Well. you definitely nailed it!! 👍🏻👍🏻. Since I had not time until now to finish my video on this topic i was very curious about your results. And wow, i am really flashed. 35+ fps is amazing. I have no Idea how you could squeeze this out of that ISA setup!! I was working for long time on my setup and I could not achieve that. Maybe coz I was using an older pure ISA board with older chipset… but damn… 35 fps 🤯. I was even 1 MHz higher on the bus than you, and as you said, the hardest part was to find an IDE controller which could handle that stable. Congratulations 🍾 to this absolutely great video. I enjoyed it from the first till the last second! Thank you! Cheers, Peter
Thanks so much, Peter! I really appreciate it. Yes indeed, the IDE controller seems to be the weakest link for this. But the CPU, cache and memory performance also makes a big difference in Doom, that's for sure. Would be really interesting to see if 25FPS could be acheivable with a much slower DX2 or something similar. I may give this a try in a follow-up video. I have already beat my previous score by a fair bit too :)
Given the rarity of Mach32 ISA card I’d be interested in what other ISA cards could be overclocked that much. The Mach32 chipset is also used on PCI bus, so that explains its overclockability.
I've had some success with the Tseng ET4000 after completing this video and will be doing a follow up at some point. Would be interested in hearing what others had success with too.
the trick is picking ISA card using chipset that also supports VLB, that way it will happily tolerate up to 30MHz. For example Cirrus Logic GD5424-9 or WD90C33.
Impressive results! If you want to cheat a bit more, try the 25 MHz ISA bus setup with FastDoom's mode 13H. This mode renders everything on the RAM, and then copies only required parts to VRAM via the ISA bus, thus reducing a lot the bandwidth needed (I know, cheating is bad 😋)
Love how you casually smashed Peter's result and didn't even say a word - very classy, Mike. 🙂 The motherboard is definitely the key to this project, but having the adjustable divider on the CPU is also very important, so those donations have been key to your success! It's hard to imagine anyone else exceeding your 35fps result in Doom, that's for sure!
Thanks very much for watching! It was a lot of fun. I'm really looking forward to seeing what others come up with in the challenge. I'll be doing a follow-up at some point as I've already beat my score twice over. More to come! :)
I think the trick is to not only have as little as possible on the ISA bus, but to also use a vid card that was advanced enough to have versions that ran on the VLB and PCI buses as well.
This answered a lot of questions! In the 90s I grew up with my dads PC, a 486 of some sort and Doom played pretty smoothly yet I know we were on a budget so NO WAY did he have a fancy VLB system. This must be what he did - overclocking the ISA bus combined with knocking down the window size a couple of notches (we played in high detail mode too). Interesting tests - cheers!
My jaw dropped when you got the XX max fps you did haha. Great video, great work on showing what was actually possible with the ISA bus. (had to edit to hide my spoiler, sorry!)
I've been following your channel for a little while now and I enjoy your videos. I worked in a little computer shop in Winnipeg in the mid 90's so I have nostalgic memories of computers at this time too. I recently built a 486 DX-2 66 VLB system and even the VLB graphics card struggles to reach 25fps in the Doom benchmark. These ISA overclock results are very impressive! Still, Doom really isn't my motivation for this build, so I'm not sure it matters all that much. :)
Always interesting to really see these old systems pushed. Motherboards are so fleeting these days, only working for a small collection of hardware and here you are pulling 250% overclocks on the system bus with 30+ year old caps and transistors.
Oh shit! This is my chance to start my channel! I have a nice 5x86 @40fsb (160mhz) and a Wyse ET4000AX that I know will get above 23fps. Nice! 32mb EDO. 256 12ns cache. Dell XPS 4/66v motherboard (aka Micronics most likely SIS) with VESA. Was 5v but using an interposer (TrinityWorks) with an AMD 5x86 Write Through cache CPU (ADW stepping).
Think I had a 486SX at 25 MHz at the time of Doom's release - later upgraded with a DX2 Overdrive to get 50 Mhz which was staggering - but the IBM PC did have a VLB Cirrus Logic VGA card. Doom ran fine(ish) even with the SX 25 Mhz as I recall. Ofc with the DX2 Overdrive it ran super smooth, later.
I had the same cpu and yeah it wasnt the bus that was the bottleneck there it was the lack of FPU in the 486sx chip. I had no idea about the buses so this was interesting. I also tried to upgrade to the overdrive chip but being young and literally my first attempt at an upgrade I missed the "ovverdrive" detail part and tried to stick a regular 50mhz dx chip i found locally in it. AFter bending the pins a few times i realized it jsut woundt fit and didnt get it until I got on the phone with IBM and they told about the overdrive chip part, hah, which gave the cpu an FPU unit apparently which was needed to run games well back then before GPU's and I had no idea about the buses other then later when slight overclock there was an easy overclock when you couldnt overclock your cpu. When duke3d came out the 486sx was pretty much toast, doom was still playable but i really woud have liked that upgrade for duke3d. But that didnt stop me from honing my fps chops in doom and actually winning a duke3d tournmanet without actually being able to run the game. I played my friend in the final who i'd watch play duke hehe. He was running it on a cyrix 166 i think it was.
@@mikejones-vd3fg pretty much no game used FPU seriously before Duke Nukem 3D/Quake. Duke only when slopes were on screen, but to run Duke at all you already needed around DX2, and Quake required Pentium for fluid gameplay.
I remember doing the same thing, not just video cards but also the venerable NE2000 could be overclocked as long as everything on the coax bus was equally overclocked. Most motherboards had a jumper for the bus division intended for different CPU speeds, which if set “wrong”. We always ran games on 12MHz, using the turbo button to switch, but 16MHz wasn’t stable with IDE and serial controllers (leading to corrupt data), but it was with Adaptec SCSI cards.
