This is almost certainly why. The CPU being able to access the RAM (and anything else that sits on the FSB) at 33MHz instead of 25MHz is going to provide a significant performance increase. If he'd tested the RAM access speeds (after the cache runs out) between the two setups it would have been pretty obvious what was going on.
I'm running Cyrix 5x86/100 @2x50MHz FSB on SiS471 486-board and it flies compared to 3x33MHz FSB. My Cirrus 5434 VLB is faster than any 33MHz PCI display card (in DOS).
For the 4x multiplier just isolate the clkmul pin of the cpu from the board and manually pull it to ground (or check if it goes to any resistor or jumper like the clock generator did). Ive literally played the same game on mine which is a siemens nixdorf d819 motherboard. I just connected it to chassis with a wire and its happily running at 160mhz giving around 1300ish realtics in doom. Note: a 5x86 ADW didnt like the 4x40mhz no matter of any overvolt even at nearly 4V. I changed it to an 5x86 ADZ and all good.
When i was a kid, I exploded a 286 trying to overclock it by replacing the crystal on the motherboard. I had succeeded in making my 286 run super speed, and was going for Warp Speed when the chip when POP with a cloud of magic smoke and a huge blast of shattered silicon. It was AWSOME, and I'm thankful my uncle gave me the junk computer parts to play with. At school we were running 486's and upgrading to Pentiums, but at home I was moving from 8088 to 8086 to 286 to 386. always a few generations behind so parts were plentiful and i could do stuff like overclock my old pc when I got an upgrade. I was trying to make the 286 run more than 2x faster. It worked at 1.5 speed. It was working at 2x speed for something like 15 or 20 seconds before the screen stopped counting the memory. then the screen went blank, and POP.
We were actually playing around with some AMD 133mhz 486 chips last night. If you can modify the board to a 2x multi to get 4x, try leaving the fsb at 40mhz and put a heatsink on it. We ran 8 different chips last night, All ADZ models were stable tbrough some quick benching,1 ADW was too but 1 ADW wasnt able to get past post. Reading online the ADZ models are the ones that are more likely to be overclocked as they are the newer ones. Enjoying the videos mate!
We Kids tuned our 486s all the time for our ethernet / doom sessions. My brothers final config was an amd 2x50 MHz mounted to the wall w/o a case and I had an intel 3x40 MHz. Happy times until quake killed the 486 😂
Back in the 90's a co-worker learned about the clock generator's 80mhz feature, and was able to run a DX/4-100 at 120mhz. He had to add a fan to keep it stable
During the same time, there were VLB motherboards and adapters of differing quality that could run fine at 33MHz with two adapters (commonly video and HDD controllers), but the combination was more unstable with a 40MHz FSB. It was a fine art of cache and memory timings to get the best stable performance. I don't know if I miss those days.
@@IBM_Museum VLB relies a lot on what clock the card can handle. They'll all run at 33MHz no problem, some small numbers of them will crap out at 40MHz, and finding ones that run stable at a 50MHz VLB clock can be a real challenge. It was a serious problem when the first 486DX-50s came out to find VLB cards that worked reliably with them to the point that many small system builders just wouldn't sell systems configured like that.
@@JeremyLevi: I ran an AMD 486DX4-120 that was challenging to have two VLB adapters (video and HDD/FDC) on a 40MHz FSB. Once it was stable, I really made sure I could make changes before locking it in. Those were the days!
Doom always seemed to really like extra bus speed over raw speed. I found this out in 1996 when I ran my AMD 5x86-133 both at 150 and 160 mhz, and left it at 150 mhz because both Doom and Quake were faster on the 50 mhz fsb. I had a very tolerant Diamond Stealth 32 Vesa card for video, and Quake was 'playable' with that setup until I finally saved enough for a Pentium 120 later in the year.
I remember us getting Quake for Christmas 1996 but we could not play it until we got our first Pentium the following month (Pentium 133). January 1997 we FINALLY had a Pentium and HOLY SHIT was it fast! The plasma rifle in DOOM/DOOM2 no longer slowed the game down like it did with the 486sx25 which we suffered with up until then.
Totally different MB but I also had an AMD DX4/133 or 150 or something and I could run it with FSB of 50, 60, or 66 MHz and clock doubled or tripled with VLB hard disk and video and it was a screaming machine. I think it wasn't stable at 66MHz but at 50 or 60 FSB it worked.
