Is there a good explanation for why AMD and Intel were running chips at significantly higher than necessary voltages back in the day? I had no idea, undervolting really wasn't on my radar at all back when these chips were contemporary and we just wanted max performance, if anything overvolting was something we were more aware of and in some cases doing. I know with variables at play they maybe picked a 'safe' voltage and didn't want users playing with the minimums but still, it looks like they were using quite significantly more power than strictly necessary.
Good question. I don't know. Will pin your comment, maybe someone has an answer. The gap between stable and stock voltage seems to be too large for the silicone quality argument.
I would assume that some dies would run stable at certain voltages, while others wouldn't. You'd either have to decrease yield, or do some type of binning if you wanted to utilize some dies being stable at lower voltages. Here, the sample size of dies is literally just one, so who knows how the majority of dies, especially the early ones, did perform. And even for this single die, you can see that a stock core voltage of 1.8V might have been marginal, so 2.0V seems like a good and safe choice.
Probably because lower quality PSU's , single stage VRM's on motherboard etc, etc. which both contributed to higher ripples and noise in voltage, which affects stability. By increasing voltage you increase gap between maximum voltage that can be recognised as logical "0" and minimum voltage that can be recognised as logical "1". They could not guarantee motherboard quality back in the days, but they could increase voltage specification for CPU and have less unhappy customers returning their product. Additionally not every motherboard could be set for voltage this low with 0,1v steppes. By marking CPU voltage with 2 volts they made their product compatible with most motherboards available.
That overclock at 1.7v has not been properly tested in this video, and imho it is probably not stable. The prime95 is a steady high load, and NFS3 with such a slow graphics card being the bottleneck running at what looks like 20 fps is not stressing the CPU at all. To properly test this old CPUs you need to use variable loads and high fps gaming for hours. I had a K6-3+@600 with voodoo3, and it was stable with prime95, games, etc, till one day i decided to play NFS2-SE from start to finish, that game runs at like 160 fps in my CRT with that configuration, after 2-3 hours of gaming the instability started to kick in, and finally I had to reduce a bit the CPU clock.
It was most likely done to improve yields during manufacturing. Dies are manufactured on wafer (a large disk of silicon that can fit many dies). So lots of CPUs are manufactured simultaneously. But the silicon disk (wafer) isn't perfect and has random flaws in its crystal structure. Plus contamination occurs. Plus flaws happen during manufacturing. So dies on the same wafer (which were intended to be identical) will have various degrees of flaws which will cause various degrees of instability. Making these dies not identical...and only a tiny fraction of dies will come out perfect out the other end of this manufacturing process. Way back, such flawed dies were disposed of. But the cost of the wafer and manufacturing remains the same, and if you intended to make profit, the few dies that you sell have to carry that cost. But, many of the imperfect dies will still work, if you do something to increase stability. E.g. 1. Lowering the clock 2. Disabling affected features like cache. 3. Or increasing the voltage. You can now sell these "failed attempts" at lower prices and recoup some more cost of manufacturing...and make more profit or reduce the selling price of the perfect dies to be more competitive. But what also typically happens is, as manufacturing of a specific CPU architecture and the process node goes on, little issues that cause flaws are solved and the process is said to "mature". And yields increase (a higher fraction of dies are perfect because flaws are less frequent) But different consumers have different budgets and this results in different markets. High end markets, medium, low. As manufacting goes on, supply for the high end market increases due to increased yields, while supply for the low end market decreases (because there's fewer badly flawed chips to clock down, disable features on, etc). Reducing supply is bad, because the consumer will simply switch to a competitor who does have supply. So CPU manufacturers will use perfectly good dies from the higher stockpiles for lower markets, just so they don't lose market share in a specific market. And during these later more mature stages of manufacturing many consumers are unknowingly blessed with CPUs that are capable of much more than they are packaged and labeled as. These CPUs will be great overclockers due to the extra stability overhead. The same is done with GPUs. Most GPUs and CPUs are simply failed attempts that are salvaged and sold at lower prices. While the perfect dies end up in servers, professional workstations, enthusiasts who don't mind forking out a lot of cash for the fastest in a product skew available. In the case of this specific AMD CPU, I assume the closer to perfect dies ended up in laptops, due to being able to be undervolted more due to more stability overhead available to sacrifice by ubdervolting. It's why I always only buy PC parts when the manufacturing process of that specific part has matured.