It was very card-specific in late 80's/early 90's. I ran 3com NIC's successfully at 12.5MHz and it helped speed-wise with early networking. Trident was notoriously bad for a reason, but Cirrus was good. I used overclocked ISA Cirrus 5422 for a while with 386DX/40, it worked ok with 13.3MHz AT-bus (40/3). Mach32 was one of the priciest cards around (with Matrox) so it was very rare, Cirrus was much common for a reason. MCA was Very pricey to implement/license - reason it never succeeded. PCI was the real solution.
Very very interesting. 🤔 My childhood 486 (which I still own) is a 486DX/50 with the FSB running at 50MHz. It was always vexing as a curious kid why I was stuck with ISA graphics since the mainboard had VLB, but it never worked. (To know then, what I know now.) I'm adding a project to my list: Explore "ISA overclock options". I have a Cirrus Logic 5434 (late model ISA video card to experiment with) and it'd be neat if I can bridge the performance gap with my VLB DX2/66 board. Great video!
VLB became bit unstable at 40MHz with only 1 VLB-card. Even running two 33MHz VLB-cards like multi-IO and VGA was slighly unstable with some cards/motherboards. 40MHz was almost guaranteed to fail somehow, mostly IO-problems (like corrupted HD-transfers). Serial ports still worked ok, at least to 115200-speeds when used with 33.6k modems. Remember to overclock your 5434 with MCLK, it made visible difference in some DOS-games with 542x-series cards.
@@tomiluukkonen4035 Indeed, and the board in that system does support the DX2/66, which is what confirmed for me that VLB was capable of working. Once I figured out that I'd been trying to overclock the VLB video card (a Cirrus Logic 5429) to 50MHz as well, my 16yr-old self back then started looking for what I could maybe set to split the clocks... If I down-clocked the DX-50 to 33MHz, it worked but that left quite a lot of performance on the table. Jacking up the wait states didn't work either... 50MHz was just too spicy. (The board doesn't support 40MHz unfortunately. It uses a dedicated 33MHz crystal and a 50MHz crystal) Now I want to tinker with it again.
I remember setting jumpers on my 486 to eke a bit more speed out of my processor, but this is next level stuff - really enjoyable and a little trip down memory lane! My second ever PC was an IBM PS/2 (386) with MCA. Proprietary and expensive.
Would like to see how far a VLB system can go and compare it to these results. I remember being able to push my DX2 80Mhz system pretty much up to full tilt, in terms of cache/memory timings & bus divider, without any issues but even if I had done benchmarks back then, I wouldn't remember the results now. Anyway, very entertaining video, thanks!
Cool ISA bus over clocking! I've tried doing a little of that on my homebrew 8088 machine as it's ISA bus is clocked directly to the CPU clock and could only get it to 11MHz before my Trident based VGA graphics cards (9000 and 8900D) refused. I think my U3 EGA card could get to about that level too before it glitches out. Thanks for showing the glitch out screen for a graphics card, that's really helpful to see on someone else's machine. My machine also probably doesn't have the best signal integrity so I'm pleased I can clock it that fast. Defo no DOOM on my 8088 (V20) tho I can just about play CGA Wolfenstein 😅
Thanks Chris! Yes, I think that would be a much bigger challenge. Doom is very CPU intensive on 486s so that would be interesting to see. I plan to do a follow-up on this one and may give it a try with some older chips too.
Great stuff. I remember overclocking my ISA-only 486DX50 board. It had some strange SCSI controller card for HDD and I later purchased a Cirrus Logic GD52424 for an already remarkable graphics boost. I was a kid of practically absent knowledge and only had read about overclocking in a magazine but I just had to try setting the relevant jumpers on the board. And indeed I got Doom running significantly faster and so did the SCSI. Of course higher settings would produce graphics garbage but I was still very happy with what I could achieve that way.
Interesting :). I wonder if you could overclock the ISA bus on a Slot 1 motherboard (Award Bios) with in this case Pentium a PIII 700 mhz?, in some DOS games it seems with a CPU faster than 550 mhz the sound does not work as it should with for example a Sound Blaster 16, CT2910 in this case, so i wonder if overclocking could help there?? hmmm.
Good question - if the BIOS supports changing the ISA divider it should be possible on more modern systems. I didn't realize that the CT2910 was speed sensitive. It would be interesting to see if speeding up the ISA bus could help.
@@vswitchzero Yes, also the motherboard and BIOS version could play a role?. Newer Windows 95 - 98 and DOS games work fine, it's just some older DOS games that have a problem with the speed of the CPU and the ISA bus it seems. Just to mention a few DOS games, Star Wars X-Wing (there is a fix for that one), Star Wars Tie Fighter (works most times as it should), Space Quest 4, Prince of Persia, Alone in the Dark (intro does not work every time as it should, or some sounds are left out). Keep up the good work :).
Impressive results. While i dont remember the fps for Doom on my 50mhz bus setup with a 5x86 and a Diamond Stealth 32 VLB card back in the day, i know it was a lot faster running 3x50 than 4x40. I always assumed that was due to memory bandwidth but this makes me wonder if the vesa bus was the limiting factor. Sadly the board died in the late 90s and i got rid of it so i cant go back and check, but i still have the cpu and both vesa cards for video and i/o. I may have to give it a try if i get time and can get the random 386/486 board to play along. I know I'll need to drop to 3v as well.
