I would made a Australia/Austria Joke but I think only people of these countries find those funny.

😅😉

Our skills back then must have been godlike.

lol yeah 10 FPS = playable back then , now some gamers ree when you dont have 120hz screen

@@spitefulwar Looking at how my son beats me in deathmatch now, yes absolutely. I'm just useless today. I can only frag him by camping...

@@Shmbler My son also plays a mean game of Europa Universalis and Victoria II, I got nothing on him.

@@ThunderClawShocktrix Yeah, the '90s. We played Goldeneye 64 deathmatch at single-digit framerates and liked it.

thanks for your great input! 👍🏻

Maybe, the performance difference between the rapid cad and 486 without L1 cache comes from the different die revision you mentioned? I know for a fact that die revisions and different steppings can have a slight and unexplainable performance difference between the same cpu type.

@@bloeckmoep Most performance vraience between intel revisions that i have seen in socket 3 was mostly down to write back/write through differences. But in this case you could be correct. It could be even as subtle as the 486 socket being further away from the bus interface chip too. Only reason I thought of microcode, is when you look at the UMC Super33, it is a literal carbon copy of the intel SX, but has up to a 30% performance advantage, and that is solely from micro code optimization. Showing they had a lot of variance. Bios may also bee using slightly different sub-timing on the rapidcad and thinking it is actually a 386,

"I would have liked to have seen a Doom test as well" what do you mean? There is Doom on the video and the results are there.

@@ccanaves woops... Im a moron... :( Not sure why I didn't register seeing it. And yeah the results are the same as I was thinking it would be had I saw them the first time I watched it lol..

Coleção Nintendo Dr. Mario por aqui também ! Na época, vi apenas o Rapid CAD 1 apenas.

Nice to read about your experience

*looks at Fusion 360 * More things change... the more they stay the same. Always need more MB's & Mhz.

I suspect that they weren't that popular when new either

You're assuming that the RapidCad chips were from failed dies. That might have been the case, but I doubt it. Knowing Intel, I would suspect they were just intentionally crippled. Remember 486 came out in 1989 and RapidCad came out 3 years later in 1992. Their yields at that time should have been high enough that they weren't dealing with very many failed dies. Intel has a history of intentionally crippling chips to sell them at a lower price point without cannibalizing the flagship line. Sometimes it bites them, as with the Celeron 300 vs. the Pentium II 450.

thank you for visiting my channel. Yes, all benchmarks are always done without any soundcards.

Nah, I'm only 30 and still old hardware is far more interesting of modern hardware :)

definitely yes. 👍🏻

Ja Danke. Grüsse aus Villach 😅

@@CPUGalaxy - Nach der Flucherei im Mainboard-Video wusste ich... Es kann nur ein Villacher sein.

lol, echt wahr? hört man da sogar die stadt raus? 🤣

🤔

so the rapid cad is running stable at 40 MHz. I tested that already 😉

@@CPUGalaxy what about some custom oscillators for MB or soldering in something like Si5351 (just an idea, no idea if it would work)

It's probably physically disconnected inside the package.

no idea. l2 is at the other side of the board and fully populated. 🤔

The two sockets there are for configuring the VLB slots for Master or Slave mode (with appropriately configured PAL16L8 chips installed), so they will be slave mode only with them missing.

thx for the info!

lol. yeah. you are up to date now. If you don’t know the past and history you are not ready for the future. 😉. Thanks for the video hint of non mmx pentiums. good idea.

+Cutting Edge Retro Quake on Pentium 75 is already "playable". In 320x200 of course. My guess is around 20 fps (demo1). I can test it, I have built Pentium 1 (non MMX) system right now, but it's on TX motherboards. I have there 150 Pentium, but as it's QDI motherboard, it can be downclocked to 75 pentuium in BIOS. Abit and QDI had first socket 7 jumperless motherboards. ( www.anandtech.com/show/491/10 ) So I dont have to open case, only enter BIOS. Most Pentium 75's were built on older slower chipsets, like Intel FX. The differance will be up to 10-15% anyway. For sure it will be lower than 30 fps, because I know, that to get to 30 fps in Quake, you need good Pentium 1 system. Not sure if 133 or 166, but something like that. So Pentium 75 will have around 20 fps, it's my guess.

Yeah I'm thinking the same.. Doom ran better than this on my 386SX/25 (fully playable, even had death matches via serial link - just put the graphics mode to low and reduce the screen window slightly, disable sound blaster 16 as this lowered frames on the 386) and it ran perfect full screen on my 486DX/33..out of interest I have been searching for videos on Doom performance on 386 on KZfaq and it runs garbage on most, but I did find a DX40 setup where it runs great

No, it can not be paired. But it can be paired with a 3167 weitek math coprocessors which I will cover in another video.

des freit mi ☺️

Hmm... well, in theory, there's actually nothing stopping you from putting a much more modern CPU architecture on an older motherboard, other than them being too power hungry for older motherboards and not supporting the right RAM. I'm thinking you would hit a bottleneck trying to build a fast new CPU for an older system like this, though. For one thing, the RAM is very slow, and so is the bus speed of the motherboard. Also, those old motherboards were not designed to push a lot of power through the CPU... remember they weren't even designed for CPUs that needed active cooling. So you'd probably need something like a low-power Atom that was geared to work on a slower bus, I'm guessing those old 486 or 386 boards probably top out at around low levels of Pentium II level performance regardless of what CPU you put in them, and even that is pushing it. The really cool thing that you could get with a more modern CPU in an older socket, though, is newer instruction sets... you could definitely benefit from SSE2 or something like that.

