Tetrium wrote on 2020-05-27, 15:55:

And besides, newer parts don't automatically equal more performance anyway. Not when it comes to 486 machines.

That's not what I was saying.

I think the more accurate thing to say is that more modern PCI peripherals on a 486 offer better performance up to a point at which they become bottlenecked by the processor and chipset.

On a 486, you will certainly get faster graphics performance from a PCI S3 Trio64+ than a VLB Cirrus Logic CL-GD5428, but not much past that makes any real sense.

I'd certainly not try to stick a Voodoo or a Riva TNT in a 486 and expect some kind of miracle.

boxpressed wrote on 2020-05-27, 16:02:

So would you say that PCI video cards newer than your Diamond card offer diminishing--if any--gains? I have a Socket 3 board with VLB and PCI and use a Virge 325 2MB just because it is small and has great compatibility. Use DOS 95% of the time.

I would say that's exactly right. Newer graphics cards on a 486 will only get you so far before they're bottlenecked by the processor and chipset.

I think your Virge card is in the sweet spot on and going any further will not gain you much more (if any) performance. S3 cards from that era are the most DOS-compatible cards I have ever tested.

In fact, if you go with a 4MB card and try to run a resolution past 1024x768 with true color, the CPU will struggle to keep up even doing 2D things.

appiah4 wrote on 2020-05-27, 13:23:

And compare I will.

And very interested I am. 😀

My cache tests where done with the SV2GX4 only. And i forgot the tests without L2 ...

In theory you could also argue something like the Virge or Voodoo helps performance as it takes work off the CPU in games that support it.
I say in theory as in those same games even the fastest 486 is bottlneckng the game enough not to call plying it enjoyable.

Really VLB is desired as it's unique to the 486 more then performance gains (yes I know it also featured on 1 or 2 Pentium M/B's but it was already legacy be then)
PCI is more practical in just about every way, but, it's also kinda, well boring

starcube wrote on 2020-05-27, 16:12:

That's not what I was saying. […]
Show full quote

Tetrium wrote on 2020-05-27, 15:55:

And besides, newer parts don't automatically equal more performance anyway. Not when it comes to 486 machines.

That's not what I was saying.

I think the more accurate thing to say is that more modern PCI peripherals on a 486 offer better performance up to a point at which they become bottlenecked by the processor and chipset.

On a 486, you will certainly get faster graphics performance from a PCI S3 Trio64+ than a VLB Cirrus Logic CL-GD5428, but not much past that makes any real sense.

I'd certainly not try to stick a Voodoo or a Riva TNT in a 486 and expect some kind of miracle.

In this case you don’t compare local bus topologies. You compare a 64bit S3 graphics card from 1995 with a 32bit CL card from 1993 on different bus topologies. I have a S3 968 VLB, lets compare it with a S3 928 PCI. You toggle with too many parameters to get any reliable statement.
Ok, a S3 968 VLB is nearly unobtainable and the PCI pendant is easy to obtain for 20€ and below. That's advantage of PCI over VLB for a 486. According to general opinion, the graphics chips based on S3 Vision line (864/964/868/968/Trio/Virge) are the optimum (speed and compatibility) for 486 purposes.

In my experience, performance difference between PCI and VLB cards with the same chipset is negligible in a 486, especially considering that most cards that come in the 2 flavors are pretty much bottlenecked by the CPU.

In games and benchmarks I get pretty much the same results on a VLB Trio64 (using an ASUS VL/I-486SV2GX4) than a PCI Trio64 (using a Gigabyte 486AM/S) with a DX4-100 WB. Considering these 2 boards are among the fastest for VLB and PCI respectively, I don't think boards with other chipsets can make too much of a difference.

While newer PCI cards may be potentially faster, it won't be noticeable on a 100MHz 486. The main advantage of PCI vs VLB in a 486 is the greater variety of options, availability and much lower cost.

PC-Engineer wrote on 2020-05-27, 17:05:

In this case you don’t compare local bus topologies.