What a great video! I used DOS computers in this time but I was not even 10 years old and didn't know almost all of this. As you probably know, Doom is limited to 35fps for normal gameplay (not DoomBench timedemo) so you've built the ultimate Doom machine! What is the CF-IDE adaptor you're using? I've got three and none of them boot in my PII 440BX system.
have you though about building a clock divider for the cards that dont run over 11mhz, 16mhz, 25mhz ? perhaps something that can detect clock no matter where it is, then lower it to the correct clock speed. this would be done slot for slot or card for card maybe, or perhaps somethin that can read clock, and distribute clock to the cards that fault at higher rates. you would do this for any and all isa slot cards you are using, and leave the VLB cards to themselves? perhaps in the case of VLB, ramp the clock way up and lower it in iteration till the clock stabilizes and write to say a small eeprom that connects to say a variable programmable resistor or something of the sort. speed control for the isa, timing control from peak, out of range to highest stable for the vlb.
you would have to detect address on the bus and slow (stretch) the clock for all addresses !A0000, I dont know PLLs capable of reacting that fast, fpga listening to address bus could do it. Still leaves the possibility of other cards glitching on faster clock and phantom activating or crashing while VGA is being accessed.
I just tried my non overclocked 133MHz 5x86 PC with an S3 Trio 64 PCI graphics card. It delivered 40.5fps on the default settings. So the ISA bus is still the main bottleneck, but it's not too far off. But on the other hand My fast 486 could get even more frames @3* 50MHz with tweaked cache and ram timings.
I must have had a PCI based 486. I barely remember that. The first system I ever built (with help) was an AMD K5-133. I tried running it at 166, but it was unstable. And obviously that's going to be significantly faster.
I wonder if a mach64 ISA card would help at all or if the mach32 is already saturating the bus completely. I do not have one of those as much as I want one.
Welp. I did it and put it in a video and got better results somehow. And then I did it with an ET4000AX and got a better result than that. This is my ET4000AX run at a whopping 40 FPS: ua-cam.com/video/n92DY5z8gb4/v-deo.html
While I appreciate your take on the challenge, using a CPU released in late 1996 seems to nullify the goal of the challenge. Would be much more interesting to have a rule that specifies that no hardware older than ~ the Doom release date can be used. I'd rather would like to see how it is possible to achieve the 25fps mark with the cheapest hardware possible that was available at that time and make a score based off that.
Thanks for your comment. I really like the idea of having a cutoff of the doom release date. I plan to do a follow-up on this video and may give that a try. Thanks for watching.
Good question. Would be interesting to try. I have some plans for a follow-up on this video so stay tuned for some other interesting tweaks potentially :)
I have always been curious and interested in experiments squeezing the Maximum out of an older technology like you just did. Got some questions for anyone reading this: Was it possible to overclock MCA and EISA back in the day? could they have came close (or even keep up) with VLB and the first PCI? Would they have been able to power some early 3D Accelerators from 1994-1996 (cards for both CAD and Gaming) without significant bottlenecking?
One of those boards had a whole bunch of 74ls chips for interfacing to the isa bus, I wonder if that was at all the limitation as I seem to recall the ls family kinda tops out around that speed anyway. Could be a fun experiment to try a newer ttl-compatible family and see what happens 🤔
I used to OC the ISA bus on my 486SLC way back in the day. It didn't buy me nearly as much, as that chip was slow as hell, just like the rest of the machine.
I wish I had known about overclocking the ISA bus back in the '90s while we were suffering with the budget 486sx25 from Sams. It would have totally been worth facing my father's wrath if caught tinkering with the board if it meant DOOM would run faster.
ISA has 8 MB/s bandwidth. 320x200x8 frame takes less than 64KB and it takes less than 4.5 MB/s of bandwidth to update 320x200 screen 70 times per second. DOOM (in-game) has 35 fps limit, so it should need less than 2.5 MB/s of bandwidth. However it seems that graphics cards are limited to about 1.3 MB, maybe a bit more since not every pixel is drawn once (yet DOOM engine is optimized to avoid overdraw). Anyway now I'm interested how fast real bandwidth is and how it is divided between multiple ISA devices.
This gave me two thoughts. One was that obviously some devices deal with an overclock better than others. So is it possible to have some sort of reclocking ISA to ISA bridge and two speed domains, one running nominal speed for most devices and one overclocked for just a handful devices which would survive that? And the other that ISA streaming digital audio is a weird beast. For cost, soundcards do not contain a sample buffer of more than a couple samples in capacity, certainly not a whole page full. For simplicity's sake, just about none of the ISA soundchips support true Bus Master, where they would be able to request data from system memory as needed. Instead they fire an interrupt and have software cook a page of audio data and set up DMA to supply samples when they're needed. DMAC runs at a low frequency i think 3.5MHz ish nominal (i forget, my memory is vague) and while it performs operations, slows down the whole ISA bus - which is also why DMA video wasn't viable and ISA DMA disk access would be a horrible idea as well, it would be too slow. This lead me to think, why would digital audio become unstable when you overclock ISA? Perhaps the whole problem is that the DMAC speed isn't set correctly, and ISA bus speed that you set explicitly isn't even the issue?
Thanks for the comment! I wondered that too - like a frequency divider for more fussy devices. I don't really know enough about the ISA bus' inner workings to know how practical that could be to implement but an interesting idea none the less. Interesting about the DMAC clock frequency too - I hadn't thought of that. I'd be curious to see if it gets thrown out of spec along with the ISA bus but not sure how to measure it.
@@vswitchzero Let's assume that a mainboard exists which has an explicit clock signal for ISA DMAC and an explicit clock signal for ISA, something perhaps less integrated, maybe discrete 8237a; perhaps you can use that sort of board to piggyback an Overdrive/Accelerator system of some sort on it; perhaps something where the RAM isn't on mainboard at all but on CPU accelerator board? Not that i even recall anything like that exiting. I have a strong suspicion that it will not lead to a very successful ISA overclock since even if you have a suitable topology, you'll have low speed grade components to deal with. And yet i'd LOVE to be proven wrong and it's just fun to think about. What if.