The reason why higher frequencies generate more heat is simple: With every clockimpulse, the transistors have to shift signals from 0 to 1 or vice-versa. The parasitic capacitance of the whole circuit acts like a short-circuit, for a fraction of a microsecond. Thus, the more often you switch, the more often you "short" the circuit and generate heat. Like charging a capacitor trough a battery and then reversing polarity, only for discharging and recharging it the other way around.
Great video! I love seeing old CPUs being optimized to their ideal performance point. I'd like to see a video of this CPU in a different motherboard being pushed to the temperature limit with sufficient cooling. You might be able to squeeze 750Mhz or more from it!
Wow. You have gone above and beyond with this video! Edit: Typo This really emhasises how much voltage headroom was in older cpus compared to modern ones. I did some rough back of the napkin math (not counting power draw) voltage/frequency scaling with your minimum voltage data @10:17. Simple equation (Frequency/voltage = arbitrary scaling value) The higher the scaling value the better the scaling. (highest is best) 400Mhz/1.3v=307.69, 450Mhz/1.3v=346.15, 500Mhz/1.35v=370.37, 550Mhz/1.5v=366.67, 570Mhz/1.55v=367.74, 600Mhz/1.7v=352.94 We can conclude that around 500Mhz @ 1.35v was likely the V/F 'sweet spot' for your K6-II+ (Technically K6-III+) CPU. (my results may be flawed but it gets a good idea of the V/F curve) Of note, we can observe that 600Mhz @1.7v the V/F scaling begins to decreace significantly from 570Mhz @1.55v. This begins to idicate the limits of your silicon. You do amazing work here BuB! Keep at it!
Thank you for taking the content further! I think you're spot on with your calculations - and that is probably why AMD stopped at 570 and not 600 MHz. Maybe my CPU sample is a later one with "good" quality since I could unlock the level 2 cache without issues.
@@bitsundbolts You could very well be on to something here. I have a suspicion that AMD took (at least some) full K6-III+ dies with 256kb L2$ and locked them down the cheaper K6-II+ with 128kb L2$ just to move sales volume despite the chips being fully functional with their cache. We’d probably need a larger sample size to confirm that though. Regarding further V/F scaling, a real K6-III+ would need to be compared to for sake of scaling comparison. Keep up the awesomeness BuB!!
Nice video! If included in the Kernel the Linux powernow-k6 driver is actually capable of dynamically scaling the frequency by altering the multiplier during runtime even on desktop motherboards.
I have a DFI 586IPVG that I bought back in the day, it has the VX chipset not the VIA chipset, but it was awesome back in the day having USB so early. I used it for a very long time, and I am going to put it back together soon.
Undervolting has less to to with Manufacturing lables and more to do with signal noise inside the CPU. E.g. Imagine an alternate reality where AMD and Intel decided to lable their manufacturing processes with two different names like "Cucumber" and "Bycicle" instead of using the name "0.18micron" for two different ways of manufacting something. Or imagine if cake recipes were given names like "5nm", "7nm", "10nm", etc...what would a recipe name like "10nm" mean in relation to how the cake is made? Using such names in CPUs had become equally as nonsensical marketing shenanigans since the moment more than 1x transistor that are different were combined in a circuit.
back then i gave absolutely zero facks about undervolting or underclocking, overclocking was still an exploit and not a feature, so it was actually worth it. when my second athlon cpu broke, i was forced to use a pentium 2 400mhz for a while. i ran that thing at 600mhz for about half a year straight. performance was actually pretty decent. completely usable even stepping down from a 1.1ghz athlon. obviously it wasn't anywhere near the same level, but it still allowed me to play my unreal tournament with decent performance in most maps. poor thing certainly wasn't comfortable at this speed though. 150mhz bus speed meant most pci cards wouldn't even work. the ideal overclock for it was 533mhz, everything still worked at this speed. that cpu still lived for a good 10 years as an office pc after i was done with it. still overclocked to 533 the entire time.
