N64 emulation is all graphical glitches unless you go the Angrylion / SoftGraphic route...

Also a lot of system requirements from hobbyist projects such as emulators are just rough estimates and the line they'll draw for acknowledging bugs/technical support than a technical requirement. Jabo's Direct3D6 by design isn't Geforce256 demanding...

Some years ago I wrote an atmospheric simulation and for it my first opencl kernel to compute condensation and was surprised that the speedup of gpu over vanilla cpu was 100x.

I'm currently being surprised by how awful some later nVidia drivers (21.81, 21.83, and 45.23) perform on my K6 MVP3 system. It's particularly bad in Unreal and Unreal Tournament for some reason (half the fps, sometimes worse). It's so severe a dropoff that I need to go back and reinstall one of the older drivers to verify that they still perform as before, and that something else didn't go wrong.

Less utterly surprising, but interesting, is that I think the K6-3 and K6-3+ don't actually perform exactly the same. They're so close that it's not worth worrying about, but after switching CPUs recently, in a minority of tests I saw some miniscule but consistent differences between them. I'm talking less than 0.5%. In most tests they do measure identically, but not all.
All attached hardware is the same and it's the same motherboard with same BIOS, so they should be running exactly the same clock frequency down to the decimal points.
"Speedsys" implies the difference might be the L2 cache. It appears that the original K6-3 has a slightly faster L2. Maybe AMD tweaked the timings a bit on the "+" chips (or it's just a side effect of the redesign). Maybe it helped them to scale the clock higher, and the effect is tiny.
So tiny that I'm not going to treat them as separate CPUs, but technically I think a difference exists.
I suppose the same thing would be found with alternate core steppings of Pentium 3s and such, so perhaps it's not that remarkable.

My most surprising result when benchmarking is how much the video card matters even in old 386 and 486 PCs and dos games and how much you can cripple the system if you put a basic slow isa vga card to the system. You can turn a 486 into a 386 in games by doing so.

My most surprising result was to get a Socket 7 board, using an AMD K6-III+ CPU, slowed down using hardware de-Turbo to speeds around 1/4 of an IBM XT 8088 CPU:

Re: The IBM XT challenge: How slow can you go with i386+ (IA-32) CPUs?

This resulted in a Topbench score of 1. The benchmark score doesn't go slower than 1, which means it's the slowest computer in its database.

My own surprised benchmarks come from trying newer games in 320x240 on the PCX2 and seeing them shockingly handle them playably well compared to the better hardware of the period I played them on. UT at 60fps+ etc

Baoran wrote:

My most surprising result when benchmarking is how much the video card matters even in old 386 and 486 PCs and dos games and how much you can cripple the system if you put a basic slow isa vga card to the system. You can turn a 486 into a 386 in games by doing so.

Actully, I’ve recently been messing around with HyperWare Speed Kit, which is best know for the fantastic disk cache it provides. It also comes with a device driver to speed up video access and it’s pretty fantastic too! Basically, it speeds up drawing operations in applications that call the video BIOS( so it won’t help games that directly access VGA registers) by trapping and replacing said video BIOS with its own much more efficient code.

In benchmarks on my system, HyperScreen sped up video BIOS drawing by 10x, which is pretty impressive! It makes TUI apps feel a lot snappier.

Most surprising benchmark...

I would say 900MB/s read and 720MB/s writes for any hard/solidstate drive on PGA 478 won as my most suprising one 😎
Here's GPU benchmark for this setup :

I switched NVMe drive with GPU (since board only has one x16 slot).

However, Native Hex Booting from DOS on CF, up to Windows Server 2016 on NVMe drive within single system wasn't far behind (it lost, due to not exactly something that can be called "benchmarking" and being more ridiculous or "Mad" to Me) : LINK 😉

I downloaded TrackMania Nations, the free to play (before f2p became a warning sign!) from 2006 and it runs flawlessly on my Nvidida MX440 at 1080p! In TMN's bechmark it scores 39fps but is slightly higher in game. Didn't think I'd be doing 1080p gaming on a MX card 😜