A Dangerous game, i might add. Dont try this with VLB grafic cards... i had success up to about 12MHz... Most VLB cards wont handle that (my 801/805 S3 did..) I saw many VLB grafic cards die because of incorrectly set ISA dividers. Be careful... What i can remenber is, that it is very efficient to get a board where the DMA waitstates can be manipulatet. (AMI BIOS, mostly a hidden option) Setting this to 0 will increase the performancy significantly, especially with overclocked ISA
I remember having a 286/16 and trying a 1:1 bus ratio. My CGA card burst into flame! But once I'd patched around the damage (which was oddly at the output, so it may have had nothing to do with the ISA bus) the whole thing worked fine and was quite a bit faster! I didn't have the nerve to leave it unattended like that though, after what had happened to the CGA card.
Interesting, thanks for your comment. I always wondered this because some older "turbo" 286 systems will run the ISA bus greater than 8MHz, but I was always under the impression that the specification called out 8.33MHz as the maximum "official" frequency for ISA BCLK. Will have to take a look at the whitepaper :)
I've used the COOSPIDER CTUV-6 to erase EPROMs and had really good results. Way cheaper than the purpose built units for erasing EPROMs. Please, at the minimum use a cardboard box to house the light, the UV is not good for you.
Did you try to upgrade the clock crystal of the ACER I/O board to see if it would work at hight bus speed if the board was actually also running at a higher freq?
This video inspired me to tinker with my 386dx-40 for this challange. I don't expect to get anywhere near 25fps but will be intresting to see what I can achieve. Unfortunatly my ATI Mach8 card seems to have bad ram. I'm currently searching my parts stash for the best alternative i have on hand. I do wish i had a scope to verify bus clock though.
Thanks for your comment. For my very basic needs, it's been a good scope (especially for a little over $100). I'm sure there are much better ones out there if you have a higher budget, but I've been happy with it so far.
@@vswitchzero Thanks for the reply! My needs are pretty basic (checking cpu vpp) and I think that one will do the job! BTW, googling "isa bus overclock" returns this video on the first place! 🏆
Of course it almost rivaled VLB cards, VLB cards run at the same speed as the CPU FSB, as such DX/2 50Mhz have a 25Mhz FSB, DX/2 66 are at 33Mhz and DX/2 80 are at 40Mhz. ONLY the 486DX 50Mhz has a FSB at 50Mhz (FROM THE FACTORY, no overclock). Aside the 486DX50Mhz (The first CPU to require active cooling to remain stable in fact) Only the DX40, DX/2 80 and DX/4 120 had FSBs faster than 33Mhz and only a single 486 model ever had a native default 50Mhz FSB sadly. Now, I would recommend you to get a rarity of a motherboard, a VLB 486DX motherboard that has a Weitek FPU socket (yes it can run at the same time as the 487 as it is memory mapped ) they exist in fact, and then do yourself a favour and get AMDs 386DX 40Mhz and a 387 40Mhz, then test this. Please use a graphics card which has the chipset available in both ISA and VLB models if you truly want to compare the performance difference between these 2 interfaces. Note that VLB first and foremost is a DIRECT TO CPU link and if that is not enough it also is natively 32bit :P
@@vswitchzero heres my argument towards what you said id say if some one had that board and still had old isa card that would be because they spent all their money on the mobo and is waiting to save up to get new vlb card
@@vswitchzero acording to google 5.3MB/s 8.33 mhz 16 bit slot which means 100 mhz overclocked bus would make 64.5 MB/S it would take 8.33x12.10% = 100.793 mhz 12.10 = 1210 %
Well. you definitely nailed it!! 👍🏻👍🏻. Since I had not time until now to finish my video on this topic i was very curious about your results. And wow, i am really flashed. 35+ fps is amazing. I have no Idea how you could squeeze this out of that ISA setup!! I was working for long time on my setup and I could not achieve that. Maybe coz I was using an older pure ISA board with older chipset… but damn… 35 fps 🤯. I was even 1 MHz higher on the bus than you, and as you said, the hardest part was to find an IDE controller which could handle that stable. Congratulations 🍾 to this absolutely great video. I enjoyed it from the first till the last second! Thank you! Cheers, Peter
Thanks so much, Peter! I really appreciate it. Yes indeed, the IDE controller seems to be the weakest link for this. But the CPU, cache and memory performance also makes a big difference in Doom, that's for sure. Would be really interesting to see if 25FPS could be acheivable with a much slower DX2 or something similar. I may give this a try in a follow-up video. I have already beat my previous score by a fair bit too :)
I'll just join Peters opinion, just wow! 😮 Great video and great results!
@@necro_ware Why am I not surprised to see you here? 😀
Did you send gold trophy to him ?
Peter we need you back 🙂
Given the rarity of Mach32 ISA card I’d be interested in what other ISA cards could be overclocked that much. The Mach32 chipset is also used on PCI bus, so that explains its overclockability.
I've had some success with the Tseng ET4000 after completing this video and will be doing a follow up at some point. Would be interested in hearing what others had success with too.
I have that ATI Mach32 in my ICL MikroMikko D4/66XGi, VLB-version on motherboard. It's fast, DX2/66 runs Doom fluently.
Mach32 isnt rare and easily obtained
the trick is picking ISA card using chipset that also supports VLB, that way it will happily tolerate up to 30MHz. For example Cirrus Logic GD5424-9 or WD90C33.
@@rasz Cirrus 5434 VLB is sick fast and ISA version too if you can find one 👍
Impressive results! If you want to cheat a bit more, try the 25 MHz ISA bus setup with FastDoom's mode 13H. This mode renders everything on the RAM, and then copies only required parts to VRAM via the ISA bus, thus reducing a lot the bandwidth needed (I know, cheating is bad 😋)
Haha that's really neat. I've been meaning to try out FastDoom one of these days.
