This motherboard doesn't support FPM ram. Only SDRAM is possible according to manual.
Low bandwitch may be caused by memory interlave not enabled. https://soggi.org/files/misc/tools/Memory-Int … bler-readme.htm

You may overcome this using driver mentioned above (windows), set memory interleave in BIOS, or if there isn't such option in BIOS - patch it by bios patcher from rom.by http://www.rom.by/articles/BP/index_english.htm

You can also tweak memory timings in the bios to improve the performance.

That’s about right.
My k6-lll+@550mhz scored 616 in Speedsys.
Not much of a difference.

Nice build.

I would make sure you apply the Win98 update patches for the K6 CPU.

Link to my build :
AMD K6-III+@550mhz., Voodoo-3-3000 (tower)

Thanks for all the replys!

@zyga64

Ohh... I thought the "FastVid" utility was the only "enabler" necessary back in these times; didn't know of this RAM problem.

I am almost exclusively interested i DOS rather than Windows, so I'd have to find something which is working with pure DOS (and also on a raw machine; essential source code would be nice). Otherwise I have to hope for a BIOS solution, as you suggests.

But will it really make a huge difference anyway, considered that the lower right corner of the graph show such low numbers for the (uncached) Read/Write as well as Moving(copying). The graph for the faster machine is a bit faster with the Moves, and substantially faster when it comes to Reading (I can't see the Write graph, hiding behind another one).

So I'm thinking that, since the Reads and Writes should be unaffected by the interleaving problem (I presume, according to my reading), then there are no chance of anything becoming much faster than the level the Reads are already at.(?)

@Intel486dx33

I had a look at your nice thread about your build.

I could see that your RAM bandwidth was also about the same as the computer I refer to here (sadly I couldn't see the graphs, though).

Maybe RAM performance like this used to be the norm, and I have had a bit too high expectations coming into this.

I may add that what retro knowledge I've had is a bit rusty by now, and the Socket 4/5/7 era has always been a bit a bit of a "black hole" for me. It seems like I have a lot of reading to do...

Still... The faster machine had so much better RAM performance that I may eventially have to go with something like that, even if it has other drawbacks, like sub-100MHz bus etc. Well, unless I can indeed find a great SS7 motherboard, of course.

----

As for the advertised system - The nice big tower plus a Voodoo3 (presumably stable at 100MHz), and the rest that it comes with is probably worth the reserve price; but I will not "chase" it if the biddings get going. We'll see how it goes. At least it would be a good start to get me going, since it's a complete system (I'll just add one of my 250MB Zip drives I bought previously).

----

PS. I can often be a slow responder, since I'm not always at the computer and I don't use mobile devices. I still appreciate anyones efforts, though.

I have three SS7 Diamond Flowers and all have their original capacitors, they're good motherboards, basically the Asus of their day. With this setup, k6+ and voodoo3, you should be focusing on optimizing memory.

I notice the single DIMM is installed in memory slot 3 and normally you would use slot 1. Also the heatsink on the Voodoo is upgraded suggesting the owner had their hands in there, so there's the potential for an amateur disaster scenario.

I did some tests using an older version of speedsys. It shows that I can't simulate your numbers by simply changing slots, memory settings, or by using a high density stick, and that slot 3 appears marginally slower than slot 1.

Default 512 MB (2x256 MB):

BIOS:
Timings 'Turbo'
CAS 2

Speedsys 4.70:
index 299
throughput 170 MB/s

256 MB in slot 1, slowest settings:

BIOS:
Timings 'Normal'
CAS 3

Speedsys 4.70:
index 267
throughput 160 MB/s

256 MB slot 3, slowest settings:

BIOS:
Timings 'Normal'
CAS 3

Speedsys 4.70:
index 267
throughput 154 MB/s

High density 512 MB slot 3, slowest settings (only 256 MB is recognised):

BIOS:
Timings 'Normal'
CAS 3

Speedsys 4.70:
index 267
throughput 154 MB/s

Turbo Button wrote on 2020-12-24, 02:30:

Thanks for all the replys! […]
Show full quote

Thanks for all the replys!

@zyga64

Ohh... I thought the "FastVid" utility was the only "enabler" necessary back in these times; didn't know of this RAM problem.

I am almost exclusively interested i DOS rather than Windows, so I'd have to find something which is working with pure DOS (and also on a raw machine; essential source code would be nice). Otherwise I have to hope for a BIOS solution, as you suggests.

But will it really make a huge difference anyway, considered that the lower right corner of the graph show such low numbers for the (uncached) Read/Write as well as Moving(copying). The graph for the faster machine is a bit faster with the Moves, and substantially faster when it comes to Reading (I can't see the Write graph, hiding behind another one).

So I'm thinking that, since the Reads and Writes should be unaffected by the interleaving problem (I presume, according to my reading), then there are no chance of anything becoming much faster than the level the Reads are already at.(?)

I remember quite a big performance increase after applying patch on VIA Apollo (slot1/socket370) motherboards back in time.
But MVP3 is also mentioned, so why not to try 😀
I think before BIOS patching, you may try TWKBIOS utility https://www.majorgeeks.com/files/details/tweakbios.html. It works with pure DOS.