Dr. Hardware 3.0

@@CPUGalaxy Thx ! maybe link for download ?

yeah, exactly. this board is amazing.

Yay, hello lad, Telcontar here from the old HWSW! If you need a set of RapidCAD for the test, I have one in my collection. For other visitors: this guy is legendary amongst the Hungarian hardware fans. Just one of his famous tests as an example: he tried out different materials used as thermal compound like jam, potato etc. :)

@@Kedvespatikus Hi again! Just to clarify for everyone, the legendary guy's not me, I'm just a guest author on his page ;)

Or in my opinion, they wre looking to make a bit more money from the 386 line and intentionally crippled the dies. Remember, this is 3 years after the release of the 486, so their yields should have been very high.

😂

I doubt it. After 3 years (1989 release for 486 and 1992 for RapidCad) They should have had very high yields.

CPU and FPU are tightly coupled in 486.

😂👹👍🏻

speaking of such, i'd never heard of either the cyrix 486dlc-40 or 486drx2-66, two more 486-instruction set CPUs with 386 socket compatibility - now i'm curious how these two fare too! @CPUgalaxy do you have either of these also? :D (also, thanks for this video -- love this channel!)

A N I M E N I M E

😅👍🏻

I wanted to write the same thing... To be fair, the results of this analysis won't change almost at all in this case, but it nice to do things "properly" if possible, just for the sake of it. In our case the "proper" thing to do to obtain a final 'performance score' for each system it to pick a reference system (eg, the 386+387 system) and compute the geometric mean of the ratios to the reference system, see e.g. en.wikipedia.org/wiki/Geometric_mean#Application_to_normalized_values These are the values I obtained in the way described above: ===Average Integer performance (3 tests: 3D bench, Dr Hard int, Doom) 386/387: 1.00 (reference system) 486 (no cache): 1.10 rapidCad: 1.14 486: 2.00 ===Average floating-point performance (3 tests: Quake, Dr Hard fp, Francint) 386/387: 1.00 (reference system) 486 (no cache): 1.75 rapidCad: 1.73 486: 2.49 === Average over all 6 tests 386/387: 1.00 (reference system) 486 (no cache): 1.39 rapidCad: 1.41 486: 2.23 In any case, this extremely minor point doesn't detract at all from this extremely interesting video! I've been wondering all my life about those rapidCad CPU! I remember seeing them in computer catalogues when I was a young boy in the 90s.

@@luckyluckydog123 While you are right, there is no need to normalize values. Because the end result will still have the same relative value. Example: Results from fracint were converted to frames per second (converting to frames per minute is also fine). The beauty of geomean is that units doesn't matter actually. 4 Columns: 386+387, RapidCAD, 486DX-noL1, 486DX Quake 1.5 2.5 2.6 3.9 3D Bench 1.0C 14.6 17.3 16.2 27.2 Dr. Hard Int: 8403 8782 8389 16050 Dr. Hard FPU: 2517 5035 5035 6625 FracInt: 0.00741 0.01149 0.01149 0.01667 DOOM: 6.66 8 7.9 14.92 Geomeans: 16.8455 23.6696 23.3373 37.5536 (The units can be weird) If you now look at how much faster is 486DX, or RapidCad, compared to 386+387, the same values show up: 1.00 1.41 1.39 2.23. Interpretation is also easy. RapidCAD is 1.41x faster than 386+387. So, don't bother with normalization. The reason it works, is because when you divide normalized values, normalization factors cancel out, and doesn't play a role. The geomean is important to use, an example of silly example: System A: Benchmark1: 200 FPS, Benchmark2: 5 FPS System B: Benchmark1: 194 FPS, Benchmark2: 10 FPS Anybody would see, that the system B is faster, and the minor difference in Benchmark 1, is not important, compared to improvement in Benchmark 2. If you use sum: System A: total 205 "FPS", System B: total 204 "FPS". Which would sugest System B is worse, or at best the same as System A. Or that system A, is 0.5% faster. Which is wrong. If you use geomean: System A: 31.6 "FPS" ("score"). System B: 44.0 "FPS" ("score"). Showing clearly the System B is faster, on average 1.4 compared to System A. Which is indeed the truth, because in first benchmark it is 0.97 faster, and in second benchmark is it 2.0 faster, the geomean ("sqrt(0.97*2)") is indeed 1.4. Also, sum or normal average only works with all units are the same. Example: System A: Benchmark1: 12000 FPM, Benchmark2: 5 FPS System B: Benchmark1: 11640 FPM, Benchmark2: 10 FPS Where, I used frames per minute for the first benchmark. If you use sum: System A: total 12005 "units unknown", System B: total 11650 "units unknown". Again suggesting system A is even faster than before. That it is 3% faster (compared to previous 0.5%). Note that we just changed units! Not values. Geomean still works: System A: 244.9 "score", System B: 341 "score". Not that If you divide these values, you still get 1.4. Which is correct. Note, that Dr. Hard, FracInt and others use all totally different units. So using just sum is absolutly wrong. Sums and averages do have their place, but only for repetition of the exactly same benchmark. In all other cases a geometric mean must be used.

yes the cryix DLC and DLC co processor was probably the best option for 386 boards

we will see what the better option was. I will do some comparison between Rapidcad overclocked @40 MHz and the Cyrix DLC + Copro 😉

Great idea, waiting for a video!

Austrian. I really want him to say, "get to da choppa!"