• Please review our updated Terms and Rules here

Fastest PCI video for text only

O

Ole Juul

Veteran Member
Joined
Aug 15, 2008
Messages
3,982
Location
Coalmont, BC, Canada
I just realized that my video card could be faster. Searching the net, it seems like I'm the only one who associates the word "text" with video cards. That's why I'm asking here.

The first card was an S3 Trio64 V+ with 2KB EDO RAM. When I ran a speed test it indicated that it was much slower than the examples. So I rummaged through a pile of old cards and found a Diamond Stealth card with the same memory but an S3 Vision 868 chipset. The result was about 10X faster - 17120 KB/sec text write.

It's still not really fast as it takes a few seconds to render long directories. For example, "du -ab >> file" gives me 125 screens on my utilities directory and doing "cat file" takes a hair over 5 seconds to render. Is this being limited by the bus or CPU or, could another video card make this faster?

PS: Machine = P133 w/ DOS 6.22
PPS: For interest sake I also tried my mono card and the same file took 34 seconds to scroll through.
 
I haven't tried your test yet, but I've got a Diamond Viper Pro. A lot will depend on the OS drivers--some cards have positively wretched performance in real mode DOS, as compared to 32-bit PM.
 
There is an old utility called kspeed4.exe that does a simple benchmark for text speed in DOS you might want to find.

Any PCI video card should be speedy for text, my favorites being Tseng ET6000 and the Matrox line (Millenium I/II Mystique).

I assume you meant 2MB EDO RAM. Cards with VRAM might be a little slower then DRAM/EDO based ones in DOS. Basically you are just worried about video RAM bandwidth and nothing more.
 
Unknown_K said:
There is an old utility called kspeed4.exe that does a simple benchmark for text speed in DOS you might want to find.

Any PCI video card should be speedy for text, my favorites being Tseng ET6000 and the Matrox line (Millenium I/II Mystique).

I assume you meant 2MB EDO RAM. Cards with VRAM might be a little slower then DRAM/EDO based ones in DOS. Basically you are just worried about video RAM bandwidth and nothing more.
Click to expand...

I couldn't find kspeed4 offhand, but of all the ones I have, DIAG.EXE seems to be the best. However, in the end it's the stopwatch test which counts. The above 17120 KB/sec text write figure from DIAG is actually pretty close to the stopwatch figure.

Yes, I meant "2MB".

I too had thought that any card was about as good as it gets, but when I replaced one PCI card with another and got a 10X increase, I realized that there was something more to it. I wouldn't mind another 10X increase (or more). I really like the idea of everything being instant (under 1/30th second). After all, that's the reason I tend to use DOS for text. The DIAG program compares to two other theoretical cards and that would suggest that an ATI Rage 128 Pro would be faster, but perhaps I've already reached my video bandwidth limit with the equipment I'm using.
 
Just for the sake of it, you may want to try running your tests from a RAM disk, so you can be sure disk access isn't affecting the results any.
 
commodorejohn said:
Just for the sake of it, you may want to try running your tests from a RAM disk, so you can be sure disk access isn't affecting the results any.
Click to expand...

Excellent idea! Actually I was, but now I tried it off the CF drive as well and, with 3 tries each, they show to be exactly the same. The file is 133,128 bytes and takes 5.51 seconds to scroll. That's 24 Kb/sec, which is better than (but close to) what DIAG reports, which is 17Kb/sec.
 
Hmm. So does the card provide its own version of the standard BIOS video interrupts, or is one card's hardware just slower? I've never been too clear on how ISA ROMs work.
 
Chuck(G) said:
I haven't tried your test yet, but I've got a Diamond Viper Pro. A lot will depend on the OS drivers--some cards have positively wretched performance in real mode DOS, as compared to 32-bit PM.
Click to expand...

I've never used a driver for video, but I suppose I could if one was available and it made text scroll real fast. I doubt anybody writes software for MS-DOS though (apart from MBrutman of course). Anyway, I would buy a reasonably cheap card if it would speed things up. Your suggestion that some cards have poor real mode DOS performance scares me a bit. In fact that is why I'm asking here for advice/experience.

Does anybody have any idea of the performance of the Diablotek ATI Rage 128 Pro 32MB? PCI is basically old technology at this point, but (apart from a couple of expensive new offerings) there seems to be a lot of old stock out there. Should I buy this one?
http://www.ebay.com/itm/Diablotek-ATI-Rage-128-Pro-32MB-PCI-VGA-Video-Card-NEVER-USED-/170710219276
 
Unknown_K said:
For DOS you have VESA BIOS extensions, univbe.
They are supposed to accelerate all video modes.. could be text modes as well

http://www.dosdriver.de/graph.php
Click to expand...

Thanks for the link. I just installed UNIVBE and tried my previous test. Scrolling is the same. Also, I tried a few colourful screen "savers" and didn't notice any difference either way. Anyway, I'll keep that on hand in case it comes in useful another time.
 
i have a hard time thinking any PCI video card would be slow for text. the only reason i can think of is that for maybe some reason it's BIOS text routines wait for vsync when writing chars. there are only 4000 bytes on a 80x25 text mode screen. there hasn't been a good reason to wait for vsync since single-ported RAM CGA cards though.

download a copy of NANSI.SYS somewhere and install a DEVICE= line for it in config.sys. see if that makes a difference. it takes over existing BIOS character write routines.
 