I had a K6-III+ on a Gigabyte 5SMM motherboard in around 2006 - this board not only was able to run at 133MHz FSB on Super 7 but it slso had voltage options from 1.3-2.05V in 50mV increments as well as 2.0-3.5V in 100mV increments. I ended up putting an AMD K-6III+ 500ATZ (1.6V nominal voltage) and managed to run it at 3x133 MHz for 400MHz and at just 1.3V - it ran without a heatsink or fan and reached 52C load according to the motherboard temperature sensor
@@bitsundbolts yep the CPU ran like that for more than a year. I still have the CPU around here. I put my finger on the heat spreader while it was under load and i could rest it there for several seconds. Also, all the K6+ CPUs that I have seen are rated to 85C (the Z at the end of the model number indicated 85C) - the non Plus chips were more like 60-65C instead)
I don't have one, but I think there were K6-III CPUs that had an engraving of 1.6 volts. We already know that AMD used the same die for K6-2+ and K6-III+ and artificially limited their lower tier models. Maybe they just didn't care about the updating the aluminum heat spreader at that point...
When I was playing with my K6-II+ it used to fail in 3D-Now! calculations at 600MHz, no matter the voltage. Otherwise it would work ok. Also a 570MHz part, kept it running at 550MHz though.
Testing for stability is hard. I prefer to use the CPU in regular day-to-day workloads. If random crashes occur, something isn't right. But good point on specialized instructions, they may not be used in benchmarks and so the illusion may appear that the CPU is stable until you fire up a 3DNow! application.
That's crazy. Back in the day, I had a 300mhz K6-2 (no +) and don't think I ever tried to OC. I wonder how much performance I left on the table, while still running a space heater....
Probably they had some headroom, but they were also manufactured in a larger 250nm process whereas the + versions were created using 180nm. So, your milage may have been limited. But I do read many posts in forums where owners complained about their K6 CPUs being slow and hot.
@@bitsundbolts Ah...did not realize the process change between the parts. I have recently added some K6-2 CPUs to my collection, as well as the same mainboard I used back in the day. Sounds like a fun experiment!
If you find a good deal on the mobile versions, don't think twice to get one. They run much cooler. The K6-III (non +) used the 250nm manufacturing process as well (=they get hot)
@@bitsundbolts that sounds possible. I think i have seen a board that could do 103mhz or 112mhz fsb for ss7 but that was about it so there's not really a room for more overclocking.
Yes, it totally depends what the board supports. I have two other boards that may be able to do this, but I'll only get them next year. If one of them supports higher FSB speeds, I'll try to get higher with this CPU, but I have a feeling it won't make it past 625 or so.
Ja, da gibt es ein paar Modelle die offiziell bei so niedrigen Voltzahlen laufen. Ich glaube aber nicht, daß es da gravierende Unterschiede gab - vielleicht etwas binning.
Honestly more of them back in the day should have seen the writing on the wall with speed scaling. All the CPUs do it now. Both up and down from nominal clockspeed. You've actually gotta go in and turn it off on desktops now if you don't want it.
If I were to tension up the retaining bracket on the cooler I have in mind, and block off the resistors with kapton tape, would I be able to run my K6-2+ delidded? I want to mod mine to just enable the full 256K of L2, but I don't really see the point in gluing the heat spreader back on when I could just do direct-die cooling. I'm aware that the die is liable to chip with poor handling, but is it any worse than the socket A Athlons for this?
I haven't done overclocking on those CPUs yet because the motherboard doesn't allow for higher than 100 MHz FSB. Once I get a board that can do that, I'll try to see how far this CPU can go. But from what I've read, it will be difficult to get much further than 650MHz.
Even if it could, I don't think there would be any CPU that could run at those speeds. The K6-2+/K6-III architecture reached its limits at around 600-650MHz. At least this is what I have read.
I need a different motherboard with good FSB overclocking support to go beyond 600 MHz. This is all I could do with the DFI board. I read somewhere that there may be a possibility to do it via software, but I haven't checked this yet.
I may replace the heatsink with a smaller one and run this CPU around 500-550 MHz. Way less heat to dissipate now. As you said, this is a possibility to keep those old CPUs cooler and last longer.