I'm very interested in the next video! Back in the 486 days I had a Packard Bell PC that came stock with a 486 DX2 66MHz. I later upgraded the machine to a AMD 5x86 133 CPU also using an in socket voltage converter. In my case I ended up having to set the board to 40 MHz FSB and used a 3X multiplier for a final speed of 120 MHz At the time there wasn't a way to set the proper multiplier and this was in the very early internet days so I had to go with the info that was etched onto the board itself. About 6 months before I did the CPU upgrade I also did a L2 cache upgrade from 0KB to 128KB. I read an article in a computer magazine where that was done on a different mainboard that had empty cache sockets that was similar to mine. I got lucky in picking out the correct cache and tag ram chips. The L2 upgrade also provided a nice speed boost in many things.
The appropriate pin on the Socket 1 diagram for the CLKMUL signal must set low (i.e., shorted to V_ss) to enable 4X on the Am5x86. That pin is possibly desigated R-17 on the Socket 1 layout.
From the technical specification: The CLK input provides the basic microprocessor timing signal. The CLKMUL input selects the multiplier value used to generate the internal operating frequency for the Am5X86 microprocessor family. All external timing parameters are specified with respect to the rising edge of CLK. The clock signal passes through an internal Phase-Lock Loop (PLL).
The microprocessor samples the CLKMUL input signal at RESET to determine the design operating frequency. An internal pull-up resistor connects to VCC, which selects Clock-tripled mode if the input is High or left floating. For Clock-quadrupled mode, the input must be pulled Low. For 133-MHz processors, this input must always be connected to VSS to ensure correct operation.
If you modify the interposer board, or otherwise run a wrap wire to the appropriate underside socket solder point (or pin on the CPU), I presume you would be attempting to run this at 160MHz. I think it would work, perhaps 10ns L2 cache might be in order (and a heatsink-fan), however, for stability?
That board is also supposed to support the Original Intel i486DX-50. I wonder how you enable the FSB to run at the same frequency as the clock generator instead of half. Then you can try to overclock the AMD CPUs to have FSB of 60 or 66 MHZ LOL! I remember when I was working at the computer store back in the 90's we had over clocked a DX50 to run at 60 or something like that. It was awesome at the time!!!
@@GigAHerZ64 my 5x86 ADZ barely gets warm in 4x40mhz mode in a few minutes. i can keep my finger on it and its only slightly warm. i do have a nice flat active cooler on it though.
I'm not surprised by the performance boost in the DX2 at 66mhz over the DX4 at 75mhz. The 486 is really limited more by the bus speed than the internal clock speed of the CPU. So running a 33mhz bus will tend to net a bigger performance gain over a 25mhz bus. A DX4 at 100mhz (33mhz bus with a 3x multiplier) would potentially get a nice boost over both, but the 5x86 is still going to beat out any 486. All this makes me want to try to overclock my IBM ThinkPad 701c with a 5x86 in place of the DX4 since I've read people use to do that to them back in the mid 90's. But as that is all SMD in a small laptop I'm afraid my limited skills would end up damaging it.
What a great hack. It's really interesting because I've got an Ambra 486SX/25 and the board is manufactured by IBM - same black PCB and the layout is similar. However, it has jumpers for 25, 33 and 50MHz. This is confusing for two reasons: 1) because of the lack of jumpers on the PS/1 board and the lack of 50MHz, and 2) because I don't know of any 487 that ran at a 50MHz FSB. I might do the hack in your previous video to see if a 50MHz DX runs stable in my system.
Cool content Roman! Right up my alley 👌 I have a similar setup on my Compaq Presario 425, which is wired to run at 33MHz and is then multiplied x4 by my Evergreen 486 upgrade which is essentially an Evergreen OEM "Socket Blaster" with an Am5x86 running at 4x33MHz=133MHz. The original Compaq Presario machines were in fact Compaq's versions of the PS/1: Cheap, low end machines for home users, which makes them even more fun to tinker with and get the most performance possible out of 😎 P.S. the DX4-75 is slower because the on board local bus graphics card and the memory is running at 25MHz instead of 33, which makes the DX2-66 faster. Memory/bus speed is often more important than raw integer speed of the CPU itself, which the DooM results demonstrate here 🤓 The i486DX2-66 was in fact the ultimate sweet spot between bus speed and raw CPU speed, making it a very stable and well performing platform at the same time, especially with its VLB implementations @ 33MHz, which is what made it the most popular 486 CPU ever made.. Love the analog synth tunes in the background. Did you make them yourself? I noticed the nice analog synth gear in previous videos ;-)
Looks like the IBM PS/1 is a bit of a sleeper 486 system. Very impressive CPU figures when you know what to hack and replace to get max performance with the resistors and the socket interposer.