Love how you casually smashed Peter's result and didn't even say a word - very classy, Mike. 🙂 The motherboard is definitely the key to this project, but having the adjustable divider on the CPU is also very important, so those donations have been key to your success! It's hard to imagine anyone else exceeding your 35fps result in Doom, that's for sure!
Thanks very much for watching! It was a lot of fun. I'm really looking forward to seeing what others come up with in the challenge. I'll be doing a follow-up at some point as I've already beat my score twice over. More to come! :)
Very cool! Now the challenge is over before it even started 😂
Let's see who can beat this one with a new trick!
I think the trick is to not only have as little as possible on the ISA bus, but to also use a vid card that was advanced enough to have versions that ran on the VLB and PCI buses as well.
This answered a lot of questions! In the 90s I grew up with my dads PC, a 486 of some sort and Doom played pretty smoothly yet I know we were on a budget so NO WAY did he have a fancy VLB system. This must be what he did - overclocking the ISA bus combined with knocking down the window size a couple of notches (we played in high detail mode too). Interesting tests - cheers!
Fun! These videos are like Chicken soup for the soul. Thank you for sharing! I know you're having a blast making these videos!
Thanks for watching! They are indeed a lot of fun :)
My jaw dropped when you got the XX max fps you did haha. Great video, great work on showing what was actually possible with the ISA bus. (had to edit to hide my spoiler, sorry!)
Thanks so much 🙂 .. I've already beat my old best score so there will be a follow-up video coming 😉👍
@@vswitchzerolooking forward to the follow up!
I've been following your channel for a little while now and I enjoy your videos. I worked in a little computer shop in Winnipeg in the mid 90's so I have nostalgic memories of computers at this time too. I recently built a 486 DX-2 66 VLB system and even the VLB graphics card struggles to reach 25fps in the Doom benchmark. These ISA overclock results are very impressive! Still, Doom really isn't my motivation for this build, so I'm not sure it matters all that much. :)
Thanks very much! 👍
Awesome work Mike! You've got me inspired to try this out for this challenge.
Thanks for the shout out Mike! Another great video! Very thorough and cool.
Thanks for the Mach32! For the longest time it was the only ISA video card I had and it got a lot of use :)
Always interesting to really see these old systems pushed. Motherboards are so fleeting these days, only working for a small collection of hardware and here you are pulling 250% overclocks on the system bus with 30+ year old caps and transistors.
Very cool! Never even thought about overclocking isa. I now need to try😁
Oh shit! This is my chance to start my channel! I have a nice 5x86 @40fsb (160mhz) and a Wyse ET4000AX that I know will get above 23fps. Nice! 32mb EDO. 256 12ns cache. Dell XPS 4/66v motherboard (aka Micronics most likely SIS) with VESA. Was 5v but using an interposer (TrinityWorks) with an AMD 5x86 Write Through cache CPU (ADW stepping).
I have found your videos from the past year incredibly interesting and entertaining. Nice to see a fellow Ontarian putting out great retro PC content!
Thanks so much! :)
That's awesome! Enjoyed this video :)
Think I had a 486SX at 25 MHz at the time of Doom's release - later upgraded with a DX2 Overdrive to get 50 Mhz which was staggering - but the IBM PC did have a VLB Cirrus Logic VGA card.
Doom ran fine(ish) even with the SX 25 Mhz as I recall. Ofc with the DX2 Overdrive it ran super smooth, later.
I had the same cpu and yeah it wasnt the bus that was the bottleneck there it was the lack of FPU in the 486sx chip. I had no idea about the buses so this was interesting. I also tried to upgrade to the overdrive chip but being young and literally my first attempt at an upgrade I missed the "ovverdrive" detail part and tried to stick a regular 50mhz dx chip i found locally in it. AFter bending the pins a few times i realized it jsut woundt fit and didnt get it until I got on the phone with IBM and they told about the overdrive chip part, hah, which gave the cpu an FPU unit apparently which was needed to run games well back then before GPU's and I had no idea about the buses other then later when slight overclock there was an easy overclock when you couldnt overclock your cpu. When duke3d came out the 486sx was pretty much toast, doom was still playable but i really woud have liked that upgrade for duke3d. But that didnt stop me from honing my fps chops in doom and actually winning a duke3d tournmanet without actually being able to run the game. I played my friend in the final who i'd watch play duke hehe. He was running it on a cyrix 166 i think it was.
@@mikejones-vd3fg pretty much no game used FPU seriously before Duke Nukem 3D/Quake. Duke only when slopes were on screen, but to run Duke at all you already needed around DX2, and Quake required Pentium for fluid gameplay.
DAMN that was high! And here I'm paranoid about using anything higher than spec.
I remember doing the same thing, not just video cards but also the venerable NE2000 could be overclocked as long as everything on the coax bus was equally overclocked. Most motherboards had a jumper for the bus division intended for different CPU speeds, which if set “wrong”. We always ran games on 12MHz, using the turbo button to switch, but 16MHz wasn’t stable with IDE and serial controllers (leading to corrupt data), but it was with Adaptec SCSI cards.
Great work and very nice videos here! I am subscribed! 👍👍
Thank you very much! :)
It was very card-specific in late 80's/early 90's. I ran 3com NIC's successfully at 12.5MHz and it helped speed-wise with early networking. Trident was notoriously bad for a reason, but Cirrus was good. I used overclocked ISA Cirrus 5422 for a while with 386DX/40, it worked ok with 13.3MHz AT-bus (40/3). Mach32 was one of the priciest cards around (with Matrox) so it was very rare, Cirrus was much common for a reason.
MCA was Very pricey to implement/license - reason it never succeeded. PCI was the real solution.
absolutely nuts to hit the fps cap on doom with such system.
that was some serious overclocking.
Very very interesting. 🤔
My childhood 486 (which I still own) is a 486DX/50 with the FSB running at 50MHz. It was always vexing as a curious kid why I was stuck with ISA graphics since the mainboard had VLB, but it never worked. (To know then, what I know now.)
