Taking a look at a modded AMD K6-2+ 570 modded into a k6-III+ and how it performs. my original video on my AMD K6-III+ Build: • AMD K6-III+ Super Sock...
There are two reasons for making K6-2+ chips out of K6-3+ chips. Making a new set of masks and developing a complete new SKU has a large fixed cost that you get rid of if you make two chips from the same die. Some of those would have a defect in the massive L2 cache. So they engineered a way to disable half the L2 and make a K6-2+ chip out of it. If they didn't have enough defective chips they could take one of the worse quality dies and disable the L2. Eventually, there were few poor quality or defective dies but they still wanted the SKU so they wouldn't have to prematurely lower the price of K6-3+ before some mobile athlon something or other came out.
Nice video! I have yet to find a K6-2+ with bad L2 cache. Would be curious to see how it behaves. I modified 3 K6-2+ CPU until now. One 450 and two 570 - all working fine with the unlocked L2 cache. The jumper-mod on the K6-2+ was an idea to see if the cache prevents the K6-2+ from clocking higher. Probably a video that I'll make in the future - when I get a better board with options higher than 100MHz FSB.
Excellent video. The Voodoo 3 is just too powerful even for the fastest K6-3s. The V3 being badly bottlenecked by the K6-3, in almost all games the performance of the V3 3500 (183 mhz) will be identical to the V3 2000 (143 mhz), in 1024x768 and lower resolutions.
It was cheaper just to make one die for all the market segments, and charge premium for the full die. It could be that when the CPU first released it had a lower yield meaning that they had a bunch of binned CPU's with only half the cache working. while the full L2 chip was sold at a premium price. But as the technology matured they've got to a higher yield anyway, but they had to gimp the lower tier CPU's to keep up market segmenting (albeit artificially). AMD has a long history with hardware unlock-able CPU and GPU features.
To add to this: chip companies make most of their profits off their most expensive products, but they need the volume of low-cost parts to cover the expenses of manufacturing and R&D. The high end architectures of the 80s and 90s like SPARC ultimately went nowhere because while the vendors could sell each chip for a lot of money, the low numbers being sold meant that the manufacturing/R&D overheads per chip were very high, and that limited the amount of investment into future designs they could. Intel and AMD's trick has been to sell huge numbers of low end chips which reduce the overhead costs per chip to nothing, and then sell high end chips to businesses and enthusiasts which bring the profit in.
When i was a kid, like age 13, back to 2002, i had an k6-2 400Mhz and all i want is to play CS1.5 in that era, but the FPS was so TERRIBLE, even with an TNT2 from Nvidia, i would love to see how much fps this 600/633Mhz can do on it.
I use this modded piece of awesomeness too - on a P5A-B - also 600mhz without a problem. And: with only a 16MB 3dfx Banshee in the AGP i score like 3.960 in 3DMark99 Max. With a single Voodoo 2 12MB only 2.600. I would say that a Banshee is the perfect GPU for this setup. Try it with a Banshee and let us know if you also score close to 4.000. I use the X-Mojo driver, Cheers
Unfortunately I sold my only Banshee card awhile back. Im surprised the banshee would score better than the voodoo 3 setup, maybe it has to do with my use of reference drivers.
@@AncientElectronics I had tried a Voodoo 3 3000 too and i never managed to beat the Banshee no matter what driver i used for the V3. The V3 is just depending on much more processing power which the K6 line cant hold up with. Weak FPU etc.. A Banshee handles that way better.
Yes, a Banshee is more than enough for a Super Socket 7 system, thats why I use a Banshee AGP (ELSA Victory II clocked to 120MHz core and 125MHz memory) for my K6-III+ system, maybe a not so often seeing Compaq V3 1000 16MB AGP is also ok, the slowest 16MB V3 version with 125MHz, but enough for a system like this.
