This video takes me back to my original K6 days. I had a K6 200 Mhz CPU and used a Megatrends HX83 mainboard. I wasn't able to overclock the K6 very far past 200 Mhz using that board but I was able to use the 83 Mhz FSB speed with a 2.5x (instead of 3x) multiplier and ended up with a 208 Mhz K6 chip. Even though the CPU was only running an extra 8 Mhz faster than stock the faster FSB made noticeable gains in many games.
If you were lucky and brave enough to tinker with your hardware back then, you could easily get more performance out of those systems by changing the FSB to a higher clock. As you said, the 8 extra MHz are almost irrelevant - the main benefit comes from the faster FSB.
I had read about some of that stuff back then but i was pretty young and didn't understand it enough hah. I also didn't want to fry the PC because I may not have gotten a new one.
So far I've modified two K6-2+ 570s and one K6-2+ 533, and all three have been stable. Both of the 570s are in MVP3 Super 7 motherboards running at 616 MHz (5.5x112) and the 533 is running at 500 (6x83) in an Asus TXP4. I had a genuine K6-3+ 450 that I was running at 500 in the TXP4, but when we discovered the resistor mod on Vogons I decided to modify the 533 since I wasn't using it, and I sold the K6-3+ 450.
I am surprised that this mod is so successful. I would expect some not to work! Usually I am the one who gets the dud or breaks everything while trying. Congratulations though! I still look for a well priced K6-3+ just for collection purposes. Thanks for your comment!
My personal experience with the K6-2+ was that pretty much all of them could hit 500 MHz but few of them could hit 550 -- and those that could, were sold as such.
Hello, have you found any info about any utility to check the various levels of cache? I only use the CACHECHK V7 (included in the Phils Computer Lab benchmark pack) but I also tried Celem Cache Test v2.69 and Checkit (at Winworld) to test if my combination of EDO RAM was making weird things to my M919 Level2 Cache card.
these from our exp, k6-3 or 2+ will often do 650mhz after a delid, the TIM they used was BAD, the IHS is not great either honestly.. if you get ahold of another or more then one, try making a shim and doing direct die, you will be shocked how much lower the temps are, opening up higher clocks, the best we managed on one was a buddies that was HORRIBLE he go it free from a shop because it ran hot regardless, we removed the lid, no tim... used some quality stuff we already had, and... the dif was shocking.. his brother got us some copper IHS replacements for those...not sure if he bought or made them at work or what but...the dif was shocking... genuinely, 766mhz on one 100% stable using a silly arse thick copper base plate with alum pins for fins cooler that, well, the systems still functional, he did upgrade the board though, he got ahold of a genuine gem of a board that will do 150buss stable, currently hes using 133 same as we managed on that other board, that was also a last gen super7 board... this ones got a better chipset though, handles clocking better then the via set did and better memory and ide perf when you run stuff out of sync.. something the sis socket A chipsets also share... honestly, this old stuff is fun, some boards like the one im talking about, are insane for how advanced they had become for their day, the pci/agp/isa are isolated from the cpu fsb such that, you can tweak them independent of the main fsb, much like far newer systems... the only downside was he had to re-cap the board and replace the old dallas rtc with abutton cell replacement... hes got all 3 of our old main chips still, only the one IHS hes got in use though, i asked, and the board he swapped out is in process of being recapped by somebody for him, in trade for allowing the fellow to clone the bios chip using his reader, as his board has a much much newer bios then any online archive has saved, that wont backup with software for some reason... we also each had dual 370 boards with 466 chips clocked up, boards would do 150buss easy... our buddy who kept his also has ours that he recapped with audio grade caps and ones his main "retro game rig" as its got dual 1.4ghz celeron p3 tu chips, the only dif between the ones he got vs the full p3 was the fsb being 100 by default not 133, and..well.. both chips have been doing 2.1ghz for years now, no problems, he even had some server ram given to him that, was both faster then what he was using and, 4x typical capacity... and worked in the board with the last bios abit released.... genuinely crazy for how old it is, how well it runs modern linux... hes actually got it setup in a uefi loader based multi boot with win95, 98se, 2k, xp, and a linux distro he really likes that still keeps a 32bit branch around and working, for its era.. it destroys any p-4/ or p-d you put it up against... it did take custom clips to attach the coolers with the TU adatpers on the board sockets, but... well.. it already had custom to deal with the pelters double stacked under the coolers on the old chips... so..yeah... it really wasnt an issue... he did endup water cooling it since he was able to modify clips for a water block to work on the sockets with the pelters without much effort, to this day its a beast.. dual voodoo2, dual matrox m3d, best agp card they ever put out that would work in the board, think its a 1950 or something... either way... its a genuine beastly rig for its era... and able to play pretty much any title you throw at it... in style.. (i mean of the era...)