I'm adding a project to my list: Explore "ISA overclock options". I have a Cirrus Logic 5434 (late model ISA video card to experiment with) and it'd be neat if I can bridge the performance gap with my VLB DX2/66 board.
Great video!
Thanks very much! :)
VLB became bit unstable at 40MHz with only 1 VLB-card. Even running two 33MHz VLB-cards like multi-IO and VGA was slighly unstable with some cards/motherboards. 40MHz was almost guaranteed to fail somehow, mostly IO-problems (like corrupted HD-transfers). Serial ports still worked ok, at least to 115200-speeds when used with 33.6k modems. Remember to overclock your 5434 with MCLK, it made visible difference in some DOS-games with 542x-series cards.
@@tomiluukkonen4035 Indeed, and the board in that system does support the DX2/66, which is what confirmed for me that VLB was capable of working. Once I figured out that I'd been trying to overclock the VLB video card (a Cirrus Logic 5429) to 50MHz as well, my 16yr-old self back then started looking for what I could maybe set to split the clocks... If I down-clocked the DX-50 to 33MHz, it worked but that left quite a lot of performance on the table. Jacking up the wait states didn't work either... 50MHz was just too spicy. (The board doesn't support 40MHz unfortunately. It uses a dedicated 33MHz crystal and a 50MHz crystal)
Now I want to tinker with it again.
@@tomiluukkonen4035 Thanks for the tip on MCLK, I didn't realize those cards could be overclocked. I will have to give this a try.
My 486 motherboard has the same chipset, you're tempting me!!
I remember setting jumpers on my 486 to eke a bit more speed out of my processor, but this is next level stuff - really enjoyable and a little trip down memory lane! My second ever PC was an IBM PS/2 (386) with MCA. Proprietary and expensive.
Would like to see how far a VLB system can go and compare it to these results. I remember being able to push my DX2 80Mhz system pretty much up to full tilt, in terms of cache/memory timings & bus divider, without any issues but even if I had done benchmarks back then, I wouldn't remember the results now. Anyway, very entertaining video, thanks!
I don't know why but seeing ":\DOS H O E H" when the ISA bus was underclocked made me laugh, like DOS is laughing at you or something.
Cool ISA bus over clocking! I've tried doing a little of that on my homebrew 8088 machine as it's ISA bus is clocked directly to the CPU clock and could only get it to 11MHz before my Trident based VGA graphics cards (9000 and 8900D) refused. I think my U3 EGA card could get to about that level too before it glitches out. Thanks for showing the glitch out screen for a graphics card, that's really helpful to see on someone else's machine. My machine also probably doesn't have the best signal integrity so I'm pleased I can clock it that fast. Defo no DOOM on my 8088 (V20) tho I can just about play CGA Wolfenstein 😅
Would also be fun to know if a stock 486DX2 66MHz could get to 25fps with the right ISA bus clock.
Thanks Chris! Yes, I think that would be a much bigger challenge. Doom is very CPU intensive on 486s so that would be interesting to see. I plan to do a follow-up on this one and may give it a try with some older chips too.
great job Mike!
Great stuff. I remember overclocking my ISA-only 486DX50 board. It had some strange SCSI controller card for HDD and I later purchased a Cirrus Logic GD52424 for an already remarkable graphics boost. I was a kid of practically absent knowledge and only had read about overclocking in a magazine but I just had to try setting the relevant jumpers on the board. And indeed I got Doom running significantly faster and so did the SCSI. Of course higher settings would produce graphics garbage but I was still very happy with what I could achieve that way.
GD5424 I think is the correct name of the gfx chipset.
4:50 I do remember this colour scheme but I have no idea which drugs the engineers were on when they chose these...
Interesting :). I wonder if you could overclock the ISA bus on a Slot 1 motherboard (Award Bios) with in this case Pentium a PIII 700 mhz?, in some DOS games it seems with a CPU faster than 550 mhz the sound does not work as it should with for example a Sound Blaster 16, CT2910 in this case, so i wonder if overclocking could help there?? hmmm.
Good question - if the BIOS supports changing the ISA divider it should be possible on more modern systems. I didn't realize that the CT2910 was speed sensitive. It would be interesting to see if speeding up the ISA bus could help.
@@vswitchzero Yes, also the motherboard and BIOS version could play a role?. Newer Windows 95 - 98 and DOS games work fine, it's just some older DOS games that have a problem with the speed of the CPU and the ISA bus it seems. Just to mention a few DOS games, Star Wars X-Wing (there is a fix for that one), Star Wars Tie Fighter (works most times as it should), Space Quest 4, Prince of Persia, Alone in the Dark (intro does not work every time as it should, or some sounds are left out). Keep up the good work :).
Very impressive results
Impressive results. While i dont remember the fps for Doom on my 50mhz bus setup with a 5x86 and a Diamond Stealth 32 VLB card back in the day, i know it was a lot faster running 3x50 than 4x40. I always assumed that was due to memory bandwidth but this makes me wonder if the vesa bus was the limiting factor. Sadly the board died in the late 90s and i got rid of it so i cant go back and check, but i still have the cpu and both vesa cards for video and i/o. I may have to give it a try if i get time and can get the random 386/486 board to play along. I know I'll need to drop to 3v as well.
I would like to see an EISA video card in this way...
Hopefully one day I'll find an EISA board for a decent price. They are getting pretty hard to find these days :)
@@vswitchzeroYeah, I noticed 😢. Perhaps one can find or fab an EISA backplane.
Overclocking was so much more fun back in the day.