It was rumored .. 'in the day', that the IBM 80 bus could easily be clocked up to 80mHz..perhaps even to 100hHz. I had a PS/2 50z that I upgraded with a socket conversion to a 486/7 AMD type chip, but the bus was still at 25mHz.. and the add-on video board and RAM was the reason that I didn't even think of up clocking the bus.
back in the day with dx2's we used to build always had a heatsink and fan... sx25's dx33's didnt need them, a lot of old vesa cards used to choke @40 and 50mhz , for DX50 rather than a DX2-50 or DX2-66 we used to advise to put an ISA card in for stability - and would always advise the customer to go to a DX2 CPU instead.. also memory could be problematic @ 50mhz bus speed
The buck-converter you have definitely has ripple, but you need to measure output voltage only loaded! And beware the smd trimpot is very unreliable, better measure the real resistance and replace it with the smd resistors. /*with the time the trimpot can go crazy because of moisture and dust*/
Perhaps those "clock" jumpers, that are documented in the PS/1 technical manual but in an obscure fashion (possibly not implemented the way the diagram suggests on the average 2133/2155/2168 motherboard IBM shipped) provide signal to R174, R175, R176, if they are all populated with 330Ohms and if the installed crystal and CPU combination matches the intended multiplier settings?
yes, thats what i have done at the last days, bevore the Pentium arrived me.... for fun, it was running @200Mhz, but the risk was to high, so reduced back to 166Mhz
I don't think they were locking down features. Rather than saving cost on manufacturing. Many companies did this. They make one product with the same layout but enabled different setups. And so could sell at different price points. They still do this today. Reason why so many cpus work in the "socket" was just because of backwards compatibility. This changed when Pentium II came out as much of the architecture was on the cpu itself rather than scattered on the motherboard
@@Pholiage I was more on the fact that PS/1 and PS/2 line was the time when they still tried to regain control over PC by making it hard to copy and extend. But I agree with you on reusing the same platform and that is actually good practice.
IBM was set on only running Intel CPUs - The Aptiva series (which came after PS/1s for the home market) were much better than the 'ValuePoint' models offered for businesses. It was a strange time as IBM (and other brandnames) got hammered by clones running AMD CPUs.
Now put an 83mhz Pentium Overdrive on the '486 socket! Then get a riser card that has PCI slots and add an evergreen technologies 550mhz Celeron-on-a-card! :D
DX4-75 means that the Front Side Bus (FSB) is running at 1/4 if 75Mhz (18.75Mhz) while the DX2-66 means the FSB is running at 33 Mhz. If I recall correctly it was cheaper to buy a DX4-75 than a DX2-66 way back when. Thank you for the video!
It's one of those confusing things for a name of an Intel '486DX4' to have a 3x multiplier - a DX4-75 has a FSB of 25MHz, and a DX4-100 has a FSB of 33MHz. An AMD 486DX4-120 is also a 3x multiplier of 40MHz, and the demonstrated AMD 5x86-133 is a 4x multiplier of 33MHz. Wait until you hear about the Intel "Pentium Overdrive for 486" that had a 2.5 multiplier: The 'POD63' is a 25MHz FSB, and the 'POD83' is a FSB of 33MHz.
@@RuruFIN My bad, thanks for the correction. Still the DX2-66 had better performance hence higher price. Eventually Intel realized that it could not keep up with the DX* naming scheme and decoupled the FSB from the internal clock as it moved forward.
That AMD 5x86 was the first big upgrade I bought with my own money back when I was a teenager. I went from a 486 DX2-66, so as you experienced for yourself, I also saw decent increase in speed in Doom and other games, but it wasn't as big a performance increase as I expected. Certainly not the Pentium levels of performance that AMD claimed. That's probably down to the 33MHz FSB, compared to the 50-66MHz of the Pentium. In hindsight, I probably would've been better off buying a second hand Pentium. I think this would've been around early 1996, so there definitely would've been plenty of 100/120MHz Pentiums on the second hand market by then, possibly the odd 133MHz systems. But I didn't find out about the local computer markets until later, so it would be a couple of years until I could afford to upgrade again.
