The VIA MVP3 chipset on my FIC VA-503+ has been a rock stable board for as long as I've had it. However, for any sort of modern usefulness, the VIA chipsets that I've seen have a limitation of 128mb of RAM per DIMM slot, meaning that the typical board is stuck using 256mb/384mb RAM. Some ALI boards have hacked BIOS files available that allow them to use 512mb Modules for 1.0-1.5GB Total RAM.

Both chipsets are probably good choices but I can only speak from the view that I've traditionally had good luck with the VIA boards. The only issue I could speak of is a recent experience of getting a PowerVR to work on the darn thing. The only way it would work is if no other PCI cards are installed on the bus, however that issue may be result of a conflict rather than any bus mastering issues as I haven't looked into it enough.

Also, as for finding a Super7 board with 2MB Cache, most I've seen only had 512k-1MB (mine has 1MB).

Here's a cool guide for those era of chipsets, I haven't read the whole thing over, but appears to have a ton of information on P5/P5 and SS7 era chipsets. Looking at it briefly, the TAG RAM chip of the ALi's cache system was integrated into the chipset, and has a limit of 1MB L2 Cache, whereas VIA MVP3 can handle 2MB (though I've never seen one).

http://www.anandtech.com/show/72/19

I have both the P5A and P5A-B from Asus. they are ali chipsets, they come with 512kb of cache and I havent had any issues. that being said, i ahve heard that k6-2 performance on this particular chipset is reduced compared to mvp.

That being said, I have a FEW k6 and k6-2 processors around, ( and a few pentiums) and I can give you benchmarks if you want. I REALLY with my boards had 1mb cache, but both have 512k. That being said, even with the alladin chipset, they still have add-on tag ram. ( on some)

maddmaxstar wrote:

Here's a cool guide for those era of chipsets... http://www.anandtech.com/show/72/19

Seems you didn't view the PDF I attached.

The amount of cache is not the limiting factor; it is the cacheable memory range of the cache given the employed caching scheme on the motherboard. I have seen an MVP3 with 2 MB of cache; one just sold on eBay. I didn't buy it because I didn't trust the guy selling it. It can still only cache 256 MB, or maybe 512 MB of RAM with 2 MB of cache. The ALi board, with a lot less cache (512KB on the chipset), can cache 512 MB of RAM using a different caching scheme. Refer to the attached PDF in the original post.

luckybob: I also have the P5A-B from Asus, but I'm looking to see if anyone out there has had any long term real world experience with it (circa 1999).

[Quick reply mode = On]
Does this Vogons link help? In it, there's discussion about a cache bug.

how are your guys ali5 boards running in windows2000? my gigabyte 5ax is ok in win98 but had lot of trouble in 2000.

retro games 100 wrote:

[Quick reply mode = On]
Does this Vogons link help? In it, there's discussion about a cache bug.

Yes, it was helpful to find this site in that link,
http://www.k6plus.com/

The results seem mixed. I think I'll just have to test this board on my own. While I do not plan on using a K6+ with my P5A-B board, I was a little disapointed to see this remark from Jan Stuenebrink, "Due to a hardware problem, P5A-B Rev 1.05 will work extremely slow with a K6plus CPU."

I just so happen to have v1.05.

Has anyone tried the 1011 Beta 005 bios for this v1.05 board and found it to be extremely slow with a K6+? I wonder if disabling the motherboard's L2 cache will speed things up?

Jan's K6+ site: http://web.inter.nl.net/hcc/J.Steunebrink/k6plus.htm

Does a K6-III-450 (non-plus) have the same performance as a K6-III+ 450? Wiki mentions the non-plus K6-III also has 256KB of on chip cache. Is the K6-III-450AHX or AFX better?

feipoa wrote:

Yes, it was helpful to find this site in that link, http://www.k6plus.com/ […]
Show full quote

retro games 100 wrote:

[Quick reply mode = On]
Does this Vogons link help? In it, there's discussion about a cache bug.

