controller and video, just to see what happens at 50 mHz. I'll report back if I have any success.
 
I always ran my Biostar VLB board with a 486DX2-66, so I guess the bus was at 33 mHz. I might dig it out and configure it to run my DX-50, using VLB disk controller and video, just to see what happens at 50 mHz. I'll report back if I have any success.

wow, thanks!
I suspect you're gonna experience problems... if so, how about you try the VLB at 40 MHz? If you don't have such a CPU, I guess (and ask if) it should be safe to donwn-clock a DX-50 to 40 MHz by setting it as an AMD DX 40??
 
I had a VLB video card operating at 50 MHz on the infamous PC Chips "fake cache" M919 motherboard with an AMD 5x86-133 CPU overclocked to 150 MHz (3 x 50).

AMD actually did release a version of the 5x86 officially designated for 150 MHz, for OEM use. This was during the late VLB / early PCI era. Which brings up the question, how did 486-era motherboards handle the PCI slots? Did they also run above the official maximum 33 MHz spec when the bus speed was set to 40 or 50 MHz?

Am5x86-P75+_150ADW.jpg
 
Didn't know about that chip. There was an Intel 486DX50 which was sold in servers I think and then they made a DX/2-50 because of the bus speed issues.
 
PCI (the original, not PCI-X or PCI-E) never ran faster than 33 MHz, if the FSB was faster they ran asynchronously.
As for the 150 MHz chip... are you sure it isn't fake?
Back in the era I saw such chips marked "160", but they were obviously fake, it was even confirmed by AMD.
 
There was an Intel 486DX50 which was sold in servers I think and then they made a DX/2-50 because of the bus speed issues.
Not only in servers, I've seen plenty of desktops with DX-50, all of them with ISA only.
With VLB, 50 MHz bus was nearly impossible, so DX-50 was replaced with DX2-50 and DX2-66.
 
PCI (the original, not PCI-X or PCI-E) never ran faster than 33 MHz, if the FSB was faster they ran asynchronously..

Not true. PCI supports a high speed mode of 66 MHz but it wasn't widely implemented. All cards have to indicate support in the config space, have to run at 3.3V I/O, and it's still up to the host controller to adjust clock. The 66 MHz speed option was introduced in 1995 with the 2.1 spec - after most early machines with PCI were on the market.
 
As for the 150 MHz chip... are you sure it isn't fake?
Back when the AMD 5x86 chips were new I did hear about a 150 MHz OEM version being officially released, but only in small numbers.

It was also widely rumored that the ADW, ADY, and ADZ temperature ratings of the 5x86 chip were originally intended to be 133, 150, and 160 MHz versions, respectively, but AMD capped the advertised speed to 133 MHz to prevent it from competing with the K5 chip, which was their underperforming Pentium competitor that was soon replaced with the K6.
 
You may find the VOGONS thread at https://www.vogons.org/viewtopic.php?f=46&t=40157 interesting, then, and I have heard of a 5x86 running 50MHz VLB with the x4 multiplier at 200MHz, but I have never tried it. I only have one ADZ 5x86, and the last I tried it it would only run up to 160MHz (40x4) and that was on a late model FIC PCI-ISA only board. It should run at 150MHz (50x3) without trouble, though, if the motherboard chipset and cache are up to it.

By the time I had a 5x86 ADZ-rated chip, I had sold the VLB board that I ran with a 50MHz VLB, so I've not tried the 150MHz variant, either.
 
I always ran my Biostar VLB board with a 486DX2-66, so I guess the bus was at 33 mHz. I might dig it out and configure it to run my DX-50, using VLB disk controller and video, just to see what happens at 50 mHz. I'll report back if I have any success.

I suspect you're gonna experience problems...

Well, in cursory testing / benchmarking, it seems my VLB MB (Biostar MB-1425/33/40/50UCV-D) runs completely stable at 50 mHz bus speed with a 486DX-50, VLB video (#9 Vision 330) and VLB controller (PTI-255W). Bear in mind that both of these VLB cards specifically state that they are rated for bus speeds up to 50 mHz.
 
Back when the AMD 5x86 chips were new I did hear about a 150 MHz OEM version being officially released, but only in small numbers.

It was also widely rumored that the ADW, ADY, and ADZ temperature ratings of the 5x86 chip were originally intended to be 133, 150, and 160 MHz versions, respectively, but AMD capped the advertised speed to 133 MHz to prevent it from competing with the K5 chip, which was their underperforming Pentium competitor that was soon replaced with the K6.

