PCI launched in 1992 (420TX chipset, but that was rather a proof of concept), started getting traction in 1994 when second-gen Pentium started eating into the mainstream. Got mine in 1995 (in a Pentium 75 system). There was only a handful of non-PCI Pentium systems (if any). So you can say the PCI became popular at the same time as Pentium.

The PCI was really a Pentium thing. Late 486 motherboards got it too but it was more an afterthought.

my first PCI board was an Asus Socket 7 with a Pentium 90 in 1996 if I recall correctly, a much needed upgrade from our previous 386DX40, skipped 486 altogether back then.

mpe wrote on 2020-04-05, 00:04:

PCI launched in 1992 (420TX chipset, but that was rather a proof of concept), started getting traction in 1994 when second-gen Pentium started eating into the mainstream. Got mine in 1995 (in a Pentium 75 system). There was only a handful of non-PCI Pentium systems (if any). So you can say the PCI became popular at the same time as Pentium.

Fully agree with the general sentiment, but there certainly were non-PCI Pentium systems. There was an OPTi VLB+ISA only chipset and Compaq sold early Pentiums linked up to their Triflex EISA+ISA chipset. The latter were probably a lot more common than the former.

The PCI was really a Pentium thing. Late 486 motherboards got it too but it was more an afterthought.

I'd dispute that. 486 remained common in the low end as late as 1997, and basically all 486 from 1995 onwards used PCI as the primary bus. Enthousiasts/gamers had abandoned the platform long before, but it was a valid general purpose/office type option until the integrated VGA chipsets dropped the price of Pentium so much that 486 no longer made sense even from a price perspective.

Yep... The first PCI machine I had was a Cyrix 5x86/100 I got at probably the very end of 1996. It was based on a PC Chips M919, which also had a VLB slot, but I never had any VLB cards for it. I think I had a Cirrus Logic PCI card in that system.

Intel wasn't a chipsets provider back then when they launched the 1st generation Pentium, the 60MHz and 66MHz versions. The chipsets market for PC back then was dominated by VLSI, OPTi, SiS and ALi. They were traditional chipsets suppliers for 386/486. I don't remember if VIA was there. As the PC prices continued to decline, those chipsets vendors were reluctant to further platform innovations into PCI, they wished to hold on to VLB for some more time as the standard just got stabilized for a year or two. Intel dislikes VLB but it couldn't do anything to push the market. Well, as a chipsets provider at the time, it was a sound business plan because new innovation always requires capital investments in R&D. They did not enjoy the lucrative margins of selling CPUs. They probably knew very well that Intel had begun its investment in chipsets designs since Intel already launched the Saturn and Neptune PCIsets for 486 and Pentium, but not very successful in the market.

The launch of 2nd generation Pentium, the 75MHz, 90MHz and 100MHz versions radically changed the landscape of chipsets market. The Intel 430FX Triton PCIsets was extremely successful. It swept the entire market to Intel's favor and Intel became the largest chipsets supplier for PC and regained the controls on how the Intel platforms would evolve for the foreseeable future. This was the time PCI became very popular, an architecturally correct PCIsets from Intel, a contrast to SiS/OPTi offerings that trending towards a more legacy approach. The OPTi 586VLB/PCI combo chipsets was notorious of its poor PCI performance and that further fueled the adoption of Intel PCIsets.

My first PC was a Compaq Presario 9232 (120MHz Pentium with PCI) -still have it and it works great. It seems like the Pentium era really started taking advantage of it.

My first one was in 1995 intel triton motherboard and a pci S3 video card when I built a new pentium 90Mhz system back then.

My first PCI system was a last gen 486 AT board I still have in use that I ran a 486/133 at 160. It lasted me until I upgraded to a M-Tech 430HX based Pentium 133 to play Quake (still have that one also).
To be honest I don't recall what PCI video card I had in that system but I purchased a Matrox Millenium for the P133.