Yes, it was helpful to find this site in that link,
http://www.k6plus.com/

The results seem mixed. I think I'll just have to test this board on my own. While I do not plan on using a K6+ with my P5A-B board, I was a little disapointed to see this remark from Jan Stuenebrink, "Due to a hardware problem, P5A-B Rev 1.05 will work extremely slow with a K6plus CPU."

I just so happen to have v1.05.

Has anyone tried the 1011 Beta 005 bios for this v1.05 board and found it to be extremely slow with a K6+? I wonder if disabling the motherboard's L2 cache will speed things up?

Jan's K6+ site: http://web.inter.nl.net/hcc/J.Steunebrink/k6plus.htm

Does a K6-III-450 (non-plus) have the same performance as a K6-III+ 450? Wiki mentions the non-plus K6-III also has 256KB of on chip cache. Is the K6-III-450AHX or AFX better?

The AFX is better because it produces less heat.
Afaik the K6-III and K6-III+ have the exact same performance

Tetrium wrote:

The AFX is better because it produces less heat.

But it runs at the same voltage. Does the AFX use a different technology scaling (aka micron process)?

Anyone know the safe upward limit on overclocking an AMD K6-III?

Tetrium wrote:

feipoa wrote:

Yes, it was helpful to find this site in that link, http://www.k6plus.com/ […]
Show full quote

retro games 100 wrote:

[Quick reply mode = On]
Does this Vogons link help? In it, there's discussion about a cache bug.

Yes, it was helpful to find this site in that link,
http://www.k6plus.com/

The results seem mixed. I think I'll just have to test this board on my own. While I do not plan on using a K6+ with my P5A-B board, I was a little disapointed to see this remark from Jan Stuenebrink, "Due to a hardware problem, P5A-B Rev 1.05 will work extremely slow with a K6plus CPU."

I just so happen to have v1.05.

Has anyone tried the 1011 Beta 005 bios for this v1.05 board and found it to be extremely slow with a K6+? I wonder if disabling the motherboard's L2 cache will speed things up?

Jan's K6+ site: http://web.inter.nl.net/hcc/J.Steunebrink/k6plus.htm

Does a K6-III-450 (non-plus) have the same performance as a K6-III+ 450? Wiki mentions the non-plus K6-III also has 256KB of on chip cache. Is the K6-III-450AHX or AFX better?

The AFX is better because it produces less heat.
Afaik the K6-III and K6-III+ have the exact same performance

I don't see how that could be. The K6-2+ is supposed be the old K6-III core ramped up to higher speeds and with lower voltages and it has half the cache as the K6-III+ so the K6-III should have the same performance on a per clock basis as a K6-2+, not a III+.

feipoa wrote:

Tetrium wrote:

The AFX is better because it produces less heat.

But it runs at the same voltage. Does the AFX use a different technology scaling (aka micron process)?

Anyone know the safe upward limit on overclocking an AMD K6-III?

Partially relevant info: I got an AMD K6-III+ overclocked from 400MHz to 600MHz using 2.2 volts. You can read about it here on Vogons. Please see page 2 of that thread, with "Success" in the first line of one of my posts.

Tetrium wrote:

Rev 1.04 and before has a bug in the chipset that allows only 128MB to be cached but will work with K6 mobiles (K6-2+ and K6-3+).
Rev 1.05 and beyond should cache 256MB (or more if theres 1MB L2 cache on the board) but won't work with the mobiles, so you'll be stuck with K6-2 and K6-3.

How do you tell how much L2 cache your v1.05 board has by looking at the data store cache on the PCB?

sliderider wrote:

I don't see how that could be. The K6-2+ is supposed be the old K6-III core ramped up to higher speeds and with lower voltages and it has half the cache as the K6-III+ so the K6-III should have the same performance on a per clock basis as a K6-2+, not a III+.