...but I wasn't comparing just bus topologies.

I was comparing the practical application of VLB vs PCI in a 486 system, which includes the advantage of faster peripherals available for the latter.

And the original discussion was actually about whole-system performance where the chipset matters more than just the bus itself.

You can have a slow PCI system and a fast VLB system built around the same processor, and vice versa, depending on your choice of chipset and peripherals.

Things like the performance of the integrated memory and disk controllers, the individual chipset's implementation of the local bus, supported memory types and speeds, and so on.

It's about how the entire system performs, not just what bus it has.

starcube wrote on 2020-05-27, 17:50:

...but I wasn't comparing just bus topologies. […]
Show full quote

PC-Engineer wrote on 2020-05-27, 17:05:

In this case you don’t compare local bus topologies.

...but I wasn't comparing just bus topologies.

I was comparing the practical application of VLB vs PCI in a 486 system, which includes the advantage of faster peripherals available for the latter.

And the original discussion was actually about whole-system performance where the chipset matters more than just the bus itself.

You can have a slow PCI system and a fast VLB system built around the same processor, and vice versa, depending on your choice of chipset and peripherals.

Things like the performance of the integrated memory and disk controllers, the individual chipset's implementation of the local bus, supported memory types and speeds, and so on.

It's about how the entire system performs, not just what bus it has.

Then i must have misunderstood you:

starcube wrote on 2020-05-24, 19:29:

And for the record:
2. PCI actually makes the 486 significantly faster over VLB, as both the IDE controller and the graphics card benefit from the much faster bus.

Nevertheless I cannot agree your statement. I have the same experiences like TheMobRules.
You did a very good job with your system. Do you need such vague thesis to justify your system?

PC-Engineer wrote on 2020-05-27, 18:48:

Nevertheless I cannot agree your statement. I have the same experiences like TheMobRules.
You did a very good job with your system. Do you need such vague thesis to justify your system?

And I cannot agree with yours. But I am not seeking validation from anyone here. My system makes me happy and that's all that matters to me.

This whole discussion started because "Intel486dx33" insisted there's no difference between his and my system whereas benchmark results clearly showed otherwise.

My mistake was continuing to argue with him, dragging in others in the process, even after I said I was done. I apologize.

I was always under the impression VLB could be a few ticks faster than janky-early PCI implementations found on some 486s boards personally.

Great to see another PCI 486-100mhz build around here though! I know they get a bit of flack for being a bit "boring", but regardless of what the VLB-Squad (or maybe VLBros?) may say, there is no denying the flexibility in a good PCI board! 😁

Love the case and how well organized you wired everything inside! 😀

starcube wrote on 2020-05-27, 16:12:

That's not what I was saying. […]
Show full quote

Tetrium wrote on 2020-05-27, 15:55:

And besides, newer parts don't automatically equal more performance anyway. Not when it comes to 486 machines.

That's not what I was saying.

I think the more accurate thing to say is that more modern PCI peripherals on a 486 offer better performance up to a point at which they become bottlenecked by the processor and chipset.

On a 486, you will certainly get faster graphics performance from a PCI S3 Trio64+ than a VLB Cirrus Logic CL-GD5428, but not much past that makes any real sense.

I'd certainly not try to stick a Voodoo or a Riva TNT in a 486 and expect some kind of miracle.

Better to compare a PCI Trio64 to a VLB Trio64. Should be roughly the same performance but the PCI version should be relatively easy to find, but I think we agree with each other on this anyway 😜
There's even significant differences between different boards and different chipsets in how (and how well) PCI/VLB were implemented.

chinny22 wrote on 2020-05-27, 16:37:

In theory you could also argue something like the Virge or Voodoo helps performance as it takes work off the CPU in games that s […]
Show full quote

In theory you could also argue something like the Virge or Voodoo helps performance as it takes work off the CPU in games that support it.
I say in theory as in those same games even the fastest 486 is bottlneckng the game enough not to call plying it enjoyable.