With enough RAM, cache, and a good video card, a 5x86-133 system was close to a Pentium 66, except with FPU-intensive software like CAD. It was also less than half the price.
As the other poster indicates, with the right motherboard and a suitable BIOS (or workarounds), VLB graphics card, performance was probably similar in many instances to a bog standard P75 sytem being pushed by the likes of AST, Acer, Packard Bell, etc during the mid-1990s.
Yours was probably held back by the motherboard. I also bought a 5x86 back in the day, but paired it with a new (PCI) motherboard and ran the 5x86 at 160MHz (so 40MHz FSB). Early Pentiums could not touch that. It could still give a Pentium 90 a run for its money (not in Quake, obviously). I later upgraded to a Pentium 100, because that definitely was faster in all scenarios.
I remember wanting a Pentium so badly in 1995 and 1996. We FINALLY got one in January 1997 and I still remember the very first moments of playing on it and the ENORMOUS difference from the 486sx25 and then being mad as hell at having to suffer with that budget 486 all those years.
An 83MHz Pentium Overdrive ( PODP5V83 ), aligned approrpriately with the inner matrix of pins on the Socket 1, with free-hanging pins (and possibly a wrap/bodge wire to the write back pin and the appropriate pin on the motherboard chipset) would be faster than this Am5x86 at 133Mhz (or 160MHz). If you want to try something even crzier than this in the future.
I have an IBM PS/1 Rapid Resume tower with the same looking board that I put a Pentium Overdrive 63MHz in. The computer was given to me in 2012 and it had a 486DX2/50 and I upgraded it for fun just for DOOM/DOOM2. Now I am wondering if I can make it run at 33 FSB.
The DX2/66 putting out (slightly) more FPS than the DX4/100 doesn't surprise me, because in many cases, bus speed (as opposed to internal CPU clock) is your main bottleneck, especially in doing stuff like pushing graphics out to the graphics card, etc. With the DX2/66, you're running at 33 MHz bus, whereas with the DX4/100 you're only running at 25 MHz.
You should have a load while adjusting the voltage to the cpu. Preferably the same as the cpu uses. Can Socket blaster be rebuilt to do the cpu multiplayer?
OMG PUT A HEAT SINK ON THAT QUICK!!!! I wouldn't clock that any higher with the bus up at 40 Mhz. Not unless you do some much better cooling. Those chips have been known to run at 160 Mhz stable but is the minor increase really worth the risk? A big portion of the speed increase you are getting is already in place by overclocking the bus. If you do clock it to 160 PLEASE get a really robust cooling solution going on there. Check out CPU Galaxy and his adventures with one of these up to 200 Mhz but he had to use a peltier cooler to pull that off.
Reminds me of the original PC/AT. The motherboard had a 12MHz crystal but the schematics labelled it as 16MHz. Sure enough, a 16MHz crystal let it run just fine at 25% more speed. Then IBM noticed people were doing this and they added a speed check in the BIOS startup code so it would not boot if it was running at over 12MHz! So you needed to have a switched divider, putting out 12MHz for 10 seconds, then switching to 16MHz. Easy peasy, almost.
When the odd "586" cpus were released, I never understood the draw. Pentium was out and offered a much better architecture. But for this use case they are really great.
It's more of a straight swap. Keep in mind tech was evolving FAST at that time, so if you could stretch your system foe awhile you could wind up saving money. Also, the AMD 5x86 was essentially a die shrunk and OC'd 486, so they were cheaper than you'd think. With that clock speed bump they were about as strong in most common tasks as a low end Pentium (the reason for the P75 rating). The main caveat: FPU stuff - the Pentium wiped the floor with practically everything else there. This wasn't an issue for most gamers till Quake came out, however.
@@mibnsharpals pentium was released before the faster 486's were released. I don't think there was any phase where 486 off the shelf was faster. cheaper yes. depended a bit on what you're running how much worse the 486 is. it was already 1995 by the time of the 120mhz 486.
xD lol push amd dx4 to limited and make this super faster lol oh dont forgot stick heatsink on it because she does get hotter at 3.4v or 3.5v just save yourself a headache in future ya knew ibm using tricky ideas putting resister on it got me laughing so tiny cant see it
they hid the fact the bus speed was adjustable by using the smallest resistors they could find instead of usual jumpers. basically without redesigning the board they could sell a faster model for far higher that really had the same exact parts as this one. typical of computer companies, it's still happening to some degree.