Here are the register differences I've noticed between the VIA (HOT-591P) and ALi (ASUS-P5A-B) motherboards,

1VIA	ALI	FEATURE
2
31	   0     NoLock (CCR1)
40	   1     SADS (CCR2)
50	   1     LINBRST (CCR3)
61	   0     TOGGLE (CCR4)
70	   1     WTALLOC (CCR5)
80	   1     RCD (RCR2)
90	   1     RCD (RCR4)
100	   1     RCD (RCR5)
110	   1     RCD (RCR6)
121	   0     WL (RCR7)
1311	  00    BIT 37 & 38 (MSR10 MSB)
14SEVERAL      Several BITS (MSR10 LSB)

I haven't studied in depth what each feature is responsible for, but the information can be found in the BIOS WRITER'S GUIDE for the MII.

What is surprising is that LINBRST is disabled for the VIA board and enabled for the ALi board.

I have attached Speedsys images of the Cyrix MII-433 in both the VIA and ALI board's. Something funny (slow) is going on with the 'Move' speed graph. The overall Speedsys score of the MII-433 seems to weigh in at about a Cyrix 6x86-200PR, which is very peculiar. I wonder if disabling CPUID screws up the Speedsys score? The MII-433 should be around a Pentium II-400.

I tried changing the Cyrix register settings of the VIA board onto the ALi board, but it didn't speed things up.

LINBRST and TOGGLE are locked and cannot be changed.
SADS caused system hang-up upon changing its value.
NOLOCK and WTALLOC were okay to change, but didn't help.

Both motherboards seem to function properly with an AMD K6-2 using the same settings as the Cyrix MII-433 (3 x 100 MHz). The performance of the two boards were similar as shown in the Speedsys screen shots below. Using the AMD K6-2-300, buth motherboards had similar PCPBench and 3DBench2 results [3DBench2: ALi = 307, VIA = 303]; [PCPBench: ALi = 40.0, VIA = 37.3].

When comparing the VIA Cyrix MII-433 to the VIA AMD K6-2-300 using Landmark v2.0, the MII scored 3628/4076 (ALU/FPU), while the K6-2 scored 3083/4929 (ALU/FPU). So clock-for-clock, the MII is a bit better than the K6-2 at ALU and a little worse at FPU. As an average, they are about equivalent. PCPBench scores for both CPUs were 37.

On a Semirelated, possibly unrelated note, has anyone ever gotten Vista/7 to run on an ALi Aladdin V board? I got a Superpower A586B yesterday, put 512mb RAM, K6-2 500 and a GF4 Ti4200 on it. I just tried getting 7 to install on it, but can't recognise the IDE controller.

The issue with the Cyrix MII and the VIA MVP3 has been semi-resolved. This particular HOT-591P (MVP3) doesn't turn on write allocate for the MII in the BIOS. When turning on Write Allocate via software, the WRITE profile of the Speedsys image greatly improves, as follows.

The FIC PA-2013 board I have natively enables on Write Allocate in the BIOS for the Cyrix MII's.

Note: I'm using a 100 MHz FSB and a 3x multiplier for the MII.
This is the only SS7 board I've tested that seems to work with a Matrox Parhelia 128 AGP, as noted on the Speedsys image.

It is still not clear to me why such a large gap between L1 read and write exists.

So no word about Vista/7 on an ALi Aladdin V chipset then, eh?

I don't own a Vista or Win7 license, so someone else will have to test for this.

maddmaxstar wrote:

On a Semirelated, possibly unrelated note, has anyone ever gotten Vista/7 to run on an ALi Aladdin V board? I got a Superpower A586B yesterday, put 512mb RAM, K6-2 500 and a GF4 Ti4200 on it. I just tried getting 7 to install on it, but can't recognise the IDE controller.

Well thats pretty ambitious 😊 You can try plugging in a PCI IDE controler instead. Like HPT370 or Promise Fasttrack - whichever has a Win7 driver.

feipoa wrote:

[Note, if your BIOS does not have an option to disable CPUID, you must disable CPUID in software to run SpeedSys with an MII].