Really VLB is desired as it's unique to the 486 more then performance gains (yes I know it also featured on 1 or 2 Pentium M/B's but it was already legacy be then)
PCI is more practical in just about every way, but, it's also kinda, well boring

^This

PC-Engineer wrote on 2020-05-27, 17:05:

starcube wrote on 2020-05-27, 16:12:

That's not what I was saying. […]
Show full quote

Tetrium wrote on 2020-05-27, 15:55:

And besides, newer parts don't automatically equal more performance anyway. Not when it comes to 486 machines.

That's not what I was saying.

I think the more accurate thing to say is that more modern PCI peripherals on a 486 offer better performance up to a point at which they become bottlenecked by the processor and chipset.

On a 486, you will certainly get faster graphics performance from a PCI S3 Trio64+ than a VLB Cirrus Logic CL-GD5428, but not much past that makes any real sense.

I'd certainly not try to stick a Voodoo or a Riva TNT in a 486 and expect some kind of miracle.

In this case you don’t compare local bus topologies. You compare a 64bit S3 graphics card from 1995 with a 32bit CL card from 1993 on different bus topologies. I have a S3 968 VLB, lets compare it with a S3 928 PCI. You toggle with too many parameters to get any reliable statement.
Ok, a S3 968 VLB is nearly unobtainable and the PCI pendant is easy to obtain for 20€ and below. That's advantage of PCI over VLB for a 486. According to general opinion, the graphics chips based on S3 Vision line (864/964/868/968/Trio/Virge) are the optimum (speed and compatibility) for 486 purposes.

^And this

On 486 class machines vlb adapters are for sure faster at dos graphics than pci counterparts, considering more or less similar time period - to say 1993 - 1996 and even early 1997.
For windows graphics is the opposite: pci > vlb.

If you test this on pci/vlb mobo - numbers will be somewhat lower than when comparing "clean" pci and vlb boards.
For dos graphics asus vli sv2gx4 is da boss.
Ark1000vl is the fastest in dos graphics, with s3 trio64, wc90c33-zz and et4000w32# breathing in its neck.
For windows graphics it is matrox millennium pci by far, then s3 trio64 vlb distant second, then the rest.
S3 trio64 vlb is in general faster than, or on par with, s3 trio64v2/dx2 pci. Same applies to et4000 vlb vs pci as well.

Now, if we strap second and third gen pci video adapters to the same 486 system, they will obviously outgun the vlb ones, but if we stick to the 93 - 96/97 period, I believe the above is true.

pshipkov wrote on 2020-05-29, 07:14:

On 486 class machines vlb adapters are for sure faster at dos graphics than pci counterparts, considering more or less similar t […]
Show full quote

On 486 class machines vlb adapters are for sure faster at dos graphics than pci counterparts, considering more or less similar time period - to say 1993 - 1996 and even early 1997.
For windows graphics is the opposite: pci > vlb.

If you test this on pci/vlb mobo - numbers will be somewhat lower than when comparing "clean" pci and vlb boards.
For dos graphics asus vli sv2gx4 is da boss.
Ark1000vl is the fastest in dos graphics, with s3 trio64, wc90c33-zz and et4000w32# breathing in its neck.
For windows graphics it is matrox millennium pci by far, then s3 trio64 vlb distant second, then the rest.
S3 trio64 vlb is in general faster than, or on par with, s3 trio64v2/dx2 pci. Same applies to et4000 vlb vs pci as well.

Now, if we strap second and third gen pci video adapters to the same 486 system, they will obviously outgun the vlb ones, but if we stick to the 93 - 96/97 period, I believe the above is true.

I'm not sure about VLB being inherently faster than PCI though. VLB was a different approach compared to PCI and it certainly had its drawbacks.
Do you have a link to any kind of benchmark results for the graphics card data? Now I wanna know 😜

This is tricky to compare. One could try to use a VIP style motherboard (the one with both VL-Bus and PCI slots). However, that way you are usually penalising one or the other bus. So you have to use two different motherboards designed for VLB and PCI with similar performance. Then you also need to equalise the cards as even when using the same VGA chips.

