Anonymous Coward wrote:

The chart shows the budget DRAM version of the Mach32, the premium VRAM version should score differently for both DOS and Windows. Most VRAM cards had a bad reputation for sub par DOS performance.

Mach32 and S3 928 were the fastest things around when they were released in 1992. This roundup is from late 1994, so it's not really fair to compare them to 64-bit chips is it?

It's not really fair, no. But for the sake of helping me choose the fastest possible vlb card, it is highly relevant though.
Reading issue 2 from 1996 makes me feel like my efforts might be futile though, as they have a build guide there, recommending none other than my current card, Matrox Millennium as the highest performer, regardless of price.
My quest is trying to find a vlb card that comes at least remotely close for a price that isn't astronomical. I mean, I found a Mach 64 for over 800 euros. Not really a price normal people could afford. 😁

The thing is the CPU on a VLB platform will be a bottleneck before a card like a ET4000 or S3 Trio. Some DOS games benefit from a Pentium 3! DirectDraw games certainly do too. As does GUI acceleration because the CPU is part of the equation with rendering applications.

What for will the card/comp be used for?

swaaye wrote:

The thing is the CPU on a VLB platform will be a bottleneck before a card like a ET4000 or S3 Trio. Some DOS games benefit from a Pentium 3! DirectDraw games certainly do too. As does GUI acceleration because the CPU is part of the equation with rendering applications.

Well, yes and no. Both those cards will probably scale in performance with a faster CPU, but when tests done on a 486DX/2-66 shows some cards performing 2-3 times faster than others, then it really doesn't matter if the CPU is a bottleneck in both cases, as the card with 3 times the performance will be superior. The slower card isn't going to scale so much better that the tables will turn at any point. And yes, some DOS games will run EVEN better on super-fast systems than they do on systems from that era, and in all honestly, that is irrelevant. If a game is simply too demanding for my CPU of choice (IntelDX4 100WB) then it isn't going to run, regardless of what graphics card I use.

kixs wrote:

What for will the card/comp be used for?

It will be used for the same things I currently use the system for, only now I'm running the system with a PCI motherboard. Primarily DOS gaming, early to mid-90's games, such as the Tex Murphy series (with emphasis on UaKM and PD), Wing Commander 3 (and possibly 4, depending on how that runs), Red Alert, all the classic lucas arts adventures... well, the list is too long to specify. Certain games however, such as the mentioned Red Alert, simply plays better in Windows 95, because it allows for higher resolution. With my current PCI setup the game runs just fine in Windows 95.

So why and I looking to rebuild a perfectly well functioning system to VLB you might ask? Because VLB "feels" more era appropriate for a 486. And because I never had a chance to explore VLB back in the days, as I went straight from an OEM (AST Advantage) 486SX with everything integrated into the motherboard, to a Pentuim 120 with obviously PCI.
Call it an experiment. If it turns out that converting this system to VLB, causes an over-all performance loss, or deterioration of user experience, then I'll just call the project a failure and go back to my PCI setup. This is why I'm also worried about having to spend too much money on this project, while also not wanting to settle for parts that are too much of a compromise.
Fortunately I'm not in a major rush, and if getting the right parts means being on the lookout for a few months, then that what I'm going to have to do. 😀

When I find the time... I'll do some more benchmarking with different VLB cards. But I don't have any late VLB chipsets, only CL-5428, Tseng ET4000/w32i, S3-805 and some ALi.

Do you recommend any benchmark (app or game) for Win95? I usually bench with Wintach and Winbench 3.1. But these are not DirectDraw/DirectX oriented as Red Alert probably is.

Just found this benchmark for DirectDraw - hopely it runs on W95.
http://www.roylongbottom.org.uk/directdraw%20results.htm

kixs wrote:

When I find the time... I'll do some more benchmarking with different VLB cards. But I don't have any late VLB chipsets, only CL […]
Show full quote

When I find the time... I'll do some more benchmarking with different VLB cards. But I don't have any late VLB chipsets, only CL-5428, Tseng ET4000/w32i, S3-805 and some ALi.

Do you recommend any benchmark (app or game) for Win95? I usually bench with Wintach and Winbench 3.1. But these are not DirectDraw/DirectX oriented as Red Alert probably is.

Just found this benchmark for DirectDraw - hopely it runs on W95.
http://www.roylongbottom.org.uk/directdraw%20results.htm

Thanks. I'll have a look at that benchmark program.
I'm going to do some benchmarks myself as soon as I find the time. I've been really busy for a while now, coming up to Christmas, baking and such. I wanna try to track down as many of the benchmarks from the CGW roundup as possible, so I can compare my own PCI cards against the ones they use. I might even consider popping my DX2/66 into my motherboard to get more comparable results. But with my motherboard lacking jumper manuals and stuff, I don't know if I can be bothered 😜
I expect that a VLB system will end up being slower than a PCI system in the end, but if it's not by much, then perhaps the higher retro factor might win out over pure performance.

It isn't that much of a difference between VLB and PCI buses but PCI has much newer graphic chipsets. Which excel in Windows.

PCI also has more reliable DMA support, meaning the potential for reduced CPU utilization. Not so much with 486s though in my experience. PCI doesn't really become solid and performant until 1996-1997 chipsets.