k6-2 where a nice chip :) I just moved jumpers on mobo. and all worked perfect. on my k6-2 300@500mhz cpu. I did not know of the L2 cache on on that series. but did alot of tweaks on later athlons. loved my Barton xp2500@xp3200 :) now we cant tweak as much. so the fun is gone. Maybe it time to go down the basement and have a look at the old stof :)
I did not do much research, but I found a forum post that said "Limit MemTest86 to 256KB" (or whatever your cache size is). But I don't think that this actually works because I have read another post that says MemTest is programmed in a way to avoid the cache as much as possible. But that could just mean that the cache is ineffective and used even more. I guess what I am trying to say is that cache allocation is the job of the CPU and you won't be able to decide what goes into the cache and what not. You may be able to influence the effectiveness of caches by utilizing certain programming techniques, but at the end, the CPU decides what goes into the cache. For now, I have not seen any test that targets the cache. I like the question and I have spent hours before trying to come up with an answer.
They become rare - there is a seller on ebay at the moment in the US selling a batch of 3 new old stock. It is around 30 USD per CPU, but you end up with two additional ones.
I have a asus P5a-b super socket 7 board. I might be willing to part. its in my old retro rig atm. might be able to come to some agreement if your interested. I also have a few k6 cpus and some intel ones. I thought about ebaying them. I need to come up with a little cash for bills smh.
I actually have bought that exact board in revision 1.04 a few days ago. Unfortunately, it is going to be shipped within Germany only and I will only get it in a few months.
@@bitsundbolts I understand that. I live in the US so idk how long shipping would take. I think mine is a 1.04rev. I love that it has at and atx power connections. If u don't mine me asking how much did it cost you? I've seen these boards vary wildly in price. Its gonna be fun seeing the same board I have being used.
I paid 35 Euros for the board including shipping. There is also a K6-2 with cooler included. I think I got it it at a really good price. Probably can be sold for double the money. It's good for me because I pay everything out of pocket for all the retro hardware I acquire at the moment. I like buying in Germany right now because the US Dollar is so strong vs Euro!
@@bitsundbolts I seen few of these boards sell for $200 to $450 at the highest. depends on the revision I guess. I know the 1.04 is ok for the most part. lots of features but I think ram speed slows down if you go over 768mb of ram. I can't remember exactly being a few years since i did any research on the thing. I just know i seen that amount mentioned and some limitations.
I'll definitely repeat that research and you will be able to see it on my channel. Probably cacheable area issues. Those boards become really expensive. I will get a DFI K6BV3+/66 which I will use in the meantime. It was quite expensive, but I can't continue with the ASUS P55T2P4 for the next projects. This DFI board cost me 140 USD incl. shipping. Was the only way to get a super socket 7 board reasonably fast.
While testing with Unreal Tournament is more fun, you should run Prime95 for a few hours to verify stability. The Torture Test will stress the CPU, cache and memory, and it will notify you if something goes wrong. Older versions like v20 will run on Win98.
MemTest via BootDisk also works very well. You can set the exact area of the CPU or motherboard cache and search for errors there. I could easily find out that the external cache on a Super7 motherboard was faulty at a FSB of 120Mhz or more.
This video takes me back to my original K6 days. I had a K6 200 Mhz CPU and used a Megatrends HX83 mainboard. I wasn't able to overclock the K6 very far past 200 Mhz using that board but I was able to use the 83 Mhz FSB speed with a 2.5x (instead of 3x) multiplier and ended up with a 208 Mhz K6 chip. Even though the CPU was only running an extra 8 Mhz faster than stock the faster FSB made noticeable gains in many games.
If you were lucky and brave enough to tinker with your hardware back then, you could easily get more performance out of those systems by changing the FSB to a higher clock. As you said, the 8 extra MHz are almost irrelevant - the main benefit comes from the faster FSB.