When comparing these two Trio64 cards:

And when the differences in chipsets is normalised and when ensuring the same MCLK then the VL-Bus version is consistently faster in tasks where the host-bus performance matters (like DOOM framerate, vidspeed, ...).

The trio64 is as fast as it gets for the VL-Bus. At least in DOS. In Windows some VRAM base cards could be faster.

I also have Mach64, Mach32, Tseng ET4000 w32p and S3 864 in PCI and VL-Bus versions and see more or less the same thing. When everything else is equal the VL-Bus tends to be slightly faster.

I plan to do this properly and eventually publish results on my blog.

VLB was a different approach compared to PCI and it certainly had its drawbacks.

VLB is tied directly to FSB, so at least it has less latency. But it's hard to overclock VLB boards past 50Mhz.

mpe wrote on 2020-05-29, 12:44:

I also have Mach64, Mach32, Tseng ET4000 w32p and S3 864 in PCI and VL-Bus versions and see more or less the same thing. When everything else is equal the VL-Bus tends to be slightly faster.

PCI is still better. A slight performance trade-off, for better reliability, compatibility and cheaper manufacturing.

VLB is a cool historical item, but at the time, PCI was a badly needed replacement. The Pentium and newer chips would have not been directly compatible with VLB, and the large card size and limited slot numbers significantly limited its future potential.

Yes. That's for sure. PCI was created for a reason and it was a much better bus, more flexible, reliable, scalable, etc. However, in absolute numbers and when transferring data from CPU to slave device the local bus could be a bit faster.

Hi Tetrium,

i keep posting info from time to time here: 3 (+3 more) retro battle stations
It is not structured as a proper "research document", so some of the points may not be obvious, but if you skim through it, you will notice the pattern.
Long ago i noticed some of that stuff about VLB and PCI graphics and kept poking at it.
I actually have a bunch of data specifically for VLB/PCI graphics that is not there. Will try to post soon.
There is also few other threads around here about this matter and they suggest the same, more or less.

Tetrium wrote on 2020-05-29, 12:02:

I'm not sure about VLB being inherently faster than PCI though. VLB was a different approach compared to PCI and it certainly had its drawbacks.
Do you have a link to any kind of benchmark results for the graphics card data? Now I wanna know 😜

chinny22 wrote on 2020-05-27, 16:37:

Really VLB is desired as it's unique to the 486 more then performance gains (yes I know it also featured on 1 or 2 Pentium M/B's but it was already legacy be then)

It was featured on a lot more than 1 or 2 Pentium motherboards. Have you ever looked at a Pentium roundup from late 1993? The vast majority of systems use VLB boards. I would say at least a few dozen Pentium boards were made with VLB slots. Also, VLB was not considered legacy in 1993 or 1994 (it came out in late 1992). The reason why you never see them around is because in 1993 and 1994 the pool of buyers for Pentiums was small, and after PCI Pentiums became a thing nobody wanted the old creaky VLB boards and tossed them out.

canthearu wrote on 2020-05-29, 13:10:

VLB is a cool historical item, but at the time, PCI was a badly needed replacement. The Pentium and newer chips would have not been directly compatible with VLB, and the large card size and limited slot numbers significantly limited its future potential.

PCI was not a replacement for VLB. They came out at pretty much the same time and were actually competing standards. Until later in 1994 it was not obvious that PCI was going to be the winner. It was only when PCI 2.0 came around that things began to change, so if you want to be fair you should compare VLB 1.0 to PCI 1.0. There was a VLB 2.0 specification which addressed many issues of 1.0, but I think it wasn't really used in anything since Intel's chipset monopoly caused most other chipset vendors to throw in the towel.

SodaSuccubus wrote on 2020-05-27, 21:44:

Great to see another PCI 486-100mhz build around here though! I know they get a bit of flack for being a bit "boring", but regardless of what the VLB-Squad (or maybe VLBros?)

Where can I sign up for VLB-Squad?