VLB can be troublesome too and can usually only handle 2 cards at once reliably. I don't remember seeing DMA on VLB at all with the various IDE cards I used. One Promise card did DMA to the drive but not to the system. Maybe NICs and SCSI adapters used DMA. VLB had problems with specification and probably board quality.

SCSI VLB cards do true DMA, it wrecks havoc with L1 cache in writeback mode though. Did any of those caching IDE cards do DMA?

For fast, reliable DMA and bus mastering in a 486 class machine, EISA might be the best bet.

Has anyone ever tried Number Nine VL Bus cards? How are they at gaming and Winodws acceleration?

Don't the older N9 cards have 512 KB Cirrus Logic chips for VGA/SVGA compatibility? If so, pretty poor for VLB but okay for games that run on ISA cards.

m1so wrote:

Don't the older N9 cards have 512 KB Cirrus Logic chips for VGA/SVGA compatibility? If so, pretty poor for VLB but okay for games that run on ISA cards.

I know some of them had an S3 chip, but not sure which number. Did Number Nine have any of their own video chips then or were they only sourcing chips from other companies?

Right, I've decided to treat myself to a Christmas present, so I've paid a little bit more than I'd intended, but the card I'm getting is a S3 Vision864, which seems to be one of the best chips you could get for the VLB. Hopefully it'll work out well when everything's set up.

sliderider wrote:

I know some of them had an S3 chip, but not sure which number. Did Number Nine have any of their own video chips then or were they only sourcing chips from other companies?

You misunderstood my post. I mentioned VGA compatibility. Number Nine had great accelerator chips of their own, but they were not VGA compatible, only usable with Windows as mentioned here http://vintage3d.org/n9.php#sthash.CNnDWrTX.dpbs . Thus, until their chips started to get VGA compatibility, their cards usually had 2 chips, the native super-fast accelerator one, and the mediocre Cirrus Logic one for VGA compatibility.

Their Imagine 128 and Imagine II cards had separate VGA for awhile but they eventually rectified that. The various Weitek cards were like this too initially.

Number Nine most definitely had some high end S3 cards too. The Motion 771 with S3 968 and VRAM was a ~ $600 card initially, for example. Later they used S3 to build mainstream products while trying to sell their Imagine and Revolution hardware with sky high pricing.

LunarG wrote:

Right, I've decided to treat myself to a Christmas present, so I've paid a little bit more than I'd intended, but the card I'm getting is a S3 Vision864, which seems to be one of the best chips you could get for the VLB. Hopefully it'll work out well when everything's set up.

Yeah that should prove to be a good choice.

I've never had a #9 DRAM based card, but their S3 964 and 968 cards were both good.

kixs wrote:

When I find the time... I'll do some more benchmarking with different VLB cards. But I don't have any late VLB chipsets, only CL […]
Show full quote

When I find the time... I'll do some more benchmarking with different VLB cards. But I don't have any late VLB chipsets, only CL-5428, Tseng ET4000/w32i, S3-805 and some ALi.

Do you recommend any benchmark (app or game) for Win95? I usually bench with Wintach and Winbench 3.1. But these are not DirectDraw/DirectX oriented as Red Alert probably is.

Just found this benchmark for DirectDraw - hopely it runs on W95.
http://www.roylongbottom.org.uk/directdraw%20results.htm

I also can help if people want.. I have a whole pile of these cards here mach 32 / S3 Vision / ET4000 / ET4000 /w32i/p and a trio 64..

Also have a lot of VLB boards here and a few trays of 486 processors.

LunarG wrote:

Anyone tried the CL-GD5434 as used on the Diamond Speedstar 64? From the technical specs, this seems like it should be a pretty good graphics chip.

A few days ago, my new (CL-GD5434-based) Kelvin 64 VLB 2MB arrived. I put it in my DX4 (2nd in signature) to replace the Stealth64.

3dbench2:
Index dropped from 67,9 (Stealth64) to 64,6 (Kelvin64).

However, doom performs slightly better (measurable, not visible)
doom -timedemo demo3
1921 (Stealth64)
1906 (Kelvin64)

doom -timedemo demo3 -nosound
1775 (Stealth64)
1766 (Kelvin64)

The vidspeed 4.0 results leave me a little confused as the Stealth's 16 and 32 bit results in SVGA don't differ whereas the Kelvin can nearly double its results between 16 and 32 bit. Does anybody have an explanation for this behavior?

CL5434 2MB VLB (Orchid Kelvin 64 VLB)
=========================================
7393W 2652R Bytes per millisecond 28.09KHz 70.22Hz 320x200x256 (VGA/MCGA)
14773W 5305R 16 bit writes/reads
29467W 10611R 32 bit writes/reads
------------------------------------
7569W 2634R Bytes per millisecond 28.83KHz 60.05Hz 640x480x256 (S-VGA)
15138W 5269R 16 bit writes/reads
30276W 10542R 32 bit writes/reads

S3 864 2MB VLB (Diamond Stealth 64 DRAM)
============================================
10922W 2184R Bytes per millisecond 27.90KHz 69.75Hz 320x200x256 (VGA/MCGA)
21845W 4371R 16 bit writes/reads
30007W 5055R 32 bit writes/reads
------------------------------------
10927W 2105R Bytes per millisecond 34.95KHz 72.80Hz 640x480x256 (S-VGA)
16976W 2595R 16 bit writes/reads
16976W 2729R 32 bit writes/reads

In pure VGA it makes no real difference. And that's my only use case 😉