Yeah first PCI board was a Pentium socket 4- 60mhz board which performed same or worse than my 486 DX100 VLB at the time and was very disappointed iirc. Know I did not get any other pentium boards until the socket 5 came out a year later. Did get a few 486 PCI boards about same time as socket 5 but they were for upgrading the 486 board. As for the models of the boards am sorta brain cramped but know the 486 PCI was an FIC and the socket 5 was Asus.

I went from a ISA+VLB AMD 5x86-133 (running at 160) to a SS7 board with a K6-233.. I think. Pretty sure that was the clock speed at least.

Pentium II-233 Compaq machine in 1997 would've been my first system with PCI.

My first PC had PCI and that was a 486dx4 100.

kjliew wrote on 2020-04-05, 02:09:

[...]

The launch of 2nd generation Pentium, the 75MHz, 90MHz and 100MHz versions radically changed the landscape of chipsets market. The Intel 430FX Triton PCIsets was extremely successful. It swept the entire market to Intel's favor and Intel became the largest chipsets supplier for PC and regained the controls on how the Intel platforms would evolve for the foreseeable future. This was the time PCI became very popular, an architecturally correct PCIsets from Intel, a contrast to SiS/OPTi offerings that trending towards a more legacy approach. The OPTi 586VLB/PCI combo chipsets was notorious of its poor PCI performance and that further fueled the adoption of Intel PCIsets.

Not quite. The Socket 5 Pentium CPUs were released over a year before i430FX. It was the latter that really made the platform compelling, combined with the fact that pricing of Pentium CPUs was less stratospheric in 1995 than it had been.

It wasn't i430FX (or indeed the rather lackluster i430NX that So5 launched with that pressed the market to PCI. The traditional chipset vendors had already made that move by 1994. You mention SiS as offering a 'legacy approach', but they had full feature parity with Intel by 1994 and the performance of their 501 was competitive with i430LX/NX. ALi's first Pentium offering actually outperformed both by a significant margin and was PCI only Via's first Pentium chipset came relatively late (1995) but was also PCI. The big driver was cost. VLB was inherently very expensive, with huge cards and worse, huge slots on the motherboard that also constrained design (there's a reason all those VLB boards looked so similar). This applied on the 486 as much as it did on the Pentium.

Non-PCI Pentium chipsets weren't easier to develop than PCI, they were much harder. Boards with VLB were also inherently more expensive due to the huge PCB real estate it demanded. OPTi went down that path on a gamble that upgrades would be a bigger thing and people would not want to throw away their expensive VLB VGA and I/O cards. Unfortunately the market didn't go that way, basically no one wanting a Pentium was keeping their legacy I/O, and just as with the contemporary VIP-IO 486 boards, the complexity translated to slower performance and less stability. By the time OPTi abandoned that gamble and came with the decent Viper-M chipset, it was too little, too late.

Engineering PCI wasn't really a big challenge - 32b, 33MHz and clear, sensible design was relatively straighforward, as evidenced by the fact pretty much every 486 vendor had PCI implementations and most of them worked decently by the second iteration. even though VLB was technically perfectly viable on that platform. The huge challenge at the time was engineering for the Pentium bus. It doubled bus width to 64b, which increased chances of skew on all the buses, it bumped speed up to 66MHz at a time when the 486 world had just given up on 50MHz because too challenging to implement reliably. I'd say it's a miracle so many of the 1993/1994 Pentium chipsets worked reliably at all.
A lot of the problems with OPTi's early Pentium chipsets came from that challenge. UMC's example is telling: they were one of the biggest if not the biggest 486 chipset vendor, with solid, reliable, fast solutions. They only ever made one Pentium chipset and it was a disaster. Why? Because they basically just shoehorned a 486 design, complete with 32b cache and memory access, onto (half of) the Pentium bus. The result was the poorest performance of any Pentium chipset ever, even worse than the slowest integrated VGA chipsets. Avoid the UM890C chipset unless you want to see how brain-dead a design could be. UMC undoubtedly chose this path to avoid R&D investment. OPTi hit the same problems with the Python despite spending heavily - juggling VLB and PCI performance was bad enough at the best of times (on a 486 that interfaced natively with VLB), trying to get something that talked natively to a 32b CPU bus talking via glue logic to a 64b CPU bus was non-trivial and inherently increased latencies. This was also a failure of marketing over technical common sense. Shame, as OPTi had pulled it off a generation earlier, with chipsets like the 295X that supported both 386 and 486 and could offer VLB to 386 CPUs without sacrificing performance significantly vs competing less flexible chipsets.