I had read about some of that stuff back then but i was pretty young and didn't understand it enough hah. I also didn't want to fry the PC because I may not have gotten a new one.
I can relate all too well to your comment!
So far I've modified two K6-2+ 570s and one K6-2+ 533, and all three have been stable. Both of the 570s are in MVP3 Super 7 motherboards running at 616 MHz (5.5x112) and the 533 is running at 500 (6x83) in an Asus TXP4. I had a genuine K6-3+ 450 that I was running at 500 in the TXP4, but when we discovered the resistor mod on Vogons I decided to modify the 533 since I wasn't using it, and I sold the K6-3+ 450.
I am surprised that this mod is so successful. I would expect some not to work! Usually I am the one who gets the dud or breaks everything while trying. Congratulations though! I still look for a well priced K6-3+ just for collection purposes. Thanks for your comment!
My personal experience with the K6-2+ was that pretty much all of them could hit 500 MHz but few of them could hit 550 -- and those that could, were sold as such.
Hello, have you found any info about any utility to check the various levels of cache? I only use the CACHECHK V7 (included in the Phils Computer Lab benchmark pack) but I also tried Celem Cache Test v2.69 and Checkit (at Winworld) to test if my combination of EDO RAM was making weird things to my M919 Level2 Cache card.
i had a k6-3. great chip for its day
these from our exp, k6-3 or 2+ will often do 650mhz after a delid, the TIM they used was BAD, the IHS is not great either honestly.. if you get ahold of another or more then one, try making a shim and doing direct die, you will be shocked how much lower the temps are, opening up higher clocks, the best we managed on one was a buddies that was HORRIBLE he go it free from a shop because it ran hot regardless, we removed the lid, no tim... used some quality stuff we already had, and... the dif was shocking.. his brother got us some copper IHS replacements for those...not sure if he bought or made them at work or what but...the dif was shocking... genuinely, 766mhz on one 100% stable using a silly arse thick copper base plate with alum pins for fins cooler that, well, the systems still functional, he did upgrade the board though, he got ahold of a genuine gem of a board that will do 150buss stable, currently hes using 133 same as we managed on that other board, that was also a last gen super7 board... this ones got a better chipset though, handles clocking better then the via set did and better memory and ide perf when you run stuff out of sync.. something the sis socket A chipsets also share... honestly, this old stuff is fun, some boards like the one im talking about, are insane for how advanced they had become for their day, the pci/agp/isa are isolated from the cpu fsb such that, you can tweak them independent of the main fsb, much like far newer systems... the only downside was he had to re-cap the board and replace the old dallas rtc with abutton cell replacement... hes got all 3 of our old main chips still, only the one IHS hes got in use though, i asked, and the board he swapped out is in process of being recapped by somebody for him, in trade for allowing the fellow to clone the bios chip using his reader, as his board has a much much newer bios then any online archive has saved, that wont backup with software for some reason...
we also each had dual 370 boards with 466 chips clocked up, boards would do 150buss easy... our buddy who kept his also has ours that he recapped with audio grade caps and ones his main "retro game rig" as its got dual 1.4ghz celeron p3 tu chips, the only dif between the ones he got vs the full p3 was the fsb being 100 by default not 133, and..well.. both chips have been doing 2.1ghz for years now, no problems, he even had some server ram given to him that, was both faster then what he was using and, 4x typical capacity... and worked in the board with the last bios abit released.... genuinely crazy for how old it is, how well it runs modern linux... hes actually got it setup in a uefi loader based multi boot with win95, 98se, 2k, xp, and a linux distro he really likes that still keeps a 32bit branch around and working, for its era.. it destroys any p-4/ or p-d you put it up against... it did take custom clips to attach the coolers with the TU adatpers on the board sockets, but... well.. it already had custom to deal with the pelters double stacked under the coolers on the old chips... so..yeah... it really wasnt an issue... he did endup water cooling it since he was able to modify clips for a water block to work on the sockets with the pelters without much effort, to this day its a beast.. dual voodoo2, dual matrox m3d, best agp card they ever put out that would work in the board, think its a 1950 or something... either way... its a genuine beastly rig for its era... and able to play pretty much any title you throw at it... in style.. (i mean of the era...)
Thanks for sharing!