How to do this feipoa?

sebaz_ri wrote:

feipoa wrote:

[Note, if your BIOS does not have an option to disable CPUID, you must disable CPUID in software to run SpeedSys with an MII].

How to do this feipoa?

There are programs which you can use to alter the register settings of Cyrix 6x86, 6x86MX, and MII CPUs.

A few names come to mind, such as 6x86cfg, 6x86ctl, Fast'n'Slow, 686Fast, 6x86opt, and IBM's utility. The easiest one to use to just turn off the CPUID is IBM's utility. Use ibmm1.exe for 6x86 CPUs and ibmm2.exe for 6x86MX and MII CPU's. Both of these files can be found on chipdb.org within the document archive section. Once you disable the CPUID, you can then run Speedsys, but in some cases, you may still need to run Speedsys with /sp to turn off PnP detection.

I happen to own both the Asus P5A and P5A-B. They might even be the 1MB L2 cache versions. Windows 2000 runs fine on them, but I couldn't get XP working on the P5A-B (forced APM mode with AT power supply) without it bombing out with weird errors in setup. I went with the Ali chipset when I bought the boards because of my past horrible experiences with VIA. They generally worked well with Windows 98SE.

feipoa wrote:

The VIA board gets a Speedsys score of 215, whereas the ALi board shows 128. BYTEMark also shows about half speed for the ALi board.

This post got me worried. Fortunately the poor performance is not present on this ALi 5 board I just tested. A Chaintech CT-5RSA0. It is exactly on par with VIA MVP3. I don't know what is wrong with your 'Asus P5A-B v1.05' / MII combination. Its ATX brother 'Asus P5A rev 1.04' also works fine, When I had it setup with the same MIIV at 3.5x66=233Mhz it scored 167,25 points. Not the same configuration, but already more then 128 points. Both Aladdin 5 boards contain the rev. 'E' North Bridge. Yours being a 'G'.

It was certainly puzzling. That revision of the motherboard I used seems to be best suited for Intel processors and non-K6+ chips.

feipoa wrote:

Upon further investigation, I noticed that SpeedSys shows the graphics card as 8x AGP. It seems like it would be more useful if […]
Show full quote

Mau1wurf1977 wrote:

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

Upon further investigation, I noticed that SpeedSys shows the graphics card as 8x AGP. It seems like it would be more useful if SpeedSys indicated what mode the graphics card was currently operating in, like 2x in this case.

I just uploaded two photos of my competing VIA MVP3 and ALi Aladdin V motherboards. I noticed that the TAG RAM on this HOT board is only 8 ns (-8). The comparison PDF I uploaded in the initial post seems to indicate that this mediocre cache speed (of 8 or 10 ns) may be problematic at 100 MHz.

Mau1wurf1977, may I know what the TAG RAM speed is on your super7 motherboards?

Anyone who has had issues with VIA MVP3 boards, do you happen to know what the TAG RAM speed is? For those who haven't had issues with VIA MVP3 boards, do you know what the TAG RAM speed is?

The 686 benchmark trials are long completed, but this still seems like a worthwhile topic to address.

From the few SS7 motherboards that I've seen in person (and my current P5A-B 1.04 and Dad's old FIC 503A system in the garage), 8 ns tag ram seems common while 4 ns cache SSRAM is common on MVP3 boards (or at least FIC boards) compared to 5 ns cache on most Aladdin V based boards I've seen (including all legible photos of P5A and P5A-B boards I've seen).

Do note that tag RAM uses normal asynchronous SRAMs (the rating is for the access time but the cycle time is equally fast) compared to synchronous SRAMs used for burst-mode cache where the cycle time is twice that of the access time. (thus 100 MHz cache needs to be at least 5 ns to run within specification and the 4 ns chips FIC boards used were rated for a nominal 125 MHz -that also means overclocking boards with 5 ns cache is genuinely pushing the cach chips out of spec). 8 ns Asynch SRAM is within spec up to 125 MHz bus operation.