Wow, that jump to 2.4 volts for an extra 50 MHz is painful! Silicone lottery. You should probably keep your K6-III+ at 550 MHz and call it a day. There isn't much benefit from those extra 50 MHz. At least not when you need to run that processor at 2.4 volts.
@@bitsundbolts Yeah i figured as such. Most likely because it's a 1.6V 400MHz model that is more binned for low power than high clocks. 550MHz was the highest frequency these were sold at anyway, so I'm quite happy.
![Twenty One Pilots - “The Line” (from Arcane Season 2) [Official Music Video]](http://i.ytimg.com/vi/E2Rj2gQAyPA/mqdefault.jpg)
Is there a good explanation for why AMD and Intel were running chips at significantly higher than necessary voltages back in the day? I had no idea, undervolting really wasn't on my radar at all back when these chips were contemporary and we just wanted max performance, if anything overvolting was something we were more aware of and in some cases doing.
I know with variables at play they maybe picked a 'safe' voltage and didn't want users playing with the minimums but still, it looks like they were using quite significantly more power than strictly necessary.
Good question. I don't know. Will pin your comment, maybe someone has an answer. The gap between stable and stock voltage seems to be too large for the silicone quality argument.
I would assume that some dies would run stable at certain voltages, while others wouldn't. You'd either have to decrease yield, or do some type of binning if you wanted to utilize some dies being stable at lower voltages. Here, the sample size of dies is literally just one, so who knows how the majority of dies, especially the early ones, did perform. And even for this single die, you can see that a stock core voltage of 1.8V might have been marginal, so 2.0V seems like a good and safe choice.
Probably because lower quality PSU's , single stage VRM's on motherboard etc, etc. which both contributed to higher ripples and noise in voltage, which affects stability.
By increasing voltage you increase gap between maximum voltage that can be recognised as logical "0" and minimum voltage that can be recognised as logical "1".
They could not guarantee motherboard quality back in the days, but they could increase voltage specification for CPU and have less unhappy customers returning their product.
Additionally not every motherboard could be set for voltage this low with 0,1v steppes. By marking CPU voltage with 2 volts they made their product compatible with most motherboards available.
That overclock at 1.7v has not been properly tested in this video, and imho it is probably not stable. The prime95 is a steady high load, and NFS3 with such a slow graphics card being the bottleneck running at what looks like 20 fps is not stressing the CPU at all. To properly test this old CPUs you need to use variable loads and high fps gaming for hours.
I had a K6-3+@600 with voodoo3, and it was stable with prime95, games, etc, till one day i decided to play NFS2-SE from start to finish, that game runs at like 160 fps in my CRT with that configuration, after 2-3 hours of gaming the instability started to kick in, and finally I had to reduce a bit the CPU clock.
It was most likely done to improve yields during manufacturing.
Dies are manufactured on wafer (a large disk of silicon that can fit many dies).
So lots of CPUs are manufactured simultaneously.
But the silicon disk (wafer) isn't perfect and has random flaws in its crystal structure. Plus contamination occurs.
Plus flaws happen during manufacturing.
So dies on the same wafer (which were intended to be identical) will have various degrees of flaws which will cause various degrees of instability. Making these dies not identical...and only a tiny fraction of dies will come out perfect out the other end of this manufacturing process.
Way back, such flawed dies were disposed of. But the cost of the wafer and manufacturing remains the same, and if you intended to make profit, the few dies that you sell have to carry that cost.
But, many of the imperfect dies will still work, if you do something to increase stability.
E.g.
1. Lowering the clock
2. Disabling affected features like cache.
3. Or increasing the voltage.
You can now sell these "failed attempts" at lower prices and recoup some more cost of manufacturing...and make more profit or reduce the selling price of the perfect dies to be more competitive.
But what also typically happens is, as manufacturing of a specific CPU architecture and the process node goes on, little issues that cause flaws are solved and the process is said to "mature".
And yields increase (a higher fraction of dies are perfect because flaws are less frequent)
But different consumers have different budgets and this results in different markets. High end markets, medium, low.
As manufacting goes on, supply for the high end market increases due to increased yields, while supply for the low end market decreases (because there's fewer badly flawed chips to clock down, disable features on, etc).
Reducing supply is bad, because the consumer will simply switch to a competitor who does have supply.