I need to get a decent socket 7 board. I've got 2 or 3 k6-2s I could try this with
Wonderful! Hope to see that baby in some more advanced motherboard ;)
Absolutely! But it may take a few weeks until I get a Super Socket 7 - stay tuned and thanks for being here!
Excellent video
k6-2 where a nice chip :) I just moved jumpers on mobo. and all worked perfect. on my k6-2 300@500mhz cpu. I
did not know of the L2 cache on on that series.
but did alot of tweaks on later athlons.
loved my Barton xp2500@xp3200 :)
now we cant tweak as much. so the fun is gone. Maybe it time to go down the basement and have a look at the old stof :)
Is there any way ACTUALLY test the cache health?
Something like how MemTest86 is for RAM, but for cache ??
I did not do much research, but I found a forum post that said "Limit MemTest86 to 256KB" (or whatever your cache size is). But I don't think that this actually works because I have read another post that says MemTest is programmed in a way to avoid the cache as much as possible. But that could just mean that the cache is ineffective and used even more.
I guess what I am trying to say is that cache allocation is the job of the CPU and you won't be able to decide what goes into the cache and what not. You may be able to influence the effectiveness of caches by utilizing certain programming techniques, but at the end, the CPU decides what goes into the cache.
For now, I have not seen any test that targets the cache. I like the question and I have spent hours before trying to come up with an answer.
Prime95 can test the cache, Karhu's RAM Test tool too. But I don't know if they run on such old Hardware/Os.
That same version of unreal tournament I have runs on win11 with 12700k just fine haha crazy.
I have a Super Socket 7 ATX board. But cant find any "+" CPU for a reasonable price.
They become rare - there is a seller on ebay at the moment in the US selling a batch of 3 new old stock. It is around 30 USD per CPU, but you end up with two additional ones.
@@bitsundbolts do you have his handle/name? I was looking for that seller too
@@bitsundbolts just means you pick the best to run with all the time, have a spare and OC the 3rd in to the ground :D
Here: premierchoicecomponents
That's pretty much what is going to happen with my batch! But I have planned something special for the chosen one!
I have a asus P5a-b super socket 7 board. I might be willing to part. its in my old retro rig atm. might be able to come to some agreement if your interested. I also have a few k6 cpus and some intel ones. I thought about ebaying them. I need to come up with a little cash for bills smh.
I actually have bought that exact board in revision 1.04 a few days ago. Unfortunately, it is going to be shipped within Germany only and I will only get it in a few months.
@@bitsundbolts I understand that. I live in the US so idk how long shipping would take. I think mine is a 1.04rev. I love that it has at and atx power connections. If u don't mine me asking how much did it cost you? I've seen these boards vary wildly in price. Its gonna be fun seeing the same board I have being used.
I paid 35 Euros for the board including shipping. There is also a K6-2 with cooler included. I think I got it it at a really good price. Probably can be sold for double the money. It's good for me because I pay everything out of pocket for all the retro hardware I acquire at the moment.
I like buying in Germany right now because the US Dollar is so strong vs Euro!
@@bitsundbolts I seen few of these boards sell for $200 to $450 at the highest. depends on the revision I guess. I know the 1.04 is ok for the most part. lots of features but I think ram speed slows down if you go over 768mb of ram. I can't remember exactly being a few years since i did any research on the thing. I just know i seen that amount mentioned and some limitations.
I'll definitely repeat that research and you will be able to see it on my channel. Probably cacheable area issues. Those boards become really expensive. I will get a DFI K6BV3+/66 which I will use in the meantime. It was quite expensive, but I can't continue with the ASUS P55T2P4 for the next projects. This DFI board cost me 140 USD incl. shipping. Was the only way to get a super socket 7 board reasonably fast.
While testing with Unreal Tournament is more fun, you should run Prime95 for a few hours to verify stability. The Torture Test will stress the CPU, cache and memory, and it will notify you if something goes wrong. Older versions like v20 will run on Win98.
I have read this, you are right. I think I will try Prime95 with an upcoming project. I will be doing the same modification on a K6-2+ 570.
MemTest via BootDisk also works very well. You can set the exact area of the CPU or motherboard cache and search for errors there. I could easily find out that the external cache on a Super7 motherboard was faulty at a FSB of 120Mhz or more.
I read contradicting information regarding MemTest. There is definitely more research required.
or... upgrade your PC to modern standards with a Ryzen 5600 or Intel 12400.....