One curious thing I forgot to mention, if I de-turbo the board with the i486SX at 40 while in LandMark, it drops score from 190 to 11.5, if I deturbo the U5SX at 40mhz, it drops from 246 to 78 I think it was. Expectation would be a max of 20, probably more like 16ish. So it would appear that some other thing isn't happening, like the L1 getting turned off. IDK if it's just this board/BIOS can't do it, or whether the U5SX can't disable it's L1. When I tried running cachechk on the i486 deturboed the other week, it crashed, so didn't have anything I did previously with the i486 to definitively say whether any caches were off or not. I believe the board sets the clock to 14.318/2 in deturbo. Anyway, so instead of running like a TurboXT/Slow286 on deturbo, the U5SX seems to run more like a 386sx25 or so, that's a bit of a wild guess off one result though.

Thanks for the explain about the W3D sitch, I was looking at my more recent pack and wondering why I couldn't find it.

analog_programmer wrote on 2024-11-24, 12:38:

So, Phill removed this Wolf3D FPS test from his bench-package back in 2017, probably due to game copyright issues for the game. And I can't find any early revisions of his DOSBench pack. But I just found compiled WOLF_286.EXE here. Unfortunately it doesn't work with shareware version of Wolf3D.

I'm pretty sure the Wolf3D 286 version I have is using the shareware files. I need to check. In case it's the shareware version, I can post the files here. Is there a forum rule against it?

MSxyz wrote on 2024-11-24, 15:47:

I'm pretty sure the Wolf3D 286 version I have is using the shareware files. I need to check.

Nice.

MSxyz wrote on 2024-11-24, 15:47:

In case it's the shareware version, I can post the files here. Is there a forum rule against it?

Only freeware (including shareware) is tolerated. Read the forum rules.

analog_programmer wrote on 2024-11-24, 18:28:

MSxyz wrote on 2024-11-24, 15:47:

Only freeware (including shareware) is tolerated. Read the forum rules.

Good, I asked because some forums frown upon posting any kind of software, just to be on the safe side...

At any rate, I won't be able to post it anyway since it's the full version game. I had the wrong impression because the zipped achieve is only 1.1MB.

I had a few, mostly video results for my Intel 486SX at 40Mhz in the same motherboard, with the fastest ISA card I have, Orchid Kelvin 64 with GD-5434 1MB. ISA at 13.3Mhz. BIOS shadows on...

Doom18s in classic bench demo, one level of border: 4,567 Realtics.
Snooper 4.13: CPU: 110 Vid: 63,330 cps
LM6.0: CPU:189 Vid: 4428.11
3Dbench 1.0: 25.6 fps

Then I did them on the U5SX in same config with Kelvin 64...

Doom18s in classic bench demo, one level of border: 4,238 Realtics.
Snooper 4.13: CPU: 160 Vid: 84,103 cps
LM6.0: CPU:245.71 Vid: 4428
3Dbench 1.0: 29.4 fps

Then in the same config, with Phil mode benchmarks.

Doom19s, fullscreen max, 4,919 realtics
PCPlayer 6.6
Speedsys 21.02 Mem:18.49 MB/sec (Saw it as UMC UMC UMC 215Mhz)
CacheChk saw it as 486 @ 179Mhz.
TopBench: "138" IDK if that's ranking or derived score but it was among dx2/66 when run from prompt.

Then two opinions from SysInfo, version 8 that shows P66 said 111.7, the version in phils old 386 pack, same as in this thread? Has DX33 as highest, that one said 111.8 so probably comparable.

Then sanity check, Doom19s from Phil's pack run with classic one level border, 4,231, within run to run variation of the 1.8s version I was using... Which is much easier to see on Trident cards BTW, 1.9 is hella dark on them.

Interestingly, when I had the Trident 8900D in before I swapped to 5434, it managed a 31 fps on 3DBench, that was with Hidden Refresh on, which Trident cards seem very sensitive to, 20-30% boost if it works on your board/RAM, but in Doom18s that was getting 4,580ish. GD-5434 seems insensitive to it, though it may put a few percent on system performance in some situations.

Anyway, this board/chipset does not seem to hamper the CPU performance any, results from same CPU benches as others at 50 seem to scale right given +25% though of course it gets it's graphics capped by the ISA bus.

Edit: sorry I am lazy and forget to convert the Doom FPS, the i486sx40 did 16.35 best in classic mode, while the U5sx did 17.65 about 8% gain. Phil mode score was 15.2 fps

EditII: So I got cachechk to run on U5 when deturboed, it does turn off L1, L2 is still working but very less effectively. Sysinfo scores it at 7.4 deturboed, and 286/8 is at 4, so guess of 386/25 might be a bit high. However, it does 3DBench at 4.5fps .. think that's more 386ey... dunno why landmark so high, favorable instruction mix maybe.