So CPU manufacturers will use perfectly good dies from the higher stockpiles for lower markets, just so they don't lose market share in a specific market.
And during these later more mature stages of manufacturing many consumers are unknowingly blessed with CPUs that are capable of much more than they are packaged and labeled as. These CPUs will be great overclockers due to the extra stability overhead.
The same is done with GPUs. Most GPUs and CPUs are simply failed attempts that are salvaged and sold at lower prices. While the perfect dies end up in servers, professional workstations, enthusiasts who don't mind forking out a lot of cash for the fastest in a product skew available.
In the case of this specific AMD CPU, I assume the closer to perfect dies ended up in laptops, due to being able to be undervolted more due to more stability overhead available to sacrifice by ubdervolting.
It's why I always only buy PC parts when the manufacturing process of that specific part has matured.
The reason why higher frequencies generate more heat is simple: With every clockimpulse, the transistors have to shift signals from 0 to 1 or vice-versa. The parasitic capacitance of the whole circuit acts like a short-circuit, for a fraction of a microsecond. Thus, the more often you switch, the more often you "short" the circuit and generate heat. Like charging a capacitor trough a battery and then reversing polarity, only for discharging and recharging it the other way around.
geiler Name, Christoph Stirnlappen hier
Great video! I love seeing old CPUs being optimized to their ideal performance point. I'd like to see a video of this CPU in a different motherboard being pushed to the temperature limit with sufficient cooling. You might be able to squeeze 750Mhz or more from it!
I'll get two more SS7 motherboards soon. Will try to get as far as possible depending on overclocking capabilities.
Thanks for watching!
Nice!! I have a K6-III+ 400 somewhere around here, will try to get it and play when I have some time 👍
Wow. You have gone above and beyond with this video!
Edit: Typo
This really emhasises how much voltage headroom was in older cpus compared to modern ones.
I did some rough back of the napkin math (not counting power draw) voltage/frequency scaling with your minimum voltage data @10:17. Simple equation (Frequency/voltage = arbitrary scaling value) The higher the scaling value the better the scaling. (highest is best)
400Mhz/1.3v=307.69, 450Mhz/1.3v=346.15, 500Mhz/1.35v=370.37, 550Mhz/1.5v=366.67, 570Mhz/1.55v=367.74, 600Mhz/1.7v=352.94
We can conclude that around 500Mhz @ 1.35v was likely the V/F 'sweet spot' for your K6-II+ (Technically K6-III+) CPU. (my results may be flawed but it gets a good idea of the V/F curve)
Of note, we can observe that 600Mhz @1.7v the V/F scaling begins to decreace significantly from 570Mhz @1.55v. This begins to idicate the limits of your silicon.
You do amazing work here BuB! Keep at it!
Thank you for taking the content further! I think you're spot on with your calculations - and that is probably why AMD stopped at 570 and not 600 MHz.
Maybe my CPU sample is a later one with "good" quality since I could unlock the level 2 cache without issues.
@@bitsundbolts You could very well be on to something here. I have a suspicion that AMD took (at least some) full K6-III+ dies with 256kb L2$ and locked them down the cheaper K6-II+ with 128kb L2$ just to move sales volume despite the chips being fully functional with their cache.
We’d probably need a larger sample size to confirm that though. Regarding further V/F scaling, a real K6-III+ would need to be compared to for sake of scaling comparison.
Keep up the awesomeness BuB!!
I like where this is going. :)
Undervolting: a saga.
Interesting video, like !
Thanks!
Nice video! If included in the Kernel the Linux powernow-k6 driver is actually capable of dynamically scaling the frequency by altering the multiplier during runtime even on desktop motherboards.
I have a DFI 586IPVG that I bought back in the day, it has the VX chipset not the VIA chipset, but it was awesome back in the day having USB so early. I used it for a very long time, and I am going to put it back together soon.
Fantastic job. cool video 💪
Thank you!
Nice vid and explanation! It makes sense to be able to do 1.7V, since the CPU is 0.18nm (like the P3 Coppermine).
Undervolting has less to to with Manufacturing lables and more to do with signal noise inside the CPU.
E.g. Imagine an alternate reality where AMD and Intel decided to lable their manufacturing processes with two different names like "Cucumber" and "Bycicle" instead of using the name "0.18micron" for two different ways of manufacting something.