CPU multipliers of 2 and 6 are the same and your jumper settings for it are also the same. The CPU is supposed to do the rest internally and map 2x to 6x. I'd be suspicious of the chip having being relabeled, personally, though so long as it works it probably doesn't really matter. They're interesting enough CPUs alright, nice to see running, but not ones I ever had any desire to run in a world that had Durons available for less than a week's groceries.
Thats what I thought as well, I'm not quite sure what's going on with it detecting as x2 and not x6. I initially just assumed it was a quirk from having the CPU modded and maybe my particular motherboard but I confirmed that another user with the same board revision and modded chip didn't have any issues.
Nice job! Lol, I got so annoyed today, I finally received my K6-2+ 500, I was looking forward to upgrade my super socket7 board running a K6-2 450 to see the difference 128KB of L2 cash can do... And my board is a Shuttle Hot-951P The manual claims it can't go higher then 300 (which I manage to run my k6-2 450) latest bios that can be found is in... sadly the K6-2+ is recognized as Unknown Processor-S 400Mhz and freezes right after the memory count. Perhaps one day Ill find a decent Super Socket 7 board for it.
It's usually more profitable to make the top end cpus with all of the chips per wafer, but at some point - the cost of production of chips goes down, and the same goes with price of top end CPUs (for example, when the company already has a higher end CPU and the previously high end is nearing its obsolescence) - but there is still demand for older CPUs since consumers have the need to buy K6-2s and K6-III and even probably some are still buying K5's at this time.... and since your CPUs are nearing end of life, you are gradually making less and less of older CPUs - meaning you no longer are ordering 2 separate wafers - one for K6-2s and the other for K6-IIIs since there is no longer the demand for as many CPUs - so it's more feasible to order just one wafer for K6-IIIs and allocate some of it to produce K6-2s - just enough to satisfy the small demand, and not leave users high and dry....
phil has a page for the yamaha card, i recommend installing the power ymf which will improve the midi capabilities. and to confirm one the previous comment built in midi does work in pure dos. I do think you get the mpu401 support though (but i have my yamaha in laptop so cant say for sure) I picked up one of those K62 processors (might have been from the new old stock), but mine is not the plus. I was upgrading from a p200mmx so overclocking meant little.
I have witnessed Expendable benchmark desynchronization with by Slot 1 build, unfortnutely I don't remember which graphics card I used. I'm frequently switching between Voodoo 3, TNT2 Ultra and Geforce 256. Could be the V3. By the way, I like that Yamaha PCI card too but it's too buggy to my liking. Cannot run Duke3D with it at all (it just crashes) and one of my favourite games, which is Little Big Adventure 2, refuses to produce any sound until I switch from VXD drivers to WDM. The drivers defeinitely hold the card back as it's otherwise very nice feature wise.
24:25 - i've observed similar "off-script" issue on benchmarking older machines with doom. The issue has always been instability of the cache in some way. ;) Raise CPU voltage and lower the cpu speed (preferably FSB speed as maybe it's an issue on motherboard) and check, if the issue persists.
Well, the box says the optimum CPU is a 450MHz Pentium II. The K6-III+ here should be about equivalent to that. I've never tried running the benchmark on anything slower than a k6-III though.
It's not the Voodoo card, it's my capture method. The V3 looks great on my PC CRT but I've never been able to dial in my capture method to look anywhere as good.
I feel that AMD really wasted their time holding out for the K6-III to give you the cache. K6-III did NOT have much success or actual release. I get that doing it my way would still have been way slower than P2 and P3, but still!
The TNT2 / Ultra looses 25-50% performance when switching from 16 to 32-bit for minimal visual improvements. It doesn’t make sense to play with the TNT2 in 32-bit. Also for a SS7 the Voodoo 3 is much much better being clearly faster especially in Glide games, having a better 16-bit image quality because of its 22-bit filter and much better game compatibility. There are many games like NFS 2 SE or Carmageddon 1 which only run in Glide mode with hardware acceleration with ZERO support for other card than 3dfx Voodoos. In those games with the NVidia ATi Matrox etc cards you have to play in SOFTWARE mode while the Voodoos perfectly run in Glide with hardware acceleration. And there are also many games from 1996-2000 which run by default with higher visual details in Glide, unavailable in direct3d or opengl - Unreal Tournament is such an example where detail textures and fog are available only in 3dfx Glide mode.
