Reply 60 of 93, by The Serpent Rider
- Rank
- l33t++
Because sound card is additional variable.
I must be some kind of standard: the anonymous gangbanger of the 21st century.
Because sound card is additional variable.
I must be some kind of standard: the anonymous gangbanger of the 21st century.
Garrett W wrote on 2020-09-06, 11:18:I never understood why Quake and other games were benchmarked with the sound turned off. Was it perhaps due to some soundcards causing issues and lowering performance compared to others? That's the only rational explanation I can think of.
Benchmarking should be as close to real world as possible and who'd want to play without sound and the awesome Trent Reznor soundtrack anyway? 😁
Back when I was doing some benchmarking on a socket-7 board, I started out running tests with sound on but it turned into a headache when I realized how sound settings were affecting results. As Serpent Rider mentioned, it adds variables that make it harder to be consistent. So I simplified things by disabling sound.
One of the variables is the sound card itself, and I didn't intend to commit myself to that particular card for the rest of my life any time I benchmark something. It was just some random SB16 clone, not even a real one.
If you ever want to quantify the impact of the sound card or settings, then you could start by reproducing previous benchmark results that were done with sound off, then start messing with sound settings to see the impact.
feipoa wrote on 2020-09-03, 00:51:Using v1.06 is the norm, not v1.08.
Is there a good reason for this? What version do people play the game with?
See attached. 23.3 fps w/Phil vs. 25.9 fps run direct.
So I've compared it on PMMX 200 Mhz with LT430TX board, and the difference was literally 0.1 fps. 49.9 fps on dosbench and 50.0 from pure Quake run (and that's only because I launched timedemo after loading into new game). As far as I can tell, that odd behaviour can't be replicated by normal means and tied to very specific hardware and/or software config. And by the way, my 5x86-180 results were also made via dosbench suite.
I must be some kind of standard: the anonymous gangbanger of the 21st century.
Basically you won't get the exact same result on the same config if you run the benchmark a second time. Or a third time.
Shame on us, doomed from the start
May God have mercy on our dirty little hearts
alvaro84 wrote on 2020-09-06, 18:21:Basically you won't get the exact same result on the same config if you run the benchmark a second time. Or a third time.
My benchmarks have been quite consistent. Straying only 0.1 FPS running them back to back.
Chadti99 wrote on 2020-09-06, 19:42:alvaro84 wrote on 2020-09-06, 18:21:Basically you won't get the exact same result on the same config if you run the benchmark a second time. Or a third time.
My benchmarks have been quite consistent. Straying only 0.1 FPS running them back to back.
Mine are consistent as well. 25.9 fps running the benchmark manually and 23.3 fps with Phil's benchmark pack.
Plan your life wisely, you'll be dead before you know it.
Upgraded to 512k single banked cache, didn’t quite get a whole frame improvement in software Quake but I did in vQuake! Up from 35.9.
Software Quake is up to 25.8, with one tick away from full screen.
Not sure what’s changed but noticed I’m seeing a consistent 37.1 now in vQuake. Lowest resolution, one tick away from full screen, no sound. Version 1.08.
Would really like to test a v1000 Verite, I’ve read it might be a bit more efficient on slower systems. Finding one has proven a challenge though.
Chadti99 wrote on 2021-06-19, 11:55:Would really like to test a v1000 Verite, I’ve read it might be a bit more efficient on slower systems. Finding one has proven a challenge though.
You'd like t V1000-P if you can find it as it has a small performance increase over the first V1000, but its even more rare.
vetz wrote on 2021-06-19, 14:47:Chadti99 wrote on 2021-06-19, 11:55:Would really like to test a v1000 Verite, I’ve read it might be a bit more efficient on slower systems. Finding one has proven a challenge though.
You'd like t V1000-P if you can find it as it has a small performance increase over the first V1000, but its even more rare.
Thanks for the tip, I’ll continue the search!
Just wanted to share screenshots of latest scores in software quake in both 1.06 and 1.08. System specs haven’t changed but plan to test a 1024KB cache config with a UMC socket 3 board soon.
POD@100MHz
Biostar 8433 ver2 w/2014 bios
512KB single bank cache 15ns
32MB EDO 50ns
Tseng Labs ET6000
Fastest dram timings in bios
*unsure why external clocks shows 66x1.5
Same machine but swap in a Diamond Riva 128 and I consistently get 25.9 and 27.8. Speedsys does show the Riva having more memory bandwidth at 43,306KB/s versus the ET6000 at 40,234KB/s, not sure if that’s what makes the difference.
I doubt the scores would prove any better but I wish I could get either my Banshee or MX400 to boot on this mobo.
Same scores as the Riva 128 on a Matrox Mystique 1064sg-h.
Best score I could get on a LuckyStar LS486e C2 revision socket 3 board. Interesting how this board and the Biostar board perform slightly better with WT mode selected for L2 cache.
POD83@100MHz
256KB Cache
16MB EDO 50ns
Riva 128 PCI
Chadti99 wrote on 2021-06-21, 22:43:Best score I could get on a LuckyStar LS486e C2 revision socket 3 board. Interesting how this board and the Biostar board perfor […]
Best score I could get on a LuckyStar LS486e C2 revision socket 3 board. Interesting how this board and the Biostar board perform slightly better with WT mode selected for L2 cache.
POD83@100MHz
256KB Cache
16MB EDO 50ns
Riva 128 PCI
If you change tag bits to '7' then write back will be faster
maxtherabbit wrote on 2021-06-22, 01:25:Chadti99 wrote on 2021-06-21, 22:43:Best score I could get on a LuckyStar LS486e C2 revision socket 3 board. Interesting how this board and the Biostar board perfor […]
Best score I could get on a LuckyStar LS486e C2 revision socket 3 board. Interesting how this board and the Biostar board perform slightly better with WT mode selected for L2 cache.
POD83@100MHz
256KB Cache
16MB EDO 50ns
Riva 128 PCIIf you change tag bits to '7' then write back will be faster
Not in this specific example, WT + 8 beats WB + 7 by half a frame or so.
Alrighty, up and running on a GA486 board and getting my best scores yet! Running with the following:
POD83@100(40fsb)
1024KB L2
16MB 60ns FPM
Riva 128 PCI
Quite happy with 29FPS in 1.08. If anyone sees anything I can optimize on the chipset settings let me know. This board appears to give better results with L2 set to WB.
*Found a fix by using the Bios from the Biostar. See next post.
This is sad though, when testing VQuake with my Diamond S220 on this GA486 board I’m getting the following message. Def takes a hit on performance compared to the Biostar board, there must be someway to enable bus mastering.