BinaryDemon wrote on 2021-11-29, 08:11:

IMO, I would disregard anything where the supporting chipset is custom designed for the cpu or gives significant advantage (MediaGX, Vortex86SX, any of the socket 5 486class cpus) because it would get difficult to separate if performance is cpu based or due to improved memory/bus latencies.

Then we may as well limit ourselves to intel and clones. does 486 mean i486 or conforming to 486 instruction set compatibility? Latter I would say.

There certainly is a lot of overlap between generations and opinions on what chip goes where. We've definitely discussed this a lot in this thread and others. Some think super-scalar is necessary, while those want to see features like branch prediction, 64-bit data bus, out of order execution, or support the complete next gen instruction sets. For my needs, "486" is whatever non-Pentium chip works in a socket 3; that's good enough. Titling the thread "World's Fastest non-Pentium socket 3" doesn't have appealing title.

Another interesting one is the cx6x86. I dug into the Cyrix 6x86 databook and of the CPUID, CMPXCHG8B, RDMSR, RDTSC, WRMSR, and RSM instructions, only CPUID and RSM are mentioned. But, the chip is super-scaler, 64-bit, has branch prediction, etc. So what category would this chip fall into? When I look up the Cyrix 6x86MX, on the other hand, all six of these Pentium instructions are included.

Are you all referring to the original Pentium as a "586" or a "686"? If 586, then PPRO being the "686"?

486 adapted to Slot 1 board. Lets gooooo!

Just for grins, I ran Quake 1.08 Timedemo Demo1 on my Cyrix MediaGX 266MHz overclocked to 300Mhz. The result is 41.6 FPS.

The Serpent Rider wrote on 2021-11-29, 12:32:

486 adapted to Slot 1 board. Lets gooooo!

Would require bus adapter chip. To name one of many considerations, but might be possible…

I think it would make more sense to adapt to socket 5/7

feipoa wrote on 2021-11-29, 08:36:

Are you all referring to the original Pentium as a "586" or a "686"? If 586, then PPRO being the "686"?

This is is how Linux sees things and how I accept it.

feipoa wrote on 2021-11-29, 08:36:

There certainly is a lot of overlap between generations and opinions on what chip goes where. We've definitely discussed this a lot in this thread and others. Some think super-scalar is necessary, while those want to see features like branch prediction, 64-bit data bus, out of order execution, or support the complete next gen instruction sets. For my needs, "486" is whatever non-Pentium chip works in a socket 3; that's good enough. Titling the thread "World's Fastest non-Pentium socket 3" doesn't have appealing title.

IMHO the "non-Pentium" bit isn't any less confusing 😀

I think I have already stated my opinion earlier in this thread.

Most likely the only reason for exclucing Pentium overdrive is that all other candidate chips have just as shitty FPUs as "pure" 486DX parts. Thus there isn't any "unfair" benefit when running Quake.

I would simply allow Pentium overdrive, 4.5 gen or upgrade chips and anything that fits into a native 486 socket (PGA-168/1689, Socket 1/2/3/6). Yes, the PODP might destroy everything else in Quake or more general in FPU heavy stuff, but not necessarily at simple graphics pushing stuff like DOOM or RAM heavy tests. In many tests a higher clocked DX4 or Am 5x86 running at overclocked FSB can beat even a full-blown Pentium at 100 or even 120 MHz. The benefit of Pentium could be reduced by weighting Quake result less in the overall benchmark mix. So if you allow anything, but state that the highest Quake score won't win this, and the weight of Quake framerate is less than 20%, then other chips like Cyrix 586 or Am5x86 can still compete. I think rather than push the "486" definition, we should define what does it mean to be "fastest".

So I installed Win95C on my 5x86 system (it flies) and thought I'd run CPU-Z Benchmark on it. Surprisingly, it seems to be roughly on par with a Pentium 90! Anyone else tried this benchmark? Is this a good score?

I had this setup a long time ago, wish I'd kept it...

- BIOSTAR MB8433UUD
- AMD 5x86 133 overclocked to 160MHz
- 256K L2 Cache
- 64MB of EDO RAM
- 80GB ATA-100 drive with dDO
- 54x CD-RW
- Some PCI 3D Card, I think it might have been an ATI Rage II 2MB

Mild stability issues due to a bad cooling, I nipped it in the kibosh by putting a decent cooler for a much newer CPU on it at one point. After that, I was hammering the crap out of it with things designed for later CPU. I wish I had owned Quake or The Sims at that time to try it out on there. Might have been a decent experience.

The Sims on 486 might worth a try indeed.

The Serpent Rider wrote on 2021-12-09, 18:33:

The Sims on 486 might worth a try indeed.

It's something I do intend to give a shot on my current VLB rig - now rigs as I just got a second one this afternoon from a co-worker (+ A super socket 7 system), I'm going to roll back to the initial release though (minus expansion packs and all that) when I try it again. I tried the full package, IIRC, the installer crashed.

