fastest vanilla doom shareware timedemo you've ever seen?

[noshutdown]

i have always been curious about this. despite being an old game from the 386 era, its quite limited by dos vga mode performance, so its not going to be infinitely fast no matter how modern your computer is.
recommended test setting is the default cfg that came with downloaded doom shareware, that is: high detail, Screen Size two steps below maximum. and run:
doom -nosound -timedemo demo3

the fastest result that i've ever got is around 330 realtics(can't remember the exact figure), thats ~225fps, on following platform:
pentium e5200 oc 3.33g
via pt880 board
nvidia geforce2ultra
2*1gb ddr2-533 ram
windows98 dos mode
i got ~360 realtics or 207fps with the same video card on 815 board and pentium3s-1.4g, so the improvement is minimal despite the cpu being 4-5 times faster.

the gf2ultra is the fastest video card that i knew of in vanilla doom, its a bit faster(10~20 tics) faster than all other nvidia cards prior to geforce4. performance of geforce4 and newer dropped by about half.

and the slowest result that i have:
cyrix486dx-25
ali m1489 board
ati mach64 isa
64mb edo ram
i forgot the number of realtics but the converted score is around 8fps.

[Maes]

That's an interesting observation. I haven't tried this test myself, but I think the real bottleneck here is the capacity of the video card to act as a pure frame buffer when used in pure VGA mode, and that saturates well before the CPU hits its limits.

Let's see... 225 fps * 64000 pixels means about 14.4 MB/sec, which should be well below the PCI's theoretical limit of 133 MB/sec, but that doesn't mean that Doom's way of writing to the PCI bus is efficient: it was designed with ISA video cards in mind, which had different transfer modes and latencies compared to PCI.

PCI and moreso, PCI Express devices, preferentially work with "packets" instead of single I/O writes, so small repeated write operations might not be as efficient as the total bandwidth might suggest. Also, those modern VGA cards might not be as efficient in pure VGA mode as a dedicated VGA card from the DOS era.

In other words, Doom is "Zerg Rushing" the PCI or PCI Express bus, plus the newer VGA cards are not as efficient in handling this sort of old, legacy display. It could go faster if somehow you had a vanilla Doom using a DOS driver for PCI videocards or modern GPUs, or if the mobo/VGA chipset has a particularly optimized legacy bridging mode.

In fact, I think you will find that all old DOS software, especially benchmarking programs, demos etc. will exhibit the same bottleneck, unless they were designed to take advantage of a PCI video card. Under these circumstances, it's entirely possible that you'll get a faster timedemo on the same machine under DOSBOX or a VM.

However, if you use a source port with a "fastdemo" option (skipping frames is allowed), the faster CPU WILL make a difference, as it should.

[VGA]

I would like to see a benchmark thread about prboom+, where people run a timedemo on identical configs on their PCs. Maybe you can take this task up Maes?

Maybe on MAP29?

[Maes]

Sorry for bumping this, but I though of looking at the famous Doom Benchmark Results page, which has some posted results over the years.

Now, the list has grown quite large and is not very readable, plus not all measurements are pure vanilla running under DOS (some are obtained with source ports and/or other OSes), but I think that a bottlenecking/saturation trend is evident, and CPUs which should be an couple of orders of magnitude faster than the ones at the bottom, do not automatically result in linearly faster speed benchmarks, except if a source port or a different OS is used.

It would be interesting to filter out the list so that only vanilla DOS running under Doom is left, and try to put it in correlation with (approximate) CPU power (even if just expressed in MIPS). The performance asymptote seems to be around 200-something FPS, though none of the machines on that list used for pure DOS was as powerful as the OP's.

However, if the legacy ISA bus is the real bottleneck, then the maximum theoretical fps for vanilla Doom under DOS (provided the motherboard has a timing-accurate legacy ISA bridge) should be about 260 fps, assuming that streaming 16 MB/sec of data through the ISA bus with no latencies or wait states is possible. So, paradoxically, Doom might run faster under a DOS box or a VM, on a given system, rather than under actual DOS, because more advanced OSes and VMs will not be bound by the legacy ISA bus.

[jerk-o]

Running Doom in dosbox? Why not freedos?

[Maes]

Well, you (we) are all free to perform our own benchmarks and post them there. In fact, there was a link on that page for a sort of Doom Shareware "LiveCD" which had just that: FreeDOS + Shareware Doom, already set for boot + automatical testing, though the link is now broken (putting it on a Russian filesharing site was not a good idea, after all).

[fraggle]

jerk-o said:

Running Doom in dosbox? Why not freedos?

Because they do different things. One is an emulator; the other is an operating system.