BitWrangler wrote on 2024-11-24, 22:45:

Then two opinions from SysInfo, version 8 that shows P66 said 111.7, the version in phils old 386 pack, same as in this thread? Has DX33 as highest, that one said 111.8 so probably comparable.

Sysinfo 7 and 8 usually report the same numbers for me. 111 points sounds about right for a U5 @ 40Mhz.

Here are some results of the same PC as in my previous post (Soyo 025P2 mobo + ET4000W32 vlb) but with the U5SX @ 40MHz

Wolf3D (286 ver.) : 611 Ticks - 81.5 FPS
Doom (screenblocks=10) : 3255 RealTicks - 22.9 FPS

Thanks for those.

Guess I'm still missing Doom potential from lack of local bus graphics, but it dawned on me that I had seen a better result on a TVGA 8900D card on this board, and that although in theory the GD-5434 is better, it does not seem to like this board much, so stuck the best 8900D in, and turned on all optimisations in BIOS. I guess I was leaving it in the config I had i486sx40 results for rather than seeing what the best it could run with U5 at 40 was.

Anyway, picked up minor improvements, a whole 'nother frame per sec in 3Dbench at 30.3 fps, and same in Doom 1.9s, with 15.97 fps in Phil mode, and 18.41 in classic mode. PC Player got an extra tick at 6.7 though for some reason it looked smoother than at 6.6 the other day, odd. Sysinfo picked up the extra decimal to match the older version score of 111.8 ... though difference there could be system clocks just either side of 40 or something silly. TOPBench score jumped to 142 overall.

So gfx definitely the weak spot, I think I might drop it to 33Mhz and try the 33/2 ISA bus speed, 40/2 is too big a leap, then if 33/2 works, and it might bring doom scores even, despite slower CPU, then I can try a 50mhz test at low ISA speed, and if no board side crashy, go to 50/3, since that would be so close to 33/2 that if that works the other should.

Though maybe it's a bit pointless trying to optimise my test mule and I should break out the UMC/VLB/EDO board.

Been trying to get one of these CPU's myself for an affordable price, but on eBay that's a little too difficult.

There's a video by vswitchzero in where he overclocks his UMC to 60MHz, and the results are very good. Faster than an Intel 486 DX-2 66Mhz. For 60Mhz though a PCI mobo is better, since many included a PCI 2/3 and/or 1/2 dividers. Some PCI Matrox cards can run with very high PCI bus speeds so you might get away with PCI running in sync at 60Mhz with the CPU. I wouldn't use a system like this to run the latest DOS games but would be very flexible with a big majority of games.

I think it's not possible to compare the performance of a dx2 vs dx, even with dividers for the first one the fsb is still double the second. That being said I don't doubt that the UMC cpu would be able to beat a dx-66 since it's always faster clock to clock.
According to the test I did some time ago increasing the clock is going to increase the performance gap.

carlostex wrote on 2024-11-28, 06:11:

Been trying to get one of these CPU's myself for an affordable price, but on eBay that's a little too difficult.

I keep thinking there's a market somewhere that must have got relatively flooded with them, South America maybe, and we're just not seeing them due to language and other barriers.

BitWrangler wrote on 2024-11-28, 15:54:

carlostex wrote on 2024-11-28, 06:11:

Been trying to get one of these CPU's myself for an affordable price, but on eBay that's a little too difficult.

I keep thinking there's a market somewhere that must have got relatively flooded with them, South America maybe, and we're just not seeing them due to language and other barriers.

Eastern Europe..

appiah4 wrote on 2024-11-29, 07:38:

Eastern Europe..

Nope, rather Eastern Asia.

dionb wrote on 2024-11-20, 14:44:

The first game that absolutely required an FPU was Quake, and good luck running that acceptably on any 486 even if you had an FPU.

False - Cyclones specifically requires a FPU. For reasons I cannot legitimately fathom, Skunny: Special Edition also requrires a FPU. Cylindrix has files dating from April 1996, though QTest1 did come out earlier.

Did the original "The Need For Speed" need an FPU? Says DX2/66 min, but notes given on moby emphasise DX 66, like it doesn't work on an SX2.

BitWrangler wrote on 2024-12-16, 05:18:

Did the original "The Need For Speed" need an FPU? Says DX2/66 min, but notes given on moby emphasise DX 66, like it doesn't work on an SX2.

All the sources online I've found only mention the 486DX2 66 . I'm afraid it won't be possible to know without trying.

BitWrangler wrote on 2024-12-16, 05:18:

Did the original "The Need For Speed" need an FPU? Says DX2/66 min, but notes given on moby emphasise DX 66, like it doesn't work on an SX2.