I think a lot of the comparison articles from back then (and indeed Anandtech's old article here http://www.anandtech.com/show/72/19 ) buys into over-conservative specification hype from the time and not real-world experience and testing. (remember that the official PC-100 specification adopted 8 ns SDRAM chips and claimed 10 ns chips would risk instability, but just a few years later it was common to use 7.5 ns SDRAMs on 133 MHz rated modules -not to mention video cards running their SDRAMs at their specified clock rates and not a conservative underclock; granted some PC-133 SDRAM did use 7 ns chips, but that's a very narrow added gap compared to the 10 vs 8 ns one)

The Anandtech article is a bit misleading in places as well (on the MVP3 listing a likely 64-256 MB cacheable range, which would imply 256k-1MB SSRAM arrangements when 512k-2MB options had already been promoted, meaning 128-512MB ranges). Their page for the Aladdin V chipset of that same article omits the on-die tag RAM bugs entirely, promoting only the theoretical gains in performance and cost without noting the need for external tag ram to be used anyway on early chipset revisions (or indeed the majority of the chipset's mainstream lifespan). And with 8-bit external tag ram used, it limited cacheable range to just 128 MB with 512k cache rather than the 512MB allowed with 10-bit tag RAM. (I'm not sure if any boards used external 10-bit tag RAM or if the chipset actually supported such)

I also assume that the chipset is only making use of 16k of the 32kx8-bit SRAMs used for tag given the internal logic would have been based around a 16k/14-bit address. (with the MVP3 designed expressly for external tag RAM, it might use the full 15-bit range and finer segmentation of cache index blocks, but I'm not sure on that; or rather finer blocks for any given cache size, given larger caches would use proportionally larger blocks)

Of course, they also oddly list the 1 GB cache range on the Aladdin V in combination with 256 kB L2 SSRAM.
http://www.anandtech.com/show/72/3

I'm not sure why 1 MB cache Aladdin V boards were so uncommon, unless that was also a limitation of the buggy initial revisions.

Sorry for the necropost.
I'm doing some test with what I consider the best of both chipsets:
Epox Mvp3g5 and Gigabyte GA-5AX (rev 5.2) both with following cpu: 6x86MX 250, MII 250, MII 300, K6-2 500, K6-2+ 500, K6-3+ 500, Pentium MMX 300. Cyrix frequency are real I have the mx pr266 (v2.7) and the 333gp low voltage, every test will be done at 100 mhz bus.
Ram: 128mb 7ns
Vga: Geforce 2 mx400, I have been adviced to use this one because is less problematic than TNT2 M64, so far I can say it's true.
Hard drive: Barracuda 7200.10 160gb (5gb for windows partition)
No sound card, no cd drive.
Windows 98 SE.
Drivers: Via 4in1 v4.26, Ali v2.13
I hope that I won't have issues with Cyrix cpu and speedsys, I regret that I sold the TMC ti5vg+ it had a bios option that let me start speedsys without issues.

For Cyrix MII chips and running Speedsys, you might need to disable CPUID. There might also be a speedsys flag to use, but I don't recall off the top of my head now.

feipoa wrote:

For Cyrix MII chips and running Speedsys, you might need to disable CPUID. There might also be a speedsys flag to use, but I don't recall off the top of my head now.

Yes I read your advice in this topic, I just hope that when I will put the cyrix cpu in there will be an option to disable it from bios, like the tmc mb.
Also on Epox there should be an option in the bios to enable it.

