rasz_pl wrote on 2024-11-21, 00:31:

UMC U5S Super40 vs. U5SX-40 + FPU? […]
Show full quote

UMC U5S Super40 vs. U5SX-40 + FPU?

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

it seems UMC decided to shut down its x86 business. I guess being mainly a foundry for third parties, UMC management must have thought it was better to be on good terms with everyone in the industry biz.

Intel sued https://www.proquest.com/docview/209696670?so … rade%20Journals
and forces UMC into settlement in January 1996 https://www.sfgate.com/business/article/off-t … ker-3160833.php

Yes, I know and -in the end- UMC and Intel settled off court, with UMC giving up any project of entering the x86 market. I was just using an 'hyperbole' to describe what is an otherwise sad, but very common, situation where an industry giant enforces its market monopoly on its weaker competitors.

Since the U5 clearly isn't a copy of the Intel 486 (unlike AMD processors), if UMC had stood up in a court, chances are that -after many years and many millions spent- it could have probably won the case. U5 is just a pipelined processor that happens to produce the same results as Intel's ones when fed some binary values on its bus. 😀 Intel did trademark some of its opcodes even in the 808x era, but that trademark, if I remember correctly, was for the name only (i.e. 'mov') rather than what the opcode does, because it would be too broad and vague. Was the 486 bus even covered by some patents? It's quite a generic bus similar to many other of the period (National Semi, for example, produced some 'glue logic' that was adaptable to both Intel, National, Motorola and other processors of the period with ease by just changing the formulas inside a PAL).

What really hurt UMC was the sales ban Intel obtained (were chipsets affected, too?) in the US and several EU markets. UMC back then wasn't as big as it is today and a litigation against a giant like Intel could have scared off other customers too. (like those buying its chipsets). Also, it wasn't an American company and it didn't produce Intel chips as second source like Texas, SGS (now ST) or IBM did. Note how the article you quote speaks of a '486 clone', while the numbers posted here clearly demonstrate that it isn't! 😳 Intel's biggest weapon even back then was the people mindset: "if it runs 486 code and it sits inside a PC, it must be a copy of an Intel processor" 😉

MSxyz wrote on 2024-11-21, 07:00:

Since the U5 clearly isn't a copy of the Intel 486 (unlike AMD processors), if UMC had stood up in a court, chances are that -after many years and many millions spent- it could have probably won the case.

intel used same https://patents.google.com/patent/US4972338 386 TLB patent to sue everyone at the time. Cant implement fast >=386 without implementing this arguably obvious idea unless your implementation is inefficient

Everyone has their own TLB patent nowadays. Arm has one, Apple, all because Intel's 338 was so successful in court.

A UMC U5S-40 resides in my 486 VLB PC and it is PERFECT for playing anything I want to play on it, including Ultima 7. It is an amazin CPU, and with FDOOM I can even play Doom 2 comfortably on it.

rasz_pl wrote on 2024-11-21, 09:41:

MSxyz wrote on 2024-11-21, 07:00:

Since the U5 clearly isn't a copy of the Intel 486 (unlike AMD processors), if UMC had stood up in a court, chances are that -after many years and many millions spent- it could have probably won the case.

intel used same https://patents.google.com/patent/US4972338 386 TLB patent to sue everyone at the time. Cant implement fast >=386 without implementing this arguably obvious idea unless your implementation is inefficient

Everyone has their own TLB patent nowadays. Arm has one, Apple, all because Intel's 338 was so successful in court.

Incredible. I didn't know this detail... However, looking at the patent it seems this method, as you say, is quite obvious and similar to what some external MMUs for early systems did. Wouldn't this patent be considered void under the premise that similar 'prior art' exists and wasn't patented?

Another question is why Cyrix fought back and was actually able to force Intel to settle. I know they used foundries which had a broad license to build Intel CPUs, but that shouldn't matter if you're using specific patented technology found in one particular CPU. Am I wrong?

MSxyz wrote on 2024-11-21, 10:07:

Another question is why Cyrix fought back and was actually able to force Intel to settle. I know they used foundries which had a broad license to build Intel CPUs, but that shouldn't matter if you're using specific patented technology found in one particular CPU. Am I wrong?

Cyrix didnt win. In US everything at this level of money is settled out of court to avoid establishing precedent that can ruin your (Intel) company. Indeed it looked like foundry having licensing agreement was enough to scare Intel.

rasz_pl wrote on 2024-11-21, 10:12:

MSxyz wrote on 2024-11-21, 10:07:

Another question is why Cyrix fought back and was actually able to force Intel to settle. I know they used foundries which had a broad license to build Intel CPUs, but that shouldn't matter if you're using specific patented technology found in one particular CPU. Am I wrong?

Cyrix didnt win. In US everything at this level of money is settled out of court to avoid establishing precedent that can ruin your (Intel) company. Indeed it looked like foundry having licensing agreement was enough to scare Intel.

There was also the fact that, IIRC, the entire x86 license debacle was on the verge of escalating into an anti-monopoly case with government. As a corporation, you don't want that. You can ask Microsoft. So Intel being evil Intel they settled out of court before things got out of whack..

As promised some actual screenshots from the thest with the U5 @ 50 MHz...

PS: Notice how this U5 completely screws the CPU speed detection algorithm of CACHECHK. I wonder what kind of code really runs nearly 18x faster than on an actual 486 50MHz... 😀 Maybe it's just the internal timer that cannot keep up.

MSxyz wrote on 2024-11-23, 08:20:

Notice how this U5 completely screws the CPU speed detection algorithm of CACHECHK.

900MHz, impressive 😁

Is there any particular reason for testing with CACHECHK version 4 instead of 7?

analog_programmer wrote on 2024-11-23, 09:54:

Is there any particular reason for testing with CACHECHK version 4 instead of 7?