UMC deserves some love. I actually used UM8891F based Socket 5 motherboard (Shuttle HOT-539) and it wasn't that bad. It was a 64 bit designed for 3.3V Pentiums up to 200 MHz (non-MMX). It even had a dual-bank 64bit L2 async cache in 16 SRAM chips, which was rather uncommon in Pentium era. SIMMs had to be installed in pairs indicating 64bit data path. Couldn't compete with Triton, but roughly comparable to SiS 501 or 430NX.

I believe the crude 486 scaled 32 designs were early pre-Viper Opti chipsets. Like OPTi 571/572 or OPTi 693/696. Or anything with VLB slots.

UMC actually had some success in early Pentium laptops as Intel's design did not scale well in that space until 430MX. Many early Pentium laptops used UMC chipsets as late as in 1996.

dionb wrote on 2020-04-05, 10:12:

It wasn't i430FX (or indeed the rather lackluster i430NX that So5 launched with that pressed the market to PCI. The traditional chipset vendors had already made that move by 1994. You mention SiS as offering a 'legacy approach', but they had full feature parity with Intel by 1994 and the performance of their 501 was competitive with i430LX/NX. ALi's first Pentium offering actually outperformed both by a significant margin and was PCI only Via's first Pentium chipset came relatively late (1995) but was also PCI.

It was, indeed. As I had already said earlier, Intel disliked VLB, drafted and promoted PCI, but traditional chipsets vendors were not interested. The earlier Intel chipsets (Saturn/Aries for 486, Mercury/Neptune for 586) were all PCIsets, but they typically landed in high-end and expensive systems. 1st generation Pentiums were expensive, too, and VLB simply dominated 486 market and PCI peripherals were not common. SiS was a more submissive chipsets vendor compared to UMC/OPTi, but SiS501 was indeed a "legacy approach" with 3-chip design and had both the Cache/Memory controller and PCI LocalBus Data controller hooking on the FSB compared to Intel's single PCI host bridge designs. The i430FX Triton PCIsets marked an unprecedented milestone, a high-integration 2-chip PCIsets and price that reached mainstream, together with cheaper and higher volume 2nd generation Pentiums, thanks to manufacturing process improvements. Well, SiS559x and VIA VP2/3 PCIsets would eventually catch up but it was too late. Intel's brand name and marketing were hard to penetrate.

There were reasons why Intel preferred a PCI host bridge design, it is a simple single CPU master and single endpoint hence simplifies logic required for arbitration. However, more importantly Intel was laying a strategic groundwork to regain the controls on CPU FSB. With the subsequent Pentium Pro/II, Intel P6 FSB became propriety and required licensing, a move that forced financially strapped AMD to invest in clean-room CPU alternatives, AMD almost went bankrupt if not for unnamed OEMs at the time who, at the critical moment, infused additional funds into AMD to keep them afloat to prevent Intel from total domination. As a side joke as I would say 😁, Intel's target was always AMD, but for those traditional chipsets vendors who weren't cooperating, it wouldn't mind the strayed fire that burned them.

dionb wrote on 2020-04-05, 10:12:

Non-PCI Pentium chipsets weren't easier to develop than PCI, they were much harder. Boards with VLB were also inherently more expensive due to the huge PCB real estate it demanded. OPTi went down that path on a gamble that upgrades would be a bigger thing and people would not want to throw away their expensive VLB VGA and I/O cards.

