Well, we are actually talking quite often as we are basically live in the same city in Austria... But ok... I do not think this could ever happen. I think he told me he had to bin over 50 CPUs to find this 200MHz one. So based also on my tests.. it is safe to say that these are very rare. Also my M/B might be modded as hell, but in the end his M/B is better as it supports EDO ram that is a bit faster. Borrowing his CPU is improbable, but we might cooperate in the future. Maybe I mod his M/B :p
@@atheatos yes, volt mod and pll mod. His graphics card should also handle it. How would it stack up to against some of the recent records posted on Vogons?
As for right now we are probably a bit behind compared to what is posted on Vogons. We might have a chance to catch up. I also have now more M/Bs and mods to test. However the real problem is what we are now just super busy with our day jobs. So yeah...
Yeah, love the doom benchmark. :D I also experienced such behavior with unstable 386. (There were some kind of cache issues with it back then) NB, it is 190% overclock. Don't forget the original speed. With no overclock, you don't say it has "100% overclock". ;) Still, impressive.
Yeah, I also think that it is the L1 cache falling here. Also... you right I messed this OC % up :/ Thank you. It is 2.9x times the speed or 190% overclcock.
I have exact same benchmark results with: Asus PVI-486SP3 @66MHz FSB, i486DX4/100 @2x66=133MHz, Cirrus 5446 PCI, 256KB cache, 32MB FPM stick. Needed to slow down cache from bios, RAM speed is actually the same in my other system: QDI V4S471, Cx5x86/100 (2x50MHz), all other settings tightest but ram set to faster. So loosening bios settings really kills the benefit of higher bus speed.
Yeah very similar conclusions to me. Getting good cache timings at high speeds is very had, and usually impossible. On this one board I also tried many different cache chips and all were actually slower. I need to try other boards too.
I love all your stuff. CPU Galaxy showed a cache interposer that went in between the socket and a 386 CPU. Is that someone a wizard like yourself could replicate with crazy tight timings and megabytes of cache or is PCBWay not up to something like that?
Sure :) I am already working on things like that. In the general case, PCBs are not the problem. But Yeah, some CPU interposers can be very hard to route. The problems are mainly with the M/B chip-set and cache coherency in general. I am trying to figure out all that now.
I met him 3 weeks ago. He is fine, just too busy. They gave him extra stuff to do at work. He also said, that he is repairing 2-3 new videos. With TH-cam it is very hard sometimes. :/ I had also a long Hiatus.
you should try replacing the crystal oscilator for a pll or an arbitrary function generator to push it higher than that, say do 2x 75mhz or so, see how far you can get with the bus and ram. Also if you are up to it, you could try separating the 5V rail from the ram (worst case you will have to rip a few pins and solder it directly....) Voltmodding the ram there should really help get the higher bus clocks stable imo.
My any_clk device is exactly that, a PLL, that replaces the default crystal oscillator. It can generate more or less any frequency, ok with some steps (
@@atheatos Well if its 70mhz max theres not much else to do. Maybe you can push it up to 6v. Not sure how good was the regulation in the XT / AT standards if any. Maybe you can play with the tolerances. Usually 5V TTL stuff can do 7 to 9 volts before breakdown and remember old machines had horrible regulation causin all sort of glitches with sleep states so much that windows didnt put the cpu to sleep in some machines to about the noise spikes caused by the change in power consumption. I bet the regulation and noise tolerance are quite lax
Yeah for sure TTL chips can handle even 7v. However the chips here have mostly CMOS specs. So only up to 5.5v it is with in the spec. Now 6v might be safe, but again I do not want to risk it. There are so many chips on the board, if one fails, it is game over. Also until now no one else has done above 66-67MHz. So yeah :)
Sure I can now do this, something like 2 x 66. Then for the Pentium I need a Socket 4 overdrive, that is rare and I do not have. So next option is just a Socket 5, there I need to find a similar chip-set M/B. (Maybe 3 x 60) Interesting idea, I might try this. Thank you.
Well sure it is cheating. Maybe not useful for speedrunners. The glitch basically messes the timedemo commands and usually results to a better result, that is for sure invalid.
Oh interesting, I know it's off-topic for the video but I noticed the DOM power mod you did. That's pretty genious!
Thank you!
I think compared to other mods here, maybe not that significant.
That is why I never mention it in any video and just did it.
Maybe CPU Galaxy could borrow his 200MHz chip 😃
Well, we are actually talking quite often as we are basically live in the same city in Austria...
But ok... I do not think this could ever happen.