Wiki mentions the K6-III (non-plus) as having 256 KB of L2 cache and the K6-2+ as having 128 KB of cache. Do you have a source that says otherwise? It is possible to alter the cache size while not altering the ALU and FPU layout.
http://en.wikipedia.org/wiki/AMD_K6-III

retro games 100 wrote:

I got an AMD K6-III+ overclocked from 400MHz to 600MHz using 2.2 volts.

The K6-III+ uses a smaller micron process than the K6-III (non-plus) and is thus "easier" to overclock. Just curious what the upward limit of the K6-III (non-plus, or desktop version) might be since my P5A-B v1.05 will be limited to this CPU.

feipoa wrote:

How do you tell how much L2 cache your v1.05 board has by looking at the data store cache on the PCB? […]
Show full quote

Tetrium wrote:

Rev 1.04 and before has a bug in the chipset that allows only 128MB to be cached but will work with K6 mobiles (K6-2+ and K6-3+).
Rev 1.05 and beyond should cache 256MB (or more if theres 1MB L2 cache on the board) but won't work with the mobiles, so you'll be stuck with K6-2 and K6-3.

How do you tell how much L2 cache your v1.05 board has by looking at the data store cache on the PCB?

sliderider wrote:

I don't see how that could be. The K6-2+ is supposed be the old K6-III core ramped up to higher speeds and with lower voltages and it has half the cache as the K6-III+ so the K6-III should have the same performance on a per clock basis as a K6-2+, not a III+.

Wiki mentions the K6-III (non-plus) as having 256 KB of L2 cache and the K6-2+ as having 128 KB of cache. Do you have a source that says otherwise? It is possible to alter the cache size while not altering the ALU and FPU layout.
http://en.wikipedia.org/wiki/AMD_K6-III

retro games 100 wrote:

I got an AMD K6-III+ overclocked from 400MHz to 600MHz using 2.2 volts.

The K6-III+ uses a smaller micron process than the K6-III (non-plus) and is thus "easier" to overclock. Just curious what the upward limit of the K6-III (non-plus, or desktop version) might be since my P5A-B v1.05 will be limited to this CPU.

Generally speaking all K6-2/3 chips struggle above 600MHz because of architecture limitations (short intruction pipeline length).

DrSwizz wrote:

Generally speaking all K6-2/3 chips struggle above 600MHz because of architecture limitations (short intruction pipeline length).

that's correct but not complete. the old k6-2 core(with no on-die cache) is already capable of 550mhz with 250nm process, but k6-3 can only reach 450mhz despite voltage officially increased from 2.2 to 2.4, hindered by its on-die cache.
k6-3+ and k6-2+ are 180nm process, therefore capable of 600mhz with on-die cache, but not much higher.

noshutdown wrote:

DrSwizz wrote:

Generally speaking all K6-2/3 chips struggle above 600MHz because of architecture limitations (short intruction pipeline length).

that's correct but not complete. the old k6-2 core(with no on-die cache) is already capable of 550mhz with 250nm process, but k6-3 can only reach 450mhz despite voltage officially increased from 2.2 to 2.4, hindered by its on-die cache.
k6-3+ and k6-2+ are 180nm process, therefore capable of 600mhz with on-die cache, but not much higher.

But even then you're still better off building a Pentium II/III/Celeron or Athlon system because even overclocked to 600mhz the FPU in the K6 line is still too far behind to keep up in games. There are also issues with AGP in some SS7 motherboards and even if you have a SS7 motherboard designed for overclocking with bus speeds over 100mhz, there's no guarantee that faster memory will work with them, either. SS7 was a good, cheap alternative to Intel in it's day but it's hard to justify building one now with Intel parts being cheap enough that price is no longer an issue.

sliderider wrote:

But even then you're still better off building a Pentium II/III/Celeron or Athlon system because even overclocked to 600mhz the FPU in the K6 line is still too far behind to keep up in games. There are also issues with AGP in some SS7 motherboards and even if you have a SS7 motherboard designed for overclocking with bus speeds over 100mhz, there's no guarantee that faster memory will work with them, either. SS7 was a good, cheap alternative to Intel in it's day but it's hard to justify building one now with Intel parts being cheap enough that price is no longer an issue.

