aries-mu wrote on 2023-10-02, 14:25:[...] […]
Show full quote
[...]
Wow!
You know what you want!!!
Questions please:
• Why "ISA only" on a 386? Shouldn't a VLB slot help with faster 386s, like DX 33 and DX40 ones? Or at least EISA?
VLB and 386 wasn't really a thing, even if boards exist (and yes, I have a couple, even one with a souped-up 386SX with 16b bus and VLB), anything that would run on VLB would run better on a 486 anyway. Then again. I'm not really that interested in running a 286 or 386 regardless, as there's not much they can do that a 486 couldn't do just as well. There's a case to be made for EISA, but that's pretty niche and makes configuration orders of magnitude more complex.
• What kind and speed of L2 cache for 386s?
Depends on the chipset but probably just asynch SRAM. This only makes sense btw if it doesn't introduce latency.
• What is USB "with legacy input support"?
A translation layer that lets you use USB peripherals that get presented to software as PS/2 so that OSs that don't support USB (i.e. DOS and early Win95) can still be operated by them. Was very common in BIOS around the turn of the millennium.
• What is podp? I googled it up
Pentium OverDrive Processor, a P54 Pentium in an extended Socket 3 form factor with onboard 5V->3.3V voltage regulator. Came in 63MHz (2.5x 25) and 83MHz (2.5x 33MHz) flavours.
• On 486 mobos, if you say "bus from 16 to 66 MHz selectable and stable" then why do you freeze PCI @ 33 MHz? Shouldn't you want it at 66 MHz?
Because the performance gain from out-of-spec PCI is negligible, but it is an infamous cause of instability, particularly data corruption on storage devices attached via PCI. Whereas there are VLB cards that can actually run at 50MHz or above (usually without any other VLB cards and frequently with wait states), PCI dies before 45MHz. So the logic is to use the VLB for a single VGA card for highest clock, but keep storage on PCI for stability.
• 72 pin RAM for 486s: What kind? FPM? EDO? What timing? 70ns? 60ns? less?
FPM vs EDO depends on chipset, but basically EDO offers no tangible benefits for 486, so FPM makes most sense. As for timing, that depends on clock - 66MHz requires 60ns, 70ns is enough for slower systems. Note that these are max speeds, so you can always run fast SIMMs slower. 60ns will always work, so I'd personally choose that.
• No L2 cache for 486s? If yes, what kind and speed?
Well, unless you are Octek, it's all asynch, and again required speeds depend on bus speed. 12ns should always be fast enough. Cachable area is however more interesting, but depends on the chipset. All things being equal I'd like the biggest possible cacheable area, although realistically speaking there's no point in running a 486 with more than 64MB anyway.
• What does "BF0, 1 and 2 hooked up" mean?
Multiplier pins.
Early So5 boards only had BF0, so could only select between 1.5x and 2.0x.
So7 boards had BF0 and BF1, adding 2.5x and 3.0x
Late So7 boards supported the third BF2 pin (used on Tillamook, K6-2 and M2 CPUs), adding 4.0x, 4.5x, 5.0x and 5.5x.
3.5x was obtained on CPUs that supported it by remapping the 1.5x setting to 3.5x
6x was obtained on CPUs that supported it by remapping the 2.0x setting to 6.0x
• What about L2 cache amount, type, and speed on Pentiums?
Again depends on chipset. Realistically we're talking Aladdin V or MVP3 here. In the case of the Aladdin V, you need at least G-revision chipset allowing anything the board can handle to be cached. For MVP3, cacheable area is determined by amount of cache. 2MB gives max area, so board should have 2MB in that case.
Type: PLB, hands down. Older async can't handle the speeds and hardly boosts performance either.
Finally, a comment:
AGP on P1: personally, when it comes to retrocomputing, it's all fuzzy... preferences... time-correctedness... all very vague and emotional. But I do know 1 thing for sure, a clear cutoff criteria. To me (but this is entirely personal), the boundary that separates what I consider retro-computing worth of my attention and post-retro I lose interest in is the AGP bus: pre-AGP, okay. Post-AGP: not interested.
Doesn't AGP feel a little too 'modern' on Pentiums?
That's a very personal consideration. Point is, for me at least, that an 'ultimate' boards should be able support any hardware that could be used on the platform, so in this case any CPU from a Pentium 75 to a K6-3+ 550. AGP and P75 is pretty odd, but a K6-3+ without it is conversely lacking - and K6-2 with Voodoo3 was a common and desirable platform in the day. But yes, it's all about that 'cutoff' and I'd go a bit later for mine - basically anything pre-XP feels vintage to me, and I tended to run newer systems with older operating systems for max speed sensation, so I was still running Windows 98 on my Athlons which most definitely were AGP platforms. But there's nothing wrong with having a different take on it. I would propose however that it's easier to have a board with AGP and not use it, than to have a board without AGP and try to get it anyway...
Also, very late AMD So7 CPUs aside, one of the joys of late So7 boards is pushing the P55C i.e. Pentium MMX to its limits. Intel killed of the P5 line with it for non-technical reasons (they wanted you to buy P2s instead), and the hardware was capable of a lot more than 233MHz. I found it easy to push them to 3.5x 100MHz on an Aladdin V platform and at that spead they easily beat a K6-2 400, and paired very well with early AGP cards.