I think he told me he had to bin over 50 CPUs to find this 200MHz one.
So based also on my tests.. it is safe to say that these are very rare.
Also my M/B might be modded as hell, but in the end his M/B is better as it supports EDO ram that is a bit faster.
Borrowing his CPU is improbable, but we might cooperate in the future. Maybe I mod his M/B :p
@@atheatos yes, volt mod and pll mod. His graphics card should also handle it. How would it stack up to against some of the recent records posted on Vogons?
As for right now we are probably a bit behind compared to what is posted on Vogons.
We might have a chance to catch up. I also have now more M/Bs and mods to test.
However the real problem is what we are now just super busy with our day jobs. So yeah...
Yeah, love the doom benchmark. :D I also experienced such behavior with unstable 386. (There were some kind of cache issues with it back then)
NB, it is 190% overclock. Don't forget the original speed. With no overclock, you don't say it has "100% overclock". ;)
Still, impressive.
Yeah, I also think that it is the L1 cache falling here.
Also... you right I messed this OC % up :/ Thank you.
It is 2.9x times the speed or 190% overclcock.
I have exact same benchmark results with: Asus PVI-486SP3 @66MHz FSB, i486DX4/100 @2x66=133MHz, Cirrus 5446 PCI, 256KB cache, 32MB FPM stick. Needed to slow down cache from bios, RAM speed is actually the same in my other system: QDI V4S471, Cx5x86/100 (2x50MHz), all other settings tightest but ram set to faster. So loosening bios settings really kills the benefit of higher bus speed.
Yeah very similar conclusions to me. Getting good cache timings at high speeds is very had, and usually impossible.
On this one board I also tried many different cache chips and all were actually slower. I need to try other boards too.
I love all your stuff. CPU Galaxy showed a cache interposer that went in between the socket and a 386 CPU. Is that someone a wizard like yourself could replicate with crazy tight timings and megabytes of cache or is PCBWay not up to something like that?
Sure :)
I am already working on things like that.
In the general case, PCBs are not the problem.
But Yeah, some CPU interposers can be very hard to route.
The problems are mainly with the M/B chip-set and cache coherency in general.
I am trying to figure out all that now.
What ever happened to CPU Galaxy? I just checked and he hasn't made a video for about six months. Love his channel.
I met him 3 weeks ago. He is fine, just too busy.
They gave him extra stuff to do at work.
He also said, that he is repairing 2-3 new videos.
With TH-cam it is very hard sometimes. :/
I had also a long Hiatus.
@@atheatos Thank you for the update! I am glad it's nothing serious like health issues or the like. :)
you should try replacing the crystal oscilator for a pll or an arbitrary function generator to push it higher than that, say do 2x 75mhz or so, see how far you can get with the bus and ram. Also if you are up to it, you could try separating the 5V rail from the ram (worst case you will have to rip a few pins and solder it directly....) Voltmodding the ram there should really help get the higher bus clocks stable imo.
My any_clk device is exactly that, a PLL, that replaces the default crystal oscillator.
It can generate more or less any frequency, ok with some steps (
@@atheatos
Well if its 70mhz max theres not much else to do.
Maybe you can push it up to 6v. Not sure how good was the regulation in the XT / AT standards if any. Maybe you can play with the tolerances. Usually 5V TTL stuff can do 7 to 9 volts before breakdown and remember old machines had horrible regulation causin all sort of glitches with sleep states so much that windows didnt put the cpu to sleep in some machines to about the noise spikes caused by the change in power consumption.
I bet the regulation and noise tolerance are quite lax
Yeah for sure TTL chips can handle even 7v. However the chips here have mostly CMOS specs.
So only up to 5.5v it is with in the spec. Now 6v might be safe, but again I do not want to risk it.
There are so many chips on the board, if one fails, it is game over.
Also until now no one else has done above 66-67MHz. So yeah :)
Interest to compare with Pentium at same bus speed and CPU clock.. :)
Sure I can now do this, something like 2 x 66.
Then for the Pentium I need a Socket 4 overdrive, that is rare and I do not have.
So next option is just a Socket 5, there I need to find a similar chip-set M/B. (Maybe 3 x 60)
Interesting idea, I might try this. Thank you.
@@atheatos Yea, compare 486, Pentium OverDrive and Pentium Soket 5 on same frequencies and memory type :)
If speedrunners would use an overclocked unstable 486 to glitch Doom for an advantage - would that count as cheating? lol
Well sure it is cheating. Maybe not useful for speedrunners. The glitch basically messes the timedemo commands and usually results to a better result, that is for sure invalid.