The immense drop in frames when translucent elements are being rendered has also baffled me when I used an A64 + NVidia FX5xxx. Rates dropped to way less than 1/4th. The finished frame should just be copied from system to video buffer... can't imagine that something's being read, modified, written from/to video memory.
Maybe sonebody who has some coding experience relating to said issue may chime in.

If you want fast performance in DOS, avoid anything newer than a Geforce 2/3, as performance dips after this. Also, don't forget to enable MTRR Write Combining using MTRRLFBE.EXE.

I don't know about that, I'd want some confirmation. The graphics card shouldn't actually care about the OS, especially since almost all the time in software rendered 3D is just copying data into video memory. The slowdown with translucency is very odd; of course similar things happen in the Doom engine but they don't seem to be GPU dependent. My experience is that Build and Quake engines have much less varying framerates than Doom engines, so that a lower average is needed to give the same perceived smoothness.

koverhbarc wrote:

The slowdown with translucency is very odd; of course similar things happen in the Doom engine but they don't seem to be GPU dependent.

I don't even try to understand Doom. Its performance above ~ P2 class is a complete hit and miss. "Luckily" it's limited to 320x200/35fps so it's highly theoretical 😁

(Of course I know that it has ports and I often play Zandronum at 1920x1200 but that has a very different renderer.)

Oh, about nolfb+Tualatin: It works well, Blood is very playable at 1280x1024. 1600x1200 is still a bit jerky, Northwood 3.06 is clearly better, even with PCI gfx. And consumes 40% more power.

koverhbarc wrote:

I don't know about that, I'd want some confirmation. The graphics card shouldn't actually care about the OS, especially since almost all the time in software rendered 3D is just copying data into video memory. The slowdown with translucency is very odd; of course similar things happen in the Doom engine but they don't seem to be GPU dependent. My experience is that Build and Quake engines have much less varying framerates than Doom engines, so that a lower average is needed to give the same perceived smoothness.

The only action in Doom where translucency is involved is when drawing the spectre(drawing a "fuzzy column") - same with the player sprite when having the "eye of protection". The variation in frame rates seem to stem from the possibilty that, some of, the routines are directly drawing to the screen(in an X mode) and not just working with a memory buffer which is always linear. I remember reading that Carmack found a way to "interleave" some of the workload when drawing the columns BUT I do not remember the source and could have misinterpreted the info. A graphic's card VGA core, next to bandwidth and processing power of course, may thus explain the fluctuating scores achieved in Doom.

Quake does not "fiddle around" and just transfers a finished frame to the screen.

koverhbarc wrote:

I don't know about that, I'd want some confirmation. The graphics card shouldn't actually care about the OS, especially since almost all the time in software rendered 3D is just copying data into video memory. The slowdown with translucency is very odd; of course similar things happen in the Doom engine but they don't seem to be GPU dependent. My experience is that Build and Quake engines have much less varying framerates than Doom engines, so that a lower average is needed to give the same perceived smoothness.

Sorry, what confirmation are you looking for?

Geforce cards are optimised for 3D APIs, long after the heyday of DOS. Not all VGA cores are made equal, nor do bus widths give an indication of VGA performance. Performance can also vary depending on video modes used, too. Regarding Nvidia VGA cores, I did run some benchmarks with fast P4s (Northwoods and Prescotts) on an i875p board with ISA, and they were significantly faster than P3 Tualatins.

After NV20 chipsets, I noticed VGA performance dropping. I also needed to use the aforementioned MTRR Write Combining to gain performance, otherwise the increase was minimal.

koverhbarc wrote:

I'd been wondering about doing this, but I don't want any possibility of ruining the board; I assume desoldering the buzzer is required first.

Since you've done it, does there need to be a DC path between + and -, or does there not? And I'm a bit surprised you can drive a 4 ohm speaker, I assumed it would be 600. I have many 8 ohm 'hobbyist' speakers, and one at least is quite loud enough - maybe too loud, since PC speaker games had no volume control ability nor does the speaker itself - although I suppose you could add one in the path if you're going to be changing it anyway!

Apparently it's not that difficult... Any crimped connector wires will do. First cut one side of the connectors off and use a wire clipper to expose the copper wires inside for soldering it to the + and - pads of the speaker, and place the connectors of the other side into a 4-pin header and connect it to the motherboard's SPEAKER pins just like the way you would with a buzzer (you can simply reuse the 4-pin header that comes with your buzzer for this, if you're already using an external buzzer that's plugged into the SPEAKER pins). On the other hand, if you have access to hot glues or insulation tapes it's better to put some on top of the soldered pads to avoid potential issues.

Also, no need to desolder the onboard buzzer anyway, as should the SPEAKER pins are connected to an external buzzer/speaker the on-board buzzer will be bypassed.

Besides, almost all motherboards can drive 4-ohm speakers from the SPEAKER pins as far as I can tell, just that the volume varies depending on many conditions, like power supply and the positioning of the speaker. Be noted that you'll often have to find a suitable place to support the speaker as they can be heavy and not suited to be placed too close to the motherboard. For that reason you may need to use relatively longer wires so you can have a good room for positioning it.

I just bought a Pentium-4 (ICH5, unknown northbridge) from an eBay seller. Looks like AGP, 5-6 PCI, 1 ISA. Northbridge is probably either E7205 or 865/875. It has an ATX12V, so not likely to be Willamette.

Apparently he has a dozen of these boards that were retired from an industrial surveillance system. The name wasn't very well listed. But I offered to fully ID the board to him in exchange for a discount.

LSS10999 wrote:

... and place the connectors of the other side into a 4-pin header and connect it to the motherboard's SPEAKER pins just like the way you would with a buzzer ...

Boards like this don't have any speaker pins, unless they're hidden under the buzzer. I expect it's actually solder pads, hence desoldering seems required (and I'm not sure you could get in there with a normal sized iron).

koverhbarc wrote:

LSS10999 wrote:

... and place the connectors of the other side into a 4-pin header and connect it to the motherboard's SPEAKER pins just like the way you would with a buzzer ...

Boards like this don't have any speaker pins, unless they're hidden under the buzzer. I expect it's actually solder pads, hence desoldering seems required (and I'm not sure you could get in there with a normal sized iron).

Just how to describe this correctly... apparently any ATX (be it ATX, mATX, or else) motherboard should have a 4-pin SPEAKER connector as part of the entire Front Panel Connector (for power/reset switches, power/HDD LEDs, etc.), even if they do include a buzzer onboard. Magnetic-driven speakers can be connected the same way as buzzers (just you more often have to solder the wires and headers on your own as speakers with headers presoldered are less available than buzzers).

I dunno what your board really is, but it's rare to come by a motherboard that doesn't leave the 4-pin SPEAKER connector open (either without pins, or simply not there).

koverhbarc wrote:

LSS10999 wrote:

... and place the connectors of the other side into a 4-pin header and connect it to the motherboard's SPEAKER pins just like the way you would with a buzzer ...

Boards like this don't have any speaker pins, unless they're hidden under the buzzer. I expect it's actually solder pads, hence desoldering seems required (and I'm not sure you could get in there with a normal sized iron).

I posted that! I was hoping that someone buying would encourage them to put up the others. I considered making the same offer, but I don't have the time right now to commit to that. I'll keep an eye open for his other auctions. Let him know that you found it from me; I would also appreciate a discount for the exposure! :p