Nemo1985 wrote:

Sorry for the necropost. I'm doing some test with what I consider the best of both chipsets: Epox Mvp3g5 and Gigabyte GA-5AX (re […]
Show full quote

Sorry for the necropost.
I'm doing some test with what I consider the best of both chipsets:
Epox Mvp3g5 and Gigabyte GA-5AX (rev 5.2) both with following cpu: 6x86MX 250, MII 250, MII 300, K6-2 500, K6-2+ 500, K6-3+ 500, Pentium MMX 300. Cyrix frequency are real I have the mx pr266 (v2.7) and the 333gp low voltage, every test will be done at 100 mhz bus.
Ram: 128mb 7ns
Vga: Geforce 2 mx400, I have been adviced to use this one because is less problematic than TNT2 M64, so far I can say it's true.
Hard drive: Barracuda 7200.10 160gb (5gb for windows partition)
No sound card, no cd drive.
Windows 98 SE.
Drivers: Via 4in1 v4.26, Ali v2.13
I hope that I won't have issues with Cyrix cpu and speedsys, I regret that I sold the TMC ti5vg+ it had a bios option that let me start speedsys without issues.

I can agree on the Epox board, I only have Mvp3g2 version but it is quick and seems to be a tiny bit faster as my DFI and Aopen board with the same chipset. (Epox 1mb vs 512kb DFI and Aopen).
However...the Ga-5ax is a very nice board but it is not the fastest. At least in my 3dmark99 and 2000 testing and also Everest benchmarks the Asus P5A was faster and P5A-B the fastest. Also have the Ga-5aa but it can like it's atx brother not beat the Asus boards.

Ali V is faster in memoryspeed and agp performance/3d, that was my conclusion. Happy to see what your results are! I did most of my tests with AMD k6-3+ and a few with mmx.

meljor wrote:

I can agree on the Epox board, I only have Mvp3g2 version but it is quick and seems to be a tiny bit faster as my DFI and Aopen board with the same chipset. (Epox 1mb vs 512kb DFI and Aopen).
However...the Ga-5ax is a very nice board but it is not the fastest. At least in my 3dmark99 and 2000 testing and also Everest benchmarks the Asus P5A was faster and P5A-B the fastest. Also have the Ga-5aa but it can like it's atx brother not beat the Asus boards.

Ali V is faster in memoryspeed and agp performance/3d, that was my conclusion. Happy to see what your results are! I did most of my tests with AMD k6-3+ and a few with mmx.

I tested the Asus P5A (rev 103) that before selling it, it was slower than Epox, but I tested it just with k6-2 500, which revision did you try?
For sure the Gigabyte has more fsb options the one I bought (well the whole system actually) has a k6-3+ at 633 mhz (115x5.5). I will compare the memory results with the p5a, it will be interesting.

I have some bad news: The Tillamook doesn't boot, it's weird because it was working on the similar Epox, while with cyrix, there is no option for linear burst (not even with modbin is visible, just wiped out, even if there is one on the manual) but there is another voice: Cyrix M2 ADS# delay, despite setting it to enabled or disabled the speedsys doesn't work properly.
Any suggestion?

I've finally completed the round up with the first mb.

Here is some notes:
1) The Tillamook won't work with latest bios, I've asked if it's possible to add the microcodes and i'm waiting for an answer, which may arrive or not, it's weird because with the other and older Epox (EP-51MVP3E-M) it was working fine.

2) The Cyrix cpus, I wish to thank feipoa for his precious help. I can confirm that 6x86mx with higher voltage are faster than M2 rated v2.2 at the same clock speed, unlucky my pr266 is unstable at 250 mhz so I wasn't able to run the full test suite, despite I tried to run it at 2.9 instead of stock voltage of 2.7.

3) There is no way to enable the linear burst on this mb, despite the mb manual says there is the voice in the bios, it has completely removed and any attempt to enable it freeze the pc.

4) Despite using very fast onboard cache there is no way to go over 100mhz fsb, I did different tries, even disable the onboard cache didn't help, the mb anyway supports the following bus over 100: bus/pci 112/37, 124/41, 133/44.

5) I didn't suffer any incompatibility issue, the only problem I had it was with geforce 2 mx400 when used with a voodoo 1, crashes, bsod and random freeze.

6) The difference between k6-2+ and 3+ is very small, but when I overclocked the first at 600 mhz it did 10 fps more in quake 320.

That's it for the mvp3, I will test the Gigabyte in the following days

GF2 has no problem? Only a problem with GF2+V1. What about GF2 + Voodoo2?