What a great video! I used DOS computers in this time but I was not even 10 years old and didn't know almost all of this. As you probably know, Doom is limited to 35fps for normal gameplay (not DoomBench timedemo) so you've built the ultimate Doom machine! What is the CF-IDE adaptor you're using? I've got three and none of them boot in my PII 440BX system.
have you though about building a clock divider for the cards that dont run over 11mhz, 16mhz, 25mhz ? perhaps something that can detect clock no matter where it is, then lower it to the correct clock speed. this would be done slot for slot or card for card maybe, or perhaps somethin that can read clock, and distribute clock to the cards that fault at higher rates. you would do this for any and all isa slot cards you are using, and leave the VLB cards to themselves? perhaps in the case of VLB, ramp the clock way up and lower it in iteration till the clock stabilizes and write to say a small eeprom that connects to say a variable programmable resistor or something of the sort. speed control for the isa, timing control from peak, out of range to highest stable for the vlb.
Thanks for your comment - that would be very interesting! A bit beyond my skill level in electronics but it sounds like it could be possible.
you would have to detect address on the bus and slow (stretch) the clock for all addresses !A0000, I dont know PLLs capable of reacting that fast, fpga listening to address bus could do it. Still leaves the possibility of other cards glitching on faster clock and phantom activating or crashing while VGA is being accessed.
I just tried my non overclocked 133MHz 5x86 PC with an S3 Trio 64 PCI graphics card. It delivered 40.5fps on the default settings. So the ISA bus is still the main bottleneck, but it's not too far off. But on the other hand My fast 486 could get even more frames @3* 50MHz with tweaked cache and ram timings.
Interesting exercise ;)
Awesome video!
I must have had a PCI based 486. I barely remember that. The first system I ever built (with help) was an AMD K5-133. I tried running it at 166, but it was unstable. And obviously that's going to be significantly faster.
Feels strange to have *very* similar hardware on hand.
Balls I may have to try to run a similar build at least temporarily.
I wonder if a mach64 ISA card would help at all or if the mach32 is already saturating the bus completely. I do not have one of those as much as I want one.
Welp. I did it and put it in a video and got better results somehow. And then I did it with an ET4000AX and got a better result than that.
This is my ET4000AX run at a whopping 40 FPS:
ua-cam.com/video/n92DY5z8gb4/v-deo.html
While I appreciate your take on the challenge, using a CPU released in late 1996 seems to nullify the goal of the challenge.
Would be much more interesting to have a rule that specifies that no hardware older than ~ the Doom release date can be used.
I'd rather would like to see how it is possible to achieve the 25fps mark with the cheapest hardware possible that was available at that time and make a score based off that.
Thanks for your comment. I really like the idea of having a cutoff of the doom release date. I plan to do a follow-up on this video and may give that a try. Thanks for watching.
go man go!
Looks like cpu galaxy will handle over to you best 486 on gold trophy to you 😉
Haha I'm sure he's got some tricks up his sleeve. But that said, I have already beat my old score and will be doing a follow-up video at some point :)
Whould changing out some of the bus chips and the roms to faster versions of them maybe make some of the less wieldy graphics cards more compliant?
Good question. Would be interesting to try. I have some plans for a follow-up on this video so stay tuned for some other interesting tweaks potentially :)
I would be interested to see if you could remove the crystal and put a frequency generator on and just keep winding it up until it stops working
I have always been curious and interested in experiments squeezing the Maximum out of an older technology like you just did. Got some questions for anyone reading this:
Was it possible to overclock MCA and EISA back in the day? could they have came close (or even keep up) with VLB and the first PCI?
Would they have been able to power some early 3D Accelerators from 1994-1996 (cards for both CAD and Gaming) without significant bottlenecking?
One of those boards had a whole bunch of 74ls chips for interfacing to the isa bus, I wonder if that was at all the limitation as I seem to recall the ls family kinda tops out around that speed anyway. Could be a fun experiment to try a newer ttl-compatible family and see what happens 🤔
Interesting! Will definitely look into this.
I used to OC the ISA bus on my 486SLC way back in the day. It didn't buy me nearly as much, as that chip was slow as hell, just like the rest of the machine.
I wish I had known about overclocking the ISA bus back in the '90s while we were suffering with the budget 486sx25 from Sams. It would have totally been worth facing my father's wrath if caught tinkering with the board if it meant DOOM would run faster.
ISA has 8 MB/s bandwidth. 320x200x8 frame takes less than 64KB and it takes less than 4.5 MB/s of bandwidth to update 320x200 screen 70 times per second. DOOM (in-game) has 35 fps limit, so it should need less than 2.5 MB/s of bandwidth.
However it seems that graphics cards are limited to about 1.3 MB, maybe a bit more since not every pixel is drawn once (yet DOOM engine is optimized to avoid overdraw).
Anyway now I'm interested how fast real bandwidth is and how it is divided between multiple ISA devices.
This gave me two thoughts. One was that obviously some devices deal with an overclock better than others. So is it possible to have some sort of reclocking ISA to ISA bridge and two speed domains, one running nominal speed for most devices and one overclocked for just a handful devices which would survive that?
And the other that ISA streaming digital audio is a weird beast. For cost, soundcards do not contain a sample buffer of more than a couple samples in capacity, certainly not a whole page full. For simplicity's sake, just about none of the ISA soundchips support true Bus Master, where they would be able to request data from system memory as needed. Instead they fire an interrupt and have software cook a page of audio data and set up DMA to supply samples when they're needed. DMAC runs at a low frequency i think 3.5MHz ish nominal (i forget, my memory is vague) and while it performs operations, slows down the whole ISA bus - which is also why DMA video wasn't viable and ISA DMA disk access would be a horrible idea as well, it would be too slow. This lead me to think, why would digital audio become unstable when you overclock ISA? Perhaps the whole problem is that the DMAC speed isn't set correctly, and ISA bus speed that you set explicitly isn't even the issue?