The machine I have right now is a VERY fast 486, I'm not saying its the fastest ever, but it's pretty shocking how far this one has gone - FIC 486-PVT Socket 3 board, AMD 486 DX4-100 SV8 with WriteBack, 64MB of RAM on 2 modules out of 4, 512K L2 Cache, S3 805 w/ 2MB on VLB, PTI-255W on VLB, and a SB AWE64. Attached to the PTI-255W is a 5.25" mobile rack and in one caddy I have a IDE to SATA connector - Kingwin ADP-06, and currently a 160GB SATA HDD attached to it. Windows 2000 SP4 boots in less than a minute on that setup IIRC. On that machine.....

- Quake runs nicely, 95 and DOS
- Doom screams, it ate my NEC Ready 9522's lunch during a network game, and that's a 100MHz Pentium with similar specs
- Duke Nukem 3D runs perfectly
- Postal runs with the screen size reduced more than well enough to be fun
- Diablo runs great, I do get a little performance loss if the action gets heavy, but surprisingly great for something the manual specifically told me not to install Diablo on

I've considered hacking the startup of Beachhead 2000 to skip CPU Detection and try running that too 😁

Probably the craziest thing I ever ran was in Win2K SP4 I ran some late (within the last 10-15 years) version of Firefox that still worked with Win2K, and while it was slow as Molasses in January, it did load on the 486 and it did run and surfed the web. Once Firefox was up it ran more like RetroZilla TBH. Runners up were running Firefox - the last version that ran on Windows 2000 SP4, which ran like a slug on valium and a bottle of Jack Daniels but it worked, and managing to buffer up a NIrvana Documentary in Opera in Windows For Workgroups 3.11 by copying over the flash plugins from WIndows 7 for Firefox (back when Secure HTTP was not required and YouTube was still flash - I miss those days).

That said, not to spark a debate, I have to wonder a bit about AMD vs. Intel, or if L1 Cache makes that much of a difference because my NEC Versa M/75, despite being 25MHz slower on bus and CPU, and half the VRAM on the 65545 Local Bus graphics, 256K L2 cache on-board, and a hard disk controller that leaves me scratching my head (and is incredibly picky), gets rather close to the desktop rig in performance.

The *new* one I just got is another AMD DX4-100, not sure what spec though.

Hey guys, just wanted to share my excitement with hitting 200MHz “stable” in Windows with an AM5x86! Dropped the voltage on the 4v jumper down to 3.75 with a resistor, as Feipoa mentioned. That got me consistently running 3/4 through Phils Quake timedemo before crashing back to dosbench. Slapped the peltier back on at 5v and I was finishing! Booted into Windows98 and was able to complete timedemos in GLQuake, Quake2, SOTE, play some MotoRacer. This is at 50x4, I’ve not been able to get it stable at 66x3 on this LS486-d board. Still some condensation forming with the peltier at 5v so it’s still impractical but I’m glad I finally made it!

Holy crap! Good job guys!

This would probably be cheating but couldn’t you make a socket 4 to socket 5 adapter use 3.3v and use a powerleap to install a k6 3+ and have jan whip up a bios to recognize it?

The bus signaling I believe is same or similar

depends how much further streching the truth. already its really fastest 586 not 486 then it would be a 686

Glad to hear the parallel resistor helped your situation. I didn't have the right size trimmers, which is why I had to use a parallel resistor for this board.

feipoa wrote on 2021-12-27, 22:19:

Glad to hear the parallel resistor helped your situation. I didn't have the right size trimmers, which is why I had to use a parallel resistor for this board.

Yes def appreciate it! If there is a trimmer you might recommend let me know, thanks!

If I recall, the mounting situation wasn't ideal. The resistors were surface mount and twisting of the screw will rip the pads. Also, there wasn't a suitable spot to glue the trimmer for support.

Well hell, to get 66x3 working all I had to do was drop down to a 32MB stick of ram.

21.9fps in Phils DosBench Quake Timedemo “option c”

Duke3D hitting 60fps on the roof top at 320x200

26.6fps in Chris SVGA 3D Bench

Impressive

Sphere478 wrote on 2021-12-27, 21:15:

Holy crap! Good job guys!

This would probably be cheating but couldn’t you make a socket 4 to socket 5 adapter use 3.3v and use a powerleap to install a k6 3+ and have jan whip up a bios to recognize it?

The bus signaling I believe is same or similar

There was an old evergreen socket 4 -> 7 adapter, typically it came with a p200, PMMX or original k6 but I’m sure we could hack in a k6-2 of some type

@chadti99
I also noticed that Voodoo3 gets us +0.1 fps at 200mhz compared to MGA, but still a bit slower in Doom for example.
What ram you are using to be a problem for 200mhz.