With older cpus that version is more reliable

rasz_pl wrote on 2024-11-23, 09:44:

MSxyz wrote on 2024-11-23, 08:20:

Notice how this U5 completely screws the CPU speed detection algorithm of CACHECHK.

900MHz, impressive 😁

On air, without even a heatsink!
Who needs 3nm CPUs? 😁

By pure coincidence, though I guess U5 was on my mind due to this thread, mine turned up, was looking for a spare Dallas to have modded ready to deploy, then saw stray CPU at bottom of a box, dead bug mode, "Not for US sale" emblazoned on the bottom and the penny dropped, THAT'S MY U5SX!!!!!1111oneone.

Sadly got some pins to straighten as a result of it's adventures outside the CPU holding pens, but looks like it will be alright. My eyes must have slid over it in the past thinking it was just another i486SX or something boring.

I have a circa '92 ISA board on the testbench at the moment, will see if it works on that and give it a spin, might not like it since it's 2 years older than the CPU.

BitWrangler wrote on 2024-11-23, 20:07:

I have a circa '92 ISA board on the testbench at the moment, will see if it works on that and give it a spin, might not like it since it's 2 years older than the CPU.

On some boards, the settings are listed as the same as the 486SX therefore I think it will work just fine even if it might not be recognized.

Another win for the most amazing pin straightening tool in the whole word, the Butler GUM flosspick, okay it doesn't do the outside rows splayed outward, you have to nudge those back with a thumbnail. So it went in the board https://theretroweb.com/motherboards/s/acer-i … gxi-model-i433a with no jumpers changed from the intel SX that was in there. Board is at 40Mhz, board had a sticker on it saying "33Mhz only" when I got it, so doubt I should be pushing my luck at 50 on this one. Anyway, booted up fine, BIOS summary thinks it still has a regular 486SX, just been messing around, getting the feel for if it's running right and if anything needs tweaking. So far, LandMark 6.0 thinks it's a 50Mhz DX and scored it at 246 AT Mhz, where it scored the 40Mhz SX at 190 AT Mhz. SpeedSys had it at 20.51, looked like a hair ahead of the DX2/50 mark on the bottom.

No further news tonight, I found this thread and rabbitholed Quake without FPU ... I am suspecting from the results in there that max would be 2 or 3 frames per sec.

Also got a wandering wondering train of thought on what could be a killer app for my U5SX... initial thought was a laptop that's stuck with an intel SX, now a buddy long ago had that kind of machine, socketed desktop chip, but I don't got one, got one with a proper SL, with copro, and two fixer uppers, one with a 486SLC2 and one with a DX4-75 so those ain't real candidates. I think the DX4-75 one might be beyond fixing. Then I start thinking about using it in the console case maybe, because it's so low heat, console case bottom of page here (relative links are off in that thread) What retro activity did you get up to today? So if I could get low power CPU in there I might not need many holes in it. I think I'd want to be running some 8 bit emulators on that though, so would have to check if there's a selection that run nice on this CPU... but then I am also thinking about half a dozen other fillings for that thing, 686L, K6-2E, etc etc

BitWrangler wrote on 2024-11-24, 05:06:

No further news tonight, I found this thread and rabbitholed Quake without FPU ... I am suspecting from the results in there that max would be 2 or 3 frames per sec.

Quake scores 6-6.5 fps on a 486DX50 so around 3 fps sounds realistic.
I'm not aware of anyone that has tried to modify the Quake source code to run on fixed point math (it has been done for the Saturn, the Game Body Advance and even for the Atari Falcon, although the first two machines can do hardware assisted texturing or -at least- deform a pixel tile to create the impression of 3D). Btw, the U5 is supposed to have a very fast integer division instruction, so it should do quite well with the primitive 3D/raycasting engines of early 90s.

With Wolf3D, my U5 at 50Mhz scores 535 Ticks, or 93.1 fps. For comparison a 486DX2-80 running on the same board scores 527 Ticks or 94.5 fps. With DOOM, however, while much faster than a 486DX-50, it lags bit behind a 486DX2-66.

MSxyz wrote on 2024-11-24, 09:14:

With Wolf3D, my U5 at 50Mhz scores 535 Ticks, or 93.1 fps. For comparison a 486DX2-80 running on the same board scores 527 Ticks or 94.5 fps.

How do you use Wolf3D for FPS tests?

A couple of years ago i did some Socket 3 benchmarks with a VLB system and 64Bit S3 Vision graphics card for DOS and Windows. Sadly i forgot an AmDX40. But related to the iDX50, the U5S33 should be on par with a AmDX40. The CxDX40 is clearly outperformed by the U5S.

The PCPBench seams to make use of the FPU, Doom clearly doesn‘t.

analog_programmer wrote on 2024-11-24, 09:59:

MSxyz wrote on 2024-11-24, 09:14:

With Wolf3D, my U5 at 50Mhz scores 535 Ticks, or 93.1 fps. For comparison a 486DX2-80 running on the same board scores 527 Ticks or 94.5 fps.

How do you use Wolf3D for FPS tests?

I'm using the modded version, which can run also on a 286. I think it was part of Phil's Benchmark Pack although it got removed in the latest revisions.

https://www.philscomputerlab.com/dos-benchmark-pack.html

MSxyz wrote on 2024-11-24, 10:58:

I'm using the modded version, which can run also on a 286. I think it was part of Phil's Benchmark Pack although it got removed in the latest revisions.

https://www.philscomputerlab.com/dos-benchmark-pack.html

V 1.3 JAN 2017
- Removed Wolfenstein 3D
Got too many support questions and users confused

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.