I have a 5x86-133 that easily oc's to 150 MHz.
 
40Mhz is not the problem, 50mhz is on VLB. You do understand that the VLB bus sits on the CPU bus and has to deal with traffic from cache and RAM at the same time as card transfers. Everybody dropped VLB because of the bottlenecks and went to PCI which was its own BUS not interfering with FSB of the CPU.

VLB wasn't dropped because of usability issues, it was dropped because the Pentium was introduced. And despite its shortcomings, it was far faster than PCI when it worked properly at 40 MHz and especially 50 MHz. Both were parallel drop buses, but VLB at most had three slots to divy bandwidth between, while PCI had up to 7 slots on consumer boards and even more on server boards.

The VLB bus is literally the 486 bus, protocol and all. When the Pentium was introduced, it was a completely different architecture that could not be easily interfaced. Pentium motherboards which had VLB slots used complicated bridge chips that were expensive and temperamental.

Not true. PCI supports a high speed mode of 66 MHz but it wasn't widely implemented. All cards have to indicate support in the config space, have to run at 3.3V I/O, and it's still up to the host controller to adjust clock. The 66 MHz speed option was introduced in 1995 with the 2.1 spec - after most early machines with PCI were on the market.

66 MHz PCI was very common in servers and some Apple machines (B&W G3.) PCI-X officially went up to 533 MHz, but hardly anyone used it since PCIe had come along and was faster and simpler to implement. The fastest PCI cards I've used were rated for 100/133 MHz operation in a PCI-X slot, and they were screamin' fast.

I have heard of a 5x86 running 50MHz VLB with the x4 multiplier at 200MHz, but I have never tried it. I only have one ADZ 5x86, and the last I tried it it would only run up to 160MHz (40x4) and that was on a late model FIC PCI-ISA only board. It should run at 150MHz (50x3) without trouble, though, if the motherboard chipset and cache are up to it.

Back when I was in community college, I had a professor with a Kingston Turbochip on a generic ISA board that ran at 200 MHz. I was stunned it even booted and more so when it was determined it was stable enough to run Windows 95 on. We never tested any games on it, but it ran any Windows application we ran on it without crashing or locking up.
 
66 MHz PCI was very common in servers and some Apple machines (B&W G3.)

I never said 66 MHz host capable controllers were not common in later model machines. All cards had to indicate they could run at 66 MHz since the PCI clock was singular and provided by the host. As soon as you insert a card that doesn't advertise 66 MHz operation either via the OD mode pin or the config bit, the host controller keeps the bus clock at 33 MHz.

PCI-X officially went up to 533 MHz, but hardly anyone used it since PCIe had come along and was faster and simpler to implement. The fastest PCI cards I've used were rated for 100/133 MHz operation in a PCI-X slot, and they were screamin' fast.

You are very much mistaken. PCI/PCI-X only operated at 33 MHz and 66 MHz. The 533 you are referencing is the MB/s (64 bits * 66.66 MHz / 8 bits per byte). I've designed a *LOT* of PCI cards in my day and even a few host controllers in Verilog and can practically recite the damned spec.
 
...
You are very much mistaken. PCI/PCI-X only operated at 33 MHz and 66 MHz. The 533 you are referencing is the MB/s (64 bits * 66.66 MHz / 8 bits per byte). I've designed a *LOT* of PCI cards in my day and even a few host controllers in Verilog and can practically recite the damned spec.
Well, this is one of the few times that I've seen that you've been mistaken. PCI-X 2.0 introduced 266MHz and 533MHz variants of PCI-X, with 2.1GB/s and 4.2GB/s bandwidths, respectively (see the PCI-SIG FAQ, where the 266MHz and 533MHz speed grades are mentioned). I have several Intel server boards (S5000PSL; see the tech spec at https://www.intel.com/content/dam/support/us/en/documents/motherboards/server/s5000psl/sb/417435_d41763_008_s5000psl_tps_rev_17.pdf) with one 133MHz PCI-X slot (and a 100MHz PCI-X slot), and several Dell Precision workstations had 133MHz PCI-X slots. The 266MHz and 533MHz slots were rare, since there was overlap with PCI-e, and 16-lane PCI-e is far faster.

But that's a bit off-topic.
 
Back
Top