As true as that may be, finding the best board possible for the Socket 7 interface would be the best possible way to run tests on as many Socket 5/7 processors as possible on the same hardware platform, seeing that many Super7 could support chips as far back as P54C all the way to K6-III+ Overclocked, which would include Cyrix/IBM/ST chips, Rise and IDT as well. It would be a great board for Retro Benchmarking.

And as for how well the K6-2/III cores did in Games, they might not have been as fast as anything Intel, but they were still fast enough for the games of their day, especially when you find all the 3DNow patches.

This conversation sounds an awful lot like "486: The Next Generation". 😜

Amd has almost ALWAYS been the underdog in the speed category. the only real exception has been during the P4 net-burst era, and that was more because of Intel spinning their tires than anything else. That being said, AMD has almost always been the better choice when you don't have to have the "fastest". Compare the fastest chip intel has right now verses amd. Amd gets their crap pushed back in! But if you consider price, then amd is the way to go for most people.

Super 7 boards were the direct result of Intel wanting people to buy the expensive and overpriced P2 chips and motherboards. Just like the 386dx-40 and the 5x86 series chips. amd K7 chips were better than the first gen p3's but were buried when coppermine came out.

However, I think the best thing to read are old issues of maximum pc. I was a subscriber until 05' and I remember most issues. You can read the old ones for free on Google books: http://books.google.com/books/serial/ISSN:152 … &sa=N&start=120

oct 99', may 00' and mar 99' would be the most relevant ones to read. Its kinda of funny, in the oct 99' issue they refer to rambus as "something to look forward too". God hindsight is pretty awesome!

The main "cool" thing for me about the K6-2+ is the on-die L2 Cache. AFAIK every chip after this used on-die L2 Cache and motherboard cache chips became a thing of the past.

Also with on-die Cache you don't have to worry about "cachable memory size". It simply doesn't matter.

Looking at the benchmarks, RAM gets accessed with 160MB/s and the mainboard cache with ~ twice that speed.

On-die L2 cache however is ~ 10x as fast as accessing RAM. So the double speed of the mainboard cache pales in comparison.

So my tip is to not worry about the cachable size and just get a board that is compatible (BIOS) with a K6-2+ or K6-3+.

Another cool thing is the unlocked multi. Through a CONFIG.SYS driver you could change the CPU multi when booting. And under Windows there are GUI tools available for this as well.

You can also set the multi through jumpers of course, giving you a very large frequency range on the higher clocked chips.

So for time-machine / slowing down with disabling internal cache projects this chip is very interesting as you can play old games, but also games using 3D acceleration.

One thing that's important though is that disabling L1 cache in Bios means "disable all on-die cache". So unlike on a pentium, you can't have L1 disabled and L2 enabled (mainboard cache is now L3). This means that you can only simulate a 386DX and not a 486DX2 like you can with a Pentium, K6 or Cyrix.

Having said that, the flexibility of also running newer games (and not being as FPS limited as with a Pentium on a Voodoo 2) is quite cool.

An older thread which should be useful: Super Socket 7 - VIA MVP3 or ALi Aladdin V?

Through BIOS updates I got a K6-2+ 550 working on two of my Super Socket mainboards.

noshutdown wrote:

The old k6-2 core(with no on-die cache) is already capable of 550mhz with 250nm process, but k6-3 can only reach 450mhz despite voltage officially increased from 2.2 to 2.4, hindered by its on-die cache. k6-3+ and k6-2+ are 180nm process, therefore capable of 600mhz with on-die cache, but not much higher.

Anyone have long term success running an AMD K6-III-450AHX (or AFX) at 500 MHz? What voltage was necessary to facilitate this speed and for how long was the system run?