Thanks for the comment! I wondered that too - like a frequency divider for more fussy devices. I don't really know enough about the ISA bus' inner workings to know how practical that could be to implement but an interesting idea none the less. Interesting about the DMAC clock frequency too - I hadn't thought of that. I'd be curious to see if it gets thrown out of spec along with the ISA bus but not sure how to measure it.
@@vswitchzero Let's assume that a mainboard exists which has an explicit clock signal for ISA DMAC and an explicit clock signal for ISA, something perhaps less integrated, maybe discrete 8237a; perhaps you can use that sort of board to piggyback an Overdrive/Accelerator system of some sort on it; perhaps something where the RAM isn't on mainboard at all but on CPU accelerator board? Not that i even recall anything like that exiting. I have a strong suspicion that it will not lead to a very successful ISA overclock since even if you have a suitable topology, you'll have low speed grade components to deal with. And yet i'd LOVE to be proven wrong and it's just fun to think about. What if.
A Dangerous game, i might add. Dont try this with VLB grafic cards... i had success up to about 12MHz... Most VLB cards wont handle that (my 801/805 S3 did..)
I saw many VLB grafic cards die because of incorrectly set ISA dividers. Be careful...
What i can remenber is, that it is very efficient to get a board where the DMA waitstates can be manipulatet. (AMI BIOS, mostly a hidden option)
Setting this to 0 will increase the performancy significantly, especially with overclocked ISA
I remember having a 286/16 and trying a 1:1 bus ratio. My CGA card burst into flame! But once I'd patched around the damage (which was oddly at the output, so it may have had nothing to do with the ISA bus) the whole thing worked fine and was quite a bit faster! I didn't have the nerve to leave it unattended like that though, after what had happened to the CGA card.
18:30 You probably need to increase the cpu voltage by 10%, and put an active fan on it at low rpm (12v case fan at 5v input).
Now this is retro gaming!
So still need to know if you could improve the speed on your DEC 486?
I'd love to do a video on it one of these days. After getting it up and running years back, I haven't really used it much.
as the ISA Bus is specified up to 12 MHz a overclocking to 25 MHz is just a little bit more than 100% overclocked not 200%.
Interesting, thanks for your comment. I always wondered this because some older "turbo" 286 systems will run the ISA bus greater than 8MHz, but I was always under the impression that the specification called out 8.33MHz as the maximum "official" frequency for ISA BCLK. Will have to take a look at the whitepaper :)
I've used the COOSPIDER CTUV-6 to erase EPROMs and had really good results. Way cheaper than the purpose built units for erasing EPROMs. Please, at the minimum use a cardboard box to house the light, the UV is not good for you.
Did you try to upgrade the clock crystal of the ACER I/O board to see if it would work at hight bus speed if the board was actually also running at a higher freq?
Could be an interesting experiment to try! I hope to revisit this at some point.
AGP is where the graphics started to make sense for me :)
This video inspired me to tinker with my 386dx-40 for this challange. I don't expect to get anywhere near 25fps but will be intresting to see what I can achieve. Unfortunatly my ATI Mach8 card seems to have bad ram. I'm currently searching my parts stash for the best alternative i have on hand. I do wish i had a scope to verify bus clock though.
Nice scope, do you recommend it?
Thanks for your comment. For my very basic needs, it's been a good scope (especially for a little over $100). I'm sure there are much better ones out there if you have a higher budget, but I've been happy with it so far.
@@vswitchzero Thanks for the reply! My needs are pretty basic (checking cpu vpp) and I think that one will do the job!
BTW, googling "isa bus overclock" returns this video on the first place!
🏆
Well, that was easy :D
I didn't know DEC made anything other than VAX and Alpha workstations,and servers
Can I use my Orchid Fahrenheit 1280 32-bit VESA Local Bus system?
wonder if a 64mhz crystal could push it further.
noice
6:21 "AMD AM5..."
me: hold up
👍
Of course it almost rivaled VLB cards, VLB cards run at the same speed as the CPU FSB, as such DX/2 50Mhz have a 25Mhz FSB, DX/2 66 are at 33Mhz and DX/2 80 are at 40Mhz. ONLY the 486DX 50Mhz has a FSB at 50Mhz (FROM THE FACTORY, no overclock). Aside the 486DX50Mhz (The first CPU to require active cooling to remain stable in fact)
Only the DX40, DX/2 80 and DX/4 120 had FSBs faster than 33Mhz and only a single 486 model ever had a native default 50Mhz FSB sadly.
Now, I would recommend you to get a rarity of a motherboard, a VLB 486DX motherboard that has a Weitek FPU socket (yes it can run at the same time as the 487 as it is memory mapped ) they exist in fact, and then do yourself a favour and get AMDs 386DX 40Mhz and a 387 40Mhz, then test this. Please use a graphics card which has the chipset available in both ISA and VLB models if you truly want to compare the performance difference between these 2 interfaces. Note that VLB first and foremost is a DIRECT TO CPU link and if that is not enough it also is natively 32bit :P
you should try mach64?
it would be interesting to see what this system would be able to do on newer games. It's got a AM5x86, so it should be able to """play""" Quake.
How much data is going through 25mhz isa bus
Good question - not sure how one could measure that.
@@vswitchzero I'v seen a benchmark video comparing speed of isa and pci
@@vswitchzero just subbed btw i wonder if u can push the long isa lvb slot by 210 percent
@@vswitchzero heres my argument towards what you said id say if some one had that board and still had old isa card that would be because they spent all their money on the mobo and is waiting to save up to get new vlb card
@@vswitchzero acording to google 5.3MB/s 8.33 mhz 16 bit slot which means 100 mhz overclocked bus would make 64.5 MB/S it would take
8.33x12.10% = 100.793 mhz 12.10 = 1210 %
Your laptop looks like a Lenovo X280
It is indeed, good eye! :) .. great little laptop.
then the doom guy died
been bussin'
how about isa-to-pci bridge