Or imagine if cake recipes were given names like "5nm", "7nm", "10nm", etc...what would a recipe name like "10nm" mean in relation to how the cake is made? Using such names in CPUs had become equally as nonsensical marketing shenanigans since the moment more than 1x transistor that are different were combined in a circuit.
back then i gave absolutely zero facks about undervolting or underclocking, overclocking was still an exploit and not a feature, so it was actually worth it. when my second athlon cpu broke, i was forced to use a pentium 2 400mhz for a while. i ran that thing at 600mhz for about half a year straight. performance was actually pretty decent. completely usable even stepping down from a 1.1ghz athlon. obviously it wasn't anywhere near the same level, but it still allowed me to play my unreal tournament with decent performance in most maps. poor thing certainly wasn't comfortable at this speed though. 150mhz bus speed meant most pci cards wouldn't even work. the ideal overclock for it was 533mhz, everything still worked at this speed.
that cpu still lived for a good 10 years as an office pc after i was done with it. still overclocked to 533 the entire time.
Overclocking was fun back in the day..
miss my overclocking days. 95 to 2006 when I stopped making my own PCs.
I had a K6-III+ on a Gigabyte 5SMM motherboard in around 2006 - this board not only was able to run at 133MHz FSB on Super 7 but it slso had voltage options from 1.3-2.05V in 50mV increments as well as 2.0-3.5V in 100mV increments. I ended up putting an AMD K-6III+ 500ATZ (1.6V nominal voltage) and managed to run it at 3x133 MHz for 400MHz and at just 1.3V - it ran without a heatsink or fan and reached 52C load according to the motherboard temperature sensor
Wow - passive cooling without a heatsink! The 52C sound like it is close to the limit those CPU were meant to run at. But I guess the CPU survived it?
@@bitsundbolts yep the CPU ran like that for more than a year. I still have the CPU around here. I put my finger on the heat spreader while it was under load and i could rest it there for several seconds. Also, all the K6+ CPUs that I have seen are rated to 85C (the Z at the end of the model number indicated 85C) - the non Plus chips were more like 60-65C instead)
Super nice!
So, they had 0.45V reserve for the stock speed. I would never expect that, its huge. Maybe 0.1-0.15v is to be expected, but this...
Exactly my thought.
I don't have one, but I think there were K6-III CPUs that had an engraving of 1.6 volts. We already know that AMD used the same die for K6-2+ and K6-III+ and artificially limited their lower tier models. Maybe they just didn't care about the updating the aluminum heat spreader at that point...
When I was playing with my K6-II+ it used to fail in 3D-Now! calculations at 600MHz, no matter the voltage. Otherwise it would work ok. Also a 570MHz part, kept it running at 550MHz though.
Testing for stability is hard. I prefer to use the CPU in regular day-to-day workloads. If random crashes occur, something isn't right.
But good point on specialized instructions, they may not be used in benchmarks and so the illusion may appear that the CPU is stable until you fire up a 3DNow! application.
This video, I very much enjoyed. xD
I am glad you enjoyed the video. Thanks for watching!
That's crazy. Back in the day, I had a 300mhz K6-2 (no +) and don't think I ever tried to OC. I wonder how much performance I left on the table, while still running a space heater....
Probably they had some headroom, but they were also manufactured in a larger 250nm process whereas the + versions were created using 180nm. So, your milage may have been limited.
But I do read many posts in forums where owners complained about their K6 CPUs being slow and hot.
@@bitsundbolts Ah...did not realize the process change between the parts. I have recently added some K6-2 CPUs to my collection, as well as the same mainboard I used back in the day. Sounds like a fun experiment!
If you find a good deal on the mobile versions, don't think twice to get one. They run much cooler. The K6-III (non +) used the 250nm manufacturing process as well (=they get hot)
man, you are sick ))))
I'll take this as a compliment ☺️
@@bitsundbolts yes sir, it is
first :)
also, i own 500Mhz variant with OC to 600Mhz. Never tried to get it higher.
I've read that more than 600 MHz is really not worth it. You probably get the best result somewhere between 550 to 625 MHz for those CPUs.