You are right, VLB is a more complicated design in term of trace length and PCB real estate. That was inevitable for a design that tried to retrofit into existing ISA bus architecture. The industry needed a solution for high performance graphics, propriety designs (OPTi, CHiPS etc) were making the situation a big mess. AIB vendors were reluctant to invest in a locked in ecosystem, so VESA was formed to take the lead to standardize without Intel blessings. The traditional chipsets vendors had a good multiple years of head start before the rise of PCI, so most of the designs quirks would have been ironed out and they wished to capitalize on it instead of listening to Intel. PCI 1.0 was in draft, it simplified AIB designs but at zero market penetration. So Intel's mission was to create the market for PCI chipsets, they couldn't do it with 486 and 1st-gen Pentiums were too expensive to reach mainstream. It was the Intel 430FX Triton PCIsets that created the critical mass for PCI to be successful.

I first saw PCI in action between December of 1994 and February of 1995. That was a Pentium or something like that. The first computer that I owned with PCI, was a K6-II-500 and Gigabyte GA5AX board, that I bought around 1998 or something like that. Until then, I only had ISA and VLB slots.

kjliew wrote on 2020-04-06, 01:36:

[...]
It was, indeed. As I had already said earlier, Intel disliked VLB, drafted and promoted PCI, but traditional chipsets vendors were not interested. The earlier Intel chipsets (Saturn/Aries for 486, Mercury/Neptune for 586) were all PCIsets, but they typically landed in high-end and expensive systems. 1st generation Pentiums were expensive, too, and VLB simply dominated 486 market and PCI peripherals were not common. SiS was a more submissive chipsets vendor compared to UMC/OPTi, but SiS501 was indeed a "legacy approach" with 3-chip design and had both the Cache/Memory controller and PCI LocalBus Data controller hooking on the FSB compared to Intel's single PCI host bridge designs. The i430FX Triton PCIsets marked an unprecedented milestone, a high-integration 2-chip PCIsets and price that reached mainstream, together with cheaper and higher volume 2nd generation Pentiums, thanks to manufacturing process improvements. Well, SiS559x and VIA VP2/3 PCIsets would eventually catch up but it was too late. Intel's brand name and marketing were hard to penetrate.

I'm not disputing the status of the i430FX as a milestone, both due to the huge performance leap and the associated compelling case for a Pentium CPU at last. But is has nothing to do with the uptake of PCI for the simple reason that every single chipset vendor for Pentium CPUs had already switched to PCI buses for both Pentium and 486 motherboards a full year before the release of the i430FX. Some vendors tried to combine VLB and PCI, mainly for 486 but rarely (OPTi...) for Pentium too, but there were no chipset releases from 1994 onwards that did not support PCI. Some boards might not implement it, but the chipsets all did. Of course successful old designs kept being sold for quite some time, but all new development had already shifted to PCI before the i430FX was launched.

You are right, VLB is a more complicated design in term of trace length and PCB real estate. That was inevitable for a design that tried to retrofit into existing ISA bus architecture. The industry needed a solution for high performance graphics, propriety designs (OPTi, CHiPS etc) were making the situation a big mess. AIB vendors were reluctant to invest in a locked in ecosystem, so VESA was formed to take the lead to standardize without Intel blessings. The traditional chipsets vendors had a good multiple years of head start before the rise of PCI, so most of the designs quirks would have been ironed out and they wished to capitalize on it instead of listening to Intel. PCI 1.0 was in draft, it simplified AIB designs but at zero market penetration. So Intel's mission was to create the market for PCI chipsets, they couldn't do it with 486 and 1st-gen Pentiums were too expensive to reach mainstream. It was the Intel 430FX Triton PCIsets that created the critical mass for PCI to be successful.

i430FX certainly created the critical mass for Pentium to be successful - and helped spark the boom in overall PC sales in the mid 1990s. But PCI had already won.