I went with the TNT2 Ultra in my original build mostly for the reason that I already had a Windows 98 build with a Voodoo 5 (video will be coming eventually) as well as a Windows 95 PC with Voodoo 2's in SLI and a DOS PC with a Voodoo 1 so the V3 felt a little redundant as I do not believe the V3 plays anything the V5 doesn't. I think the TNT2 Ultra is still a really good choice for the k6-III+ but after running both I do agree the V3 is the overall better choice.
Just like the ISA version, this PCI yamaha card has a real OPL3, it's not emulation, it's just opl3. Here's a comparison video and you can hear it sounds exactly the same (with far less noise) th-cam.com/video/2v6FqmQbs2w/w-d-xo.html As for midi on DOS games, it's normal : the yamaha XG part only works in windows. You can perfectly run these games through windows and benefit from the XG midi.
...does Expendable do the same thing at a lower clock speed...? (Just because back in the 90s the imps in Doom shot their fireballs into the ground right in front of them - when my Am5x86 ran at 160MHz and had a minute or so to warm up, perhaps a little too much for the speed/voltage it was set to. And it was fine even at a slightly slower 150MHz. So it wasn't a normal freeze or dos4gw error but a rather strange behavior.)
@@AncientElectronics Nah, I was just curious if it's this cery CPU at this very clock. Maybe it runs without problems without overclocking. Maybe not. Was it fine with the precious one?
@@alvaroacwellan9051 I've never run the benchmark on anything slower than a 550mhz k6-III+ that I recall but I've had the problem pop up on faster systems as well, Pentium III's, and early 1.5GHz and 2GHz Pentium 4s's. As I mentioned in the video Phil who also has a youtube channel has also run this game on many different CPUs and builds and told me he never encountered the error.
Dumb question: can a K6-II be used in an older intel socket7 board? I have a 400mhz chip and would like to use it underclocked as an alternative to forking out an intel mmx to upgrade my 133mhz AT pentium.
@@AncientElectronics Thanks very much, i'll have to look into it when i get my system apart - the board is a QDI legend explorer and i can't remember it saying anything about MMX, though it does support 200mhz intel Pentiums which i thought were all MMX chips...
I could be wrong but I believe there is a non-mmx pentium 200 so double check with the board. Socket 7 boards can be hit or miss with supporting dual voltage. Some just can, some require a voltage regulator added and some don't the ones that don't seem little more than socket 5.
@@AncientElectronics you are right, there is a non-mmx s7 pentium. It has model numbers ranging SL24Q, SY045 and SU114 and is marked as "i200" on the silver top that makes no statement of MMX tech.
no me gusta el k6-2 en aquel momento ya intel tenia pentium 2 a 350-400-450 mhz el amd k6-2 era para vender ordenadores baratos, voodoo 3000 funciona mucho mejor en pentium 2 a 450 mhz
There are two reasons for making K6-2+ chips out of K6-3+ chips. Making a new set of masks and developing a complete new SKU has a large fixed cost that you get rid of if you make two chips from the same die. Some of those would have a defect in the massive L2 cache. So they engineered a way to disable half the L2 and make a K6-2+ chip out of it. If they didn't have enough defective chips they could take one of the worse quality dies and disable the L2. Eventually, there were few poor quality or defective dies but they still wanted the SKU so they wouldn't have to prematurely lower the price of K6-3+ before some mobile athlon something or other came out.
Thanks for the explanation.