@@bitsundbolts that sounds possible. I think i have seen a board that could do 103mhz or 112mhz fsb for ss7 but that was about it so there's not really a room for more overclocking.
Yes, it totally depends what the board supports. I have two other boards that may be able to do this, but I'll only get them next year. If one of them supports higher FSB speeds, I'll try to get higher with this CPU, but I have a feeling it won't make it past 625 or so.
@@bitsundbolts i ran mine over 650
700 was a no go
In case one of the other two SS7 motherboards support proper overclocking settings, I am going to attempt an overclock as well!
Wenn ich mich richtig erinnere wurde der K6-3+ 400 sogar mit 1.6V CORE in Embedded Systemen verbaut
Ja, da gibt es ein paar Modelle die offiziell bei so niedrigen Voltzahlen laufen. Ich glaube aber nicht, daß es da gravierende Unterschiede gab - vielleicht etwas binning.
Honestly more of them back in the day should have seen the writing on the wall with speed scaling.
All the CPUs do it now. Both up and down from nominal clockspeed.
You've actually gotta go in and turn it off on desktops now if you don't want it.
If I were to tension up the retaining bracket on the cooler I have in mind, and block off the resistors with kapton tape, would I be able to run my K6-2+ delidded? I want to mod mine to just enable the full 256K of L2, but I don't really see the point in gluing the heat spreader back on when I could just do direct-die cooling. I'm aware that the die is liable to chip with poor handling, but is it any worse than the socket A Athlons for this?
Inhaber read that the die of those k6-2 CPUs is extremely fragile. You might risk cracking the die with direct die cooling.
I'm interested in seeing what kind of benchmarks you can get with the K6-II+ when its fully unlocked.
I'll do videos on this CPU in the future.
i had hoped it would allow you to run it at 750mhz plus since it was so good with voltage drops in the chart
I haven't done overclocking on those CPUs yet because the motherboard doesn't allow for higher than 100 MHz FSB. Once I get a board that can do that, I'll try to see how far this CPU can go. But from what I've read, it will be difficult to get much further than 650MHz.
@@bitsundbolts ah right I thought the thermal head room you had was the limiting factor
Oh I’ve taken a K6-3+ way lower. Every time I turn the laptop off it goes down to 0v.
If only the Super socket 7 had an official mvp4 chipset with a true Agp slot, a true 133 mhz support. 6x133.3 mhz = 800 mhz !
Even if it could, I don't think there would be any CPU that could run at those speeds. The K6-2+/K6-III architecture reached its limits at around 600-650MHz. At least this is what I have read.
Hi Mister, Where have you got a true 600 mhz ?
I am afraid to say, it is fake :) just my skills with a photo editing software... It is a K6-2+ 570.
change crystal osc with higher one.
this cpu can run up to 800MHz with higher frequency crystal osc but it tricky with isa
НАСКОЛЬКО Я ПОМНЮ ТАМ БОЛЬШЕ 550 МХЗ НЕ ОТОБРАЖАЕТСЯ.
MORE THAN 550MHz LIMETED TO SHOW IN MAINBOARD ~VIA 512 WT-CACHE.
but could you overclock it as much as it'll go
I need a different motherboard with good FSB overclocking support to go beyond 600 MHz. This is all I could do with the DFI board. I read somewhere that there may be a possibility to do it via software, but I haven't checked this yet.
@@bitsundbolts that's fair
undervolting is probably what you want anyways if you want to reduce the power of your old machine.
I may replace the heatsink with a smaller one and run this CPU around 500-550 MHz. Way less heat to dissipate now.
As you said, this is a possibility to keep those old CPUs cooler and last longer.
Jesus, my K6-III+ needs 1.8V for 550MHz and 2.4V for 600MHz if i want stable operation. It really doesn't like 600MHz
Wow, that jump to 2.4 volts for an extra 50 MHz is painful! Silicone lottery. You should probably keep your K6-III+ at 550 MHz and call it a day. There isn't much benefit from those extra 50 MHz. At least not when you need to run that processor at 2.4 volts.
@@bitsundbolts Yeah i figured as such. Most likely because it's a 1.6V 400MHz model that is more binned for low power than high clocks. 550MHz was the highest frequency these were sold at anyway, so I'm quite happy.
This must be the most modified k6 🤣