Just take a look at video cards, SCSI adapters, network adapters etc. - the sort of things people used VLB for and later PCI. 1994 was the breakthrough year for all of them. Everyone from Adaptec and 3Com in the high end to bottom-feeders like Trident were shipping PCI cards. I just grabbed a September 1994 edition of Byte and looked at the adverts - 486DX2 systems were being advertised with "PCI Local Bus Graphics" and "PCI Enhanced IDE interface", in fact some even offering that on 486SX systems (Zeos Pantera "Standard with every Zeos Pantera system [...] PCI local bus color graphics"). This was four months before the launch of i430FX. There were certainly still systems being sold without PCI, but PCI itself had already become a selling point, and not just on the hugely expensive and slow early Pentium systems but also on 486, even down to SX in one case.

I first had a PC with PCI in 1997, it was also a full atx board with USB, running with a p133 and 16MB of memory.

dionb wrote on 2020-04-06, 23:08:

But is has nothing to do with the uptake of PCI for the simple reason that every single chipset vendor for Pentium CPUs had already switched to PCI buses for both Pentium and 486 motherboards a full year before the release of the i430FX. Some vendors tried to combine VLB and PCI, mainly for 486 but rarely (OPTi...) for Pentium too, but there were no chipset releases from 1994 onwards that did not support PCI.

PCI 1.0 was available in 1992. It was hard to predict which standard would prevail as traditional chipsets vendors already shipped VLB chipsets. It was also hard to tell if history would have been rewritten had Intel not took the problem into their own hands with the i430FX Triton PCIsets. It became another heated "VHS vs Beta" type of debate, if availability would triumph the "architecturally correct" design on paper. i430FX launched in January 1995, but you could rest be assured that engineering samples would be available much earlier up to a full year for Intel partners to develop the PCI ecosystem. The change of tune from the traditional chipsets vendors towards PCI simply told that Intel was a force to be reckoned with, but too late and too bad for them. The aftermath, UMC and OPTi were wiped out, the more submissive SiS and VIA were rewarded and even granted license to Intel P6 bus. For the glorious years that came after Intel 430FX Triton PCIsets, the Intel CPU attachment rate even reached over 95%. If not for Intel to abandon their own Pentium platforms in favor of the new propriety licensed P6 bus leaving the market for SiS/VIA/ALi to breathe, I doubt if those would survive.

Perhaps it could be difference for the market region but for the region where I lived during mid-90's, 486 PCI was a scarcity, a cursed and typically sold to uninformed 1st-time buyers, bulk local schools/colleges orders or government agencies. It was sold with the idea that one could keep the PCI peripherals in future upgrades. Pentium 75MHz was priced similar in the high-end of 486s and that was Intel's answer to AMD finally started shipping Am486s, even though 486DX4-100 and Am486DX4-120 can be faster than Pentium 75MHz in benchmarks.

dionb wrote on 2020-04-06, 23:08:

Just take a look at video cards, SCSI adapters, network adapters etc. - the sort of things people used VLB for and later PCI. 1994 was the breakthrough year for all of them. Everyone from Adaptec and 3Com in the high end to bottom-feeders like Trident were shipping PCI cards. I just grabbed a September 1994 edition of Byte and looked at the adverts - 486DX2 systems were being advertised with "PCI Local Bus Graphics" and "PCI Enhanced IDE interface", in fact some even offering that on 486SX systems (Zeos Pantera "Standard with every Zeos Pantera system [...] PCI local bus color graphics"). This was four months before the launch of i430FX. There were certainly still systems being sold without PCI, but PCI itself had already become a selling point, and not just on the hugely expensive and slow early Pentium systems but also on 486, even down to SX in one case.

The reason why the AIB market was booming for PCI compared to VLB was simple. Despite VESA efforts to standardize, VLB was simply not as versatile as PCI being a "true local bus" that hooked off the CPU FSB and overloaded the bus in contrast to PCI "intelligent bridge" design. No VLB design can have more that 2 masters, for VLB motherboards that have 3 slots, only 2 can be masters the remaining is slave-only. SCSI and networking controllers vendors were simply left out in the cold shower by VLB and forced to stay high-end with MCA/EISA solution. It would be no surprise that they would welcome and endorse PCI for the opportunity to expand their market into mainstream products, not to mention that the veil of uncertainty had gone for a standard backed by Intel.