I have the same board (DFI P5BV3+) with a K6-2+ 533 @ 600MHz, 128 MB SDRAM PC133, a GeForce 2 MX, a Creative Sound Blaster Live SB0220 PCI & an ESS Audiodrive ES1868F. This build is extremely stable and pretty fast as far as SS7 goes.
I also tested more power hungry video cards (such as the Voodoo 3 3000) and they also work fine. Overall, have had no issues with it, works great in both Windows & DOS. So I can comfortably recommend it 😀

I don't have this same board, but I happen to have a FIC VA-503+ on my test bench at the moment, which uses the same VIA MVP3 chipset, so I ran a couple benchmarks on it. Check the attached image for my speedsys results I get with optimised BIOS memory settings (Turbo timings; Cycle Length 2; 4 Bank Interleave; Page Mode, Sustained 3T Write, Write Cache Pipeline, DRAM Read Pipeline, Read Around Write all enabled; MD-to-HD Pop + 1T and Sustain 4T Cycle disabled). Switching all those settings to their slowest/safest values, cuts my memory bandwidth/throughput almost exactly in half. So it pays off to take a couple minutes to optimise your BIOS settings. The pictures in the OP show results only a bit higher than my worst settings, so the seller clearly didn't optimise much (or is using a slow memory stick that doesn't support faster settings)

Looking at the manual for the DFI board, it's missing a couple of the settings the FIC has, but with the ones it does have (timings, interleave, cycle length, 3T write, etc.) it should still get close to these results. And you can always use a program like TweakBios, which someone already mentioned, to tweak settings not available in the BIOS.

frudi wrote on 2020-12-26, 00:32:

I don't have this same board, but I happen to have a FIC VA-503+ on my test bench at the moment, which uses the same VIA MVP3 chipset, so I ran a couple benchmarks on it.

Great motherboard ! Highly suggest anyone looking for a good SS7 to try and find one, though it will not be cheap now-a-days....

Hi, again!

In the end I did not bid on the computer in question, but from what I can read here now, it isn't a bad system. But since the seller has relisted the auction - now with the starting bid set ~17% lower than before - I'll buy it, if price do not jump. (It expires January 9:th)

Reading the SpeedSys outputs were a little confusing to begin with, but I "eye-balled" a few graphs after writing my last reply and IIRC I concluded the following:

The L1/L2/L3/Throughput data seem to represent an average of Reading, Writing and Moving.
When MMX is supported, the highest value of MMX and non-MMX instruction usage for the respective Reading, Writing and Moving type of operation is selected before averaging together these three selected numbers.

(I'll double check later, and update if I'm wrong)

As for Memory Bandwidth, I have no idea.

________

If Time (and "life") permits, I may try to get back into programming by writing my own benchmark program. But it may take a while (Somehow I'm a heavy user of "The back burner", despite the fact we hardly have any actual "burners" in Sweden).

That way I'll have full control over what is being measured, and how it is being measured.

What pops to mind is for the program to...
* Measure BLOCK-COPY ("REP MOVSD" instruction I guess). Possibly "MMX MOV" (I know; not the correct mnemonic)
* Block Copy: Memory->Memory and Memory->GFX.
* Use a blown up version of the SpeedSys style graph.
* Measurement 1: Huge BLOCK-COPIES that are non-cached. Representing worst case scenario.
* Measurement 2: Medium BLOCK-COPIES representing 32-bit 2D gaming and heavy graphical Applications/Operating systems.
* Measurement 3: Small BLOCK-COPIES representing older 16-bit 2D gaming.
* Measurement 4: Simplified DOOM-style('ish) rendering with minimal calculations per pixel representing 3D gaming.
* Measurement 5: Single random Writes/Reads. (Low priority; I'm obsessed with the 2D bitmaps 😉 )
* The Block sizes in measurement 2 and 3 above will not be fixed. During a run they will vary in size such that the total result with all the different blocks within the selected range (test category) are weighted similar to a mathematical "Normal distribution" of different sizes. This should, to a degree, smooth out any potential sudden jumps in the test results between machines where the cache sizes are not dramatically different.
* An 800x600 mode, if supported by VESA BIOS / VBE. With the graph being a pixel-perfect copy (from the standard 640x480 mode) it leaves room for less important data to be presented on screen.

But now I'm derailing. Maybe I'll open a thread about this when it's time.

Now it's rather time to answer the latest posts:
________

@Maestro

Thank you for conducting these tests!

@zyga64

Thanks' for the suggestion. I'll try the TWKBIOS utility if I win the auction.

@bloodem

You and others seem to think it is a rather solid board, so I'm interested.

@frudi

That was a nice little collection of data/info you provided there. Much appreciated.

It shows a really great improvement - Just what I hoped for! At around those speeds there aren't any imminent need for replacing the motherboard anymore.

Your cached numbers are practically a carbon copy of the ones in my first screenshot (but yours are a hair better). But beyond the realm of the L3 cache it seems to be evident that your tweaking shines, with a Memory throughput of 205MB/s vs. the 115MB/s screenshot I referred to.

As for the SpeedSys "Memory bandwidth", it goes from 176MB/s to 302MB/s. But I don't know how this number is derived, so for now I'm focusing on the graphs in the lower right corner. They are quite informative.

@Horun

Yet another testimonial. It looks like I may just get one. 😉

________

PS. Because of a "computer situation", I'm sharing a PC with others. I also don't log off, actually this computer may go for weeks without being completely shut off by anyone. So therefore I'm disabling the "online" indicator, because it can't possibly be showing a correct status. Sorry for any confusion!