Nope, no FPU required. Very playable on a Ti 486DLC-40 with a WD90c30 Trident TVGA8900CL card and a 10Mhz ISA bus, easily hitting around 15-20 FPS. The box is correct in that it does require 486 instructions.

It did however make me think about how a lot of polygonal racing games started appearing around 1995, and given that those generally want a good framerate so I decided to a bit of further digging as I haven't really documented protected-mode games that much outside of FPS and platformer games and sure enough, the following racing games that were released before Quake requires a FPU:

• Destruction Derby
• Fatal Racing
• Track Attack
• Virtual Karts
• Zone Raiders

While most of them will crash when trying to get to the main menu, Fatal Racing is fairly unique in that it never crashes but all it renders is a solid blue square during the attract mode and gameplay.

Big Red Racing, Hi-Octane, Slipstream 5000, Screamer and IndyCar Racing 2 don't require it at all - that was expected from IndyCar 2 using the same engine as NASCAR Racing, though the others were a bit surprising to see.

Did some even further digging in general, going through my GOG library for games around 1995-96 that would seem like contenders that I hadn't documented yet, and sure enough Magic Carpet 2 also requires a FPU. For some reason my install of Magic Carpet 1 is broken and won't run, probably FAT data corruption at some point. Finally did some poking around a few sim games from 1995, and the only one that required a FPU is Star Rangers, an obscure space sim from i-Magic.

I also only focused on the part of my spreadsheet that was listing "386+387" as a CPU requirement earlier and had dates listed - Hypercycles requires a 486DX to run and that was released in 1995. An obscure shareware shmup called Day of Pigs uniquely has a minimum of a 286+287 with files dated January 1996. I also didn't list a date for a DOS port of a German Atari ST pinball game called Starball, that has files dated around January 1996 - that also requires a 386+387 at minimum. The latter two are more contenders for "why does this even need a FPU requirement?".

So that's what, 13 games that require a FPU that were all released before Quake? Seems like Quake is anything but the "absolutely first" game that required a FPU - it's not even the first FPS game that requires a FPU, more like the fourth, technically five if you count Hypercycles. Cyclones is still the only game from 1994 that I've found that requires a FPU at this point.

All in all, I hit a good success rate for finding stuff yesterday evening. Bit surprised at the complete lack of FPU requirement for any of the flight sims I tried, but I didn't check out Flight Unlimited.

UPDATE: Didn't pay attention to what video card I was using, I blame the multiple WD90C30s I have sitting around on my bench which I've been meaning to test the RAMDACs for 😜

Good info thanks, some there to make a note to try on my BL3 @ 100 plus 33mhz 387 to find out how badly they need a FPU, since BL3 is around dx2 66 performance otherwise. If that keeps up then I guess there would be hope for 387 emulator with fast U5s.

I seem to recall a racing game from around then having a really annoying pentium detection routine, when it didn't really need a pentium, just needed high performance, and it would fail on 6x86 and K5.... then some time later there was a patch and it ran on 5x86 and DX4 pretty good. If I can figure out what that was, it might be one to try, as I think it actually went damn fast on "weak" FPU non-intels, so was probably mostly integer.

Im surprised by Destruction Derby, its doing fine with Fixed point math on Playstation.

rasz_pl wrote on 2024-12-17, 00:46:

Im surprised by Destruction Derby, its doing fine with Fixed point math on Playstation.

I decided to double-check just in case that I made a mistake as my testing path went 386DX -> 386DX+387 -> 486DLC -> 486DLC+387 and I stopped at 386DX+387 with DD1 - it definitely requires a FPU even with 486 instructions.

Quick video:
https://www.youtube.com/watch?v=f_BF3MBnSAg

Tangentially related, but I do recall back in early-2022 or late-2021 that I was testing Forsaken on a AMD 5x86-133 (or maybe 160Mhz, I can't seem to find my brief notes) with a PCI 32MB Savage4 at 640x480x16b and getting somewhere around 20-23 FPS. Seeing as that's another game that's also on the PS1 which is extremely well-optimised, it would be interesting to make a direct comparison between a DX2-80 and an AMD SX2-66 running at 40Mhz FSB, which my chip can easily handle. I do have a UMC U5S-40 that easily hits 50Mhz, but I don't have a mobo that supports 60/66Mhz FSB.

MSxyz wrote on 2024-11-20, 18:47:

To add some gossip to the rumour mill, the U5SD supposedly is a CPU fully compatible with the 486DX, but the built in FPU turned out to be defective, so it was deactivated.

That would have been amazing, because the UMC U5S isn't even fully Intel 486SX compatible.
For example: The Intel 486SX can run Windows 98SE, even the Cyrix 486DLC (a 386 replacement!) can run Windows 98SE, but the UMC U5S can not.