Mike Chambers said:
i have a hard time thinking any PCI video card would be slow for text. the only reason i can think of is that for maybe some reason it's BIOS text routines wait for vsync when writing chars. there are only 4000 bytes on a 80x25 text mode screen. there hasn't been a good reason to wait for vsync since single-ported RAM CGA cards though.

download a copy of NANSI.SYS somewhere and install a DEVICE= line for it in config.sys. see if that makes a difference. it takes over existing BIOS character write routines.
Click to expand...


hehe, I've been using NANSI.SYS since it was first written. :)

To put it bluntly, I'm an analogue guy in a digital world and all these little delays are depressing me. :) In this case I'm holding DOS to a different standard than a GUI, and striving to get it as fast as possible. By going from one PCI card to another I went from something like 2Kb/sec to 17Kb/sec text write speed and it's very noticeable. It's not "slow", but faster would be better. Perhaps a P133 can't go any faster, but I'm not seeing any other significant delays in normal text processing except this one.
 
Mike Chambers said:
i have a hard time thinking any PCI video card would be slow for text. the only reason i can think of is that for maybe some reason it's BIOS text routines wait for vsync when writing chars. there are only 4000 bytes on a 80x25 text mode screen. there hasn't been a good reason to wait for vsync since single-ported RAM CGA cards though.
Click to expand...

First, later PCI video cards were optimized in design for 32-bit (protected mode) access, with 16-bit access being supplied as sort of an afterthought (can't boot with a blank screen after al).

If you've got the memory, it might be interesting to see if "unreal" mode works any better. Try displaying using an OS such as DOS32, for example.

Second, BIOS character mode routines, such as you'd get by simply using DOS routines resort to moving characters (rather than moving the display aperture when scrolling. This is hideously slow--so slow, on 8088 machines, that it couldn't keep up with 9600 bps serial-port terminal programs. Many early terminal programs of the time, to get the display speed, move the starting address for the display (in the 6845 controller, rather than move the character data).

Perhaps (I hope) that's what NANSI.SYS does.
 
Even if you don't move the display aperture, you'd think that a PCI card would have hardware copy you could use to vastly accelerate scrolling...if the firmware implementors cared about text mode :/
 
commodorejohn said:
Even if you don't move the display aperture, you'd think that a PCI card would have hardware copy you could use to vastly accelerate scrolling...if the firmware implementors cared about text mode :/
Click to expand...

Since we're talking about a P1 system here, one could also improve scrolling using MMX instructions if they're available.
 
Unknown_K said:
In the BIOS is IRQ for video turned on or off?
Click to expand...

There is no clear indication. What it has is:

Code: 
PCI/VGA Palette Snoop (Disabled)
Video ROM BIOS Shadow (Enabled)
Video BIOS Cacheable (Enabled)

Slot 1 IRQ (Auto)

I tried changing that last one (which has the video card) to "9" but it didn't make any difference. I assume it's just for PnP.

My guess is that this card is running as fast as it can, but of course would be interested in trying anything that will make significant improvements. I wonder if any of the other video related settings could make a difference. I'll try them later.
 
I just bought a Var128P-32P ATI Rage 128Pro 32 MB PCI Video Card from eBay. Hopefully I'll get it, though the eBay complications are a bit beyond me so I have my doubts. Anyway, I'll report on that when I get it. At least it will be educational.
 
Ole Juul said:
I just bought a Var128P-32P ATI Rage 128Pro 32 MB PCI Video Card from eBay. Hopefully I'll get it, though the eBay complications are a bit beyond me so I have my doubts. Anyway, I'll report on that when I get it. At least it will be educational.
Click to expand...

The card just arrived. W00t!

To recount:
I started with an old S3 Trio64 V+ with 2KB EDO RAM card. Then I tried a Diamond Stealth v1.02 dated 1995 and got almost a 10x increase in speed of text rendering. That made me so excited that I thought I'd see if it would go even faster, so I ordered the ATI Rage 128 since it is supposedly much faster and cheap because it is obsolete now.

The results:
It was slightly slower! hahah Using my test of "cat testfile" in a ramdisk I got 5.52 seconds from the previous card and 5.99 seconds for the new one. A diagnostic test (DIAG.EXE) showed both text and graphics (not important to me) rendering to both be slightly lower. The faster numbers (for the Diamond) being 17,120 KB/sec for text and 22,211 KB/sec for graphics.

Conclusion:
I was going to go back to the faster card even though it was only a small difference since the new one could be used in a newer computer to better effect - although it's a sexy red which is a huge plus to me. :) However, when I went back it became obvious that the newer card actually produced noticeably clearer text. OK, that was it. That's definitely the deal maker. The speed difference is not noticeable, but the clarity definitely is - and I get to keep the nice red card.

PS: This is all barely vintage, although it is a P1. Still, using a DOS only box in this day and age is somewhat vintage. At any rate the user is.
 
You must log in or register to reply here.
Back
Top