Nice video! I have yet to find a K6-2+ with bad L2 cache. Would be curious to see how it behaves. I modified 3 K6-2+ CPU until now. One 450 and two 570 - all working fine with the unlocked L2 cache. The jumper-mod on the K6-2+ was an idea to see if the cache prevents the K6-2+ from clocking higher. Probably a video that I'll make in the future - when I get a better board with options higher than 100MHz FSB.
Excellent video. The Voodoo 3 is just too powerful even for the fastest K6-3s. The V3 being badly bottlenecked by the K6-3, in almost all games the performance of the V3 3500 (183 mhz) will be identical to the V3 2000 (143 mhz), in 1024x768 and lower resolutions.
True that... maybe you remember my comment from another channel - using a Banshee i scored almost 4.000 in 3DMark99 Max.
How about a TNT2 Ultra vs V3 183 mhz comparison video on the Pentium 3 ? That would be awesome :)
a GF2 would be best
New Sub! Dryden, MI.
Nice Job; Gonna tag along and see what else you got!
Oh my, one of my systems back in the day, including the case was pretty much the same.
Looks like a nice build.
It was cheaper just to make one die for all the market segments, and charge premium for the full die. It could be that when the CPU first released it had a lower yield meaning that they had a bunch of binned CPU's with only half the cache working. while the full L2 chip was sold at a premium price. But as the technology matured they've got to a higher yield anyway, but they had to gimp the lower tier CPU's to keep up market segmenting (albeit artificially). AMD has a long history with hardware unlock-able CPU and GPU features.
To add to this: chip companies make most of their profits off their most expensive products, but they need the volume of low-cost parts to cover the expenses of manufacturing and R&D. The high end architectures of the 80s and 90s like SPARC ultimately went nowhere because while the vendors could sell each chip for a lot of money, the low numbers being sold meant that the manufacturing/R&D overheads per chip were very high, and that limited the amount of investment into future designs they could. Intel and AMD's trick has been to sell huge numbers of low end chips which reduce the overhead costs per chip to nothing, and then sell high end chips to businesses and enthusiasts which bring the profit in.
When i was a kid, like age 13, back to 2002, i had an k6-2 400Mhz and all i want is to play CS1.5 in that era, but the FPS was so TERRIBLE, even with an TNT2 from Nvidia, i would love to see how much fps this 600/633Mhz can do on it.
I use this modded piece of awesomeness too - on a P5A-B - also 600mhz without a problem. And: with only a 16MB 3dfx Banshee in the AGP i score like 3.960 in 3DMark99 Max. With a single Voodoo 2 12MB only 2.600. I would say that a Banshee is the perfect GPU for this setup. Try it with a Banshee and let us know if you also score close to 4.000. I use the X-Mojo driver, Cheers
Unfortunately I sold my only Banshee card awhile back. Im surprised the banshee would score better than the voodoo 3 setup, maybe it has to do with my use of reference drivers.
@@AncientElectronics I had tried a Voodoo 3 3000 too and i never managed to beat the Banshee no matter what driver i used for the V3. The V3 is just depending on much more processing power which the K6 line cant hold up with. Weak FPU etc.. A Banshee handles that way better.
Yes, a Banshee is more than enough for a Super Socket 7 system, thats why I use a Banshee AGP (ELSA Victory II clocked to 120MHz core and 125MHz memory) for my K6-III+ system, maybe a not so often seeing Compaq V3 1000 16MB AGP is also ok, the slowest 16MB V3 version with 125MHz, but enough for a system like this.
CPU multipliers of 2 and 6 are the same and your jumper settings for it are also the same. The CPU is supposed to do the rest internally and map 2x to 6x. I'd be suspicious of the chip having being relabeled, personally, though so long as it works it probably doesn't really matter. They're interesting enough CPUs alright, nice to see running, but not ones I ever had any desire to run in a world that had Durons available for less than a week's groceries.
Thats what I thought as well, I'm not quite sure what's going on with it detecting as x2 and not x6. I initially just assumed it was a quirk from having the CPU modded and maybe my particular motherboard but I confirmed that another user with the same board revision and modded chip didn't have any issues.
Nice job!
Lol, I got so annoyed today, I finally received my K6-2+ 500, I was looking forward to upgrade my super socket7 board running a K6-2 450 to see the difference 128KB of L2 cash can do...
And my board is a Shuttle Hot-951P
The manual claims it can't go higher then 300 (which I manage to run my k6-2 450)
latest bios that can be found is in... sadly the K6-2+ is recognized as Unknown Processor-S 400Mhz and freezes right after the memory count.
Perhaps one day Ill find a decent Super Socket 7 board for it.
It's usually more profitable to make the top end cpus with all of the chips per wafer, but at some point - the cost of production of chips goes down, and the same goes with price of top end CPUs (for example, when the company already has a higher end CPU and the previously high end is nearing its obsolescence) - but there is still demand for older CPUs since consumers have the need to buy K6-2s and K6-III and even probably some are still buying K5's at this time.... and since your CPUs are nearing end of life, you are gradually making less and less of older CPUs - meaning you no longer are ordering 2 separate wafers - one for K6-2s and the other for K6-IIIs since there is no longer the demand for as many CPUs - so it's more feasible to order just one wafer for K6-IIIs and allocate some of it to produce K6-2s - just enough to satisfy the small demand, and not leave users high and dry....
phil has a page for the yamaha card, i recommend installing the power ymf which will improve the midi capabilities.
and to confirm one the previous comment built in midi does work in pure dos. I do think you get the mpu401 support though (but i have my yamaha in laptop so cant say for sure)
I picked up one of those K62 processors (might have been from the new old stock), but mine is not the plus. I was upgrading from a p200mmx so overclocking meant little.
I'll have to check that out, thanks!
mod your 3dfx bios to allow for fast writes and side band addressing and see if that helps.
That Yamaha YMF 724 card only has General MIDI support under Windows. In pure DOS, it can only use FM synthesis.
Did you try setting the multiplier to 2. CPU should interpret this as multiplier 6.
It's the same setting on the board.
I have witnessed Expendable benchmark desynchronization with by Slot 1 build, unfortnutely I don't remember which graphics card I used. I'm frequently switching between Voodoo 3, TNT2 Ultra and Geforce 256. Could be the V3.
By the way, I like that Yamaha PCI card too but it's too buggy to my liking. Cannot run Duke3D with it at all (it just crashes) and one of my favourite games, which is Little Big Adventure 2, refuses to produce any sound until I switch from VXD drivers to WDM. The drivers defeinitely hold the card back as it's otherwise very nice feature wise.
24:25 - i've observed similar "off-script" issue on benchmarking older machines with doom. The issue has always been instability of the cache in some way. ;)
Raise CPU voltage and lower the cpu speed (preferably FSB speed as maybe it's an issue on motherboard) and check, if the issue persists.
Well, the box says the optimum CPU is a 450MHz Pentium II. The K6-III+ here should be about equivalent to that. I've never tried running the benchmark on anything slower than a k6-III though.
That display is unusually blurry for voodoo 3. The card has great DAC, should be a bit sharper (im talking about win desktop, not games).
It's not the Voodoo card, it's my capture method. The V3 looks great on my PC CRT but I've never been able to dial in my capture method to look anywhere as good.
I have that sound card and their is software for it to do more stuff
I feel that AMD really wasted their time holding out for the K6-III to give you the cache. K6-III did NOT have much success or actual release. I get that doing it my way would still have been way slower than P2 and P3, but still!
The TNT2 / Ultra looses 25-50% performance when switching from 16 to 32-bit for minimal visual improvements. It doesn’t make sense to play with the TNT2 in 32-bit. Also for a SS7 the Voodoo 3 is much much better being clearly faster especially in Glide games, having a better 16-bit image quality because of its 22-bit filter and much better game compatibility. There are many games like NFS 2 SE or Carmageddon 1 which only run in Glide mode with hardware acceleration with ZERO support for other card than 3dfx Voodoos. In those games with the NVidia ATi Matrox etc cards you have to play in SOFTWARE mode while the Voodoos perfectly run in Glide with hardware acceleration. And there are also many games from 1996-2000 which run by default with higher visual details in Glide, unavailable in direct3d or opengl - Unreal Tournament is such an example where detail textures and fog are available only in 3dfx Glide mode.
I went with the TNT2 Ultra in my original build mostly for the reason that I already had a Windows 98 build with a Voodoo 5 (video will be coming eventually) as well as a Windows 95 PC with Voodoo 2's in SLI and a DOS PC with a Voodoo 1 so the V3 felt a little redundant as I do not believe the V3 plays anything the V5 doesn't. I think the TNT2 Ultra is still a really good choice for the k6-III+ but after running both I do agree the V3 is the overall better choice.
Yea V3 is the way to go with Socket 7. It also avoids all the AGP chipset bugs.
Just like the ISA version, this PCI yamaha card has a real OPL3, it's not emulation, it's just opl3.
Here's a comparison video and you can hear it sounds exactly the same (with far less noise) th-cam.com/video/2v6FqmQbs2w/w-d-xo.html
As for midi on DOS games, it's normal : the yamaha XG part only works in windows. You can perfectly run these games through windows and benefit from the XG midi.
I thought I remembered reading that the OPL 3 was real, thanks for confirming.
...does Expendable do the same thing at a lower clock speed...?
(Just because back in the 90s the imps in Doom shot their fireballs into the ground right in front of them - when my Am5x86 ran at 160MHz and had a minute or so to warm up, perhaps a little too much for the speed/voltage it was set to. And it was fine even at a slightly slower 150MHz. So it wasn't a normal freeze or dos4gw error but a rather strange behavior.)
I don't think I've ever tried running it on anything slower than a k6-III. the box indicated a Pentium II 450 as the optimum CPU.
@@AncientElectronics Nah, I was just curious if it's this cery CPU at this very clock. Maybe it runs without problems without overclocking. Maybe not. Was it fine with the precious one?
@@alvaroacwellan9051 I've never run the benchmark on anything slower than a 550mhz k6-III+ that I recall but I've had the problem pop up on faster systems as well, Pentium III's, and early 1.5GHz and 2GHz Pentium 4s's. As I mentioned in the video Phil who also has a youtube channel has also run this game on many different CPUs and builds and told me he never encountered the error.
@@AncientElectronics I see.
Dumb question: can a K6-II be used in an older intel socket7 board?
I have a 400mhz chip and would like to use it underclocked as an alternative to forking out an intel mmx to upgrade my 133mhz AT pentium.
If the board supports split voltage it should work, I'd think. So if the boards support MMX processors the k6-II should work.
@@AncientElectronics Thanks very much, i'll have to look into it when i get my system apart - the board is a QDI legend explorer and i can't remember it saying anything about MMX, though it does support 200mhz intel Pentiums which i thought were all MMX chips...
I could be wrong but I believe there is a non-mmx pentium 200 so double check with the board. Socket 7 boards can be hit or miss with supporting dual voltage. Some just can, some require a voltage regulator added and some don't the ones that don't seem little more than socket 5.
@@AncientElectronics Thanks for this advise Justin, you're the best
@@AncientElectronics you are right, there is a non-mmx s7 pentium. It has model numbers ranging SL24Q, SY045 and SU114 and is marked as "i200" on the silver top that makes no statement of MMX tech.
No theme song?
It's there.
@@AncientElectronics I must have forgotten soon after watching it.
no me gusta el k6-2 en aquel momento ya intel tenia pentium 2 a 350-400-450 mhz el amd k6-2 era para vender ordenadores baratos, voodoo 3000 funciona mucho mejor en pentium 2 a 450 mhz