Ah, I did not know that zooming was forbidden... I did it a little bit
+0 / -0
|
Benchmark done. I avoided zoom and such, left the mouse entirely alone, but i did type /fps and /speed Did you know that 91 displays the wrong sim speed?
+0 / -0
|
Benchmarked on my Radeon 7870 / Intel i7 3730 combo. I bricked the formatting, couldn't get it to properly split the csv. I'm up for further tests if that helps.
+0 / -0
|
I uploaded some benchmarks using headless exes, but my comp has some CPU problems, so I'm not sure if they're useful. I'll try to fix the thing and add better stats later.
+0 / -0
|
|
|
I've managed to semi-fix my hardware, so there's now real stats from both headless and non-headless on the sheet. I'll try compiling custom exes tomorrow to gather some rough timing stats. Stay tuned.
+0 / -0
|
|
|
Click on the giant "Run a Benchmark" text in @Googlefrog's OP.
+1 / -0
|
Any conclusion? 91 stuck forever?
+3 / -0
|
    GoogleFrog please make "run a benchmark" into a newspost. Shoving an official request into people's faces is a great way to make the need for more data apparent. It's a shame the process couldn't be automated into some sort of executable macro/virus for even the laziest to just click a button and it does all the setup for them and collecting data and posting on googledocs for them.
+2 / -0
|
It was useful to have a baseline and to know that the mysterious people who say dev engine works better on windows do not actually exist.
+1 / -0
|
It seems the jar 3 on the donations page is about moving to a newer version of the engine, so donate to it if you really want unstuck to 91. http://zero-k.info/Contributions
+1 / -0
|
As for me, jar 3 is "Improve the pathing"... In any case, a specific jar for engine upgrade would be a good idea and I would be willing to donate for that.
+0 / -0
|
|
|
Skasi does not know what he is talking about. The pathing in the new engine is better. As far as I am aware performance stops us switching engine but I have not sufficiently tested recent changes to know for sure.
+0 / -0
|
I don't see how units walking into map corners without being ordered, ships being accelerated on shores and movement of 100 units killing FPS is "better pathing". You even just said "performance stops us" and pathing is one of the new things that drops performance. Don't get me wrong, I appreciate units no longer humping my mexes and windgens, but "finding a path no matter the price" is not something I'll settle for. Scrap that, units still get stuck on single buildings on a flat map.
+0 / -0
|
quote:
and pathing is one of the new things that drops performance |
That might be very probable, but we have no data that proves it. What's needed is some real profiling that'll tell us what drops performance and what doesn't.
+0 / -0
|
quote:
but we have no data that proves it |
Start a game on red comet. /cheat /give 100 corfav Give them a (formation) move order. In 98+ this leads to a massive frame drop for me compared to 91.0. Another example is pathing recalculation on ground deformation. According to GF and other devs, the frequency for this is higher in 98+ than in 91.0. You can probably find proof for this in the engine code.
+0 / -0
|
That's observation, not proof. I agree what you say is likely, but theoretically it can also be something to do with orders, mobileAI, unit scripts etc. All these things fit your test just the same as pathfinding. Again, I agree pathfinding is the main suspect, however you must profile to prove it.
+4 / -0
|
https://docs.google.com/spreadsheets/d/15dq0EuuhB29ezifaOQIu2PMpR9T7hbpeGGmgqMPGACI/edit#gid=0I added simple timediff prints in both 98 and 91 to check the difference between different parts of the code. The data was averaged from games created by the regular benchmark setup (12cai, 10 games etc.). In essence, 98.0 has similar unsynced to 91.0, but its synced is significantly slower mostly due to the GameFrame event and slightly due to the pathfinder (other modules seem less significant). Next I'll have to check inside the event to see if I can pinpoint the cause (if really only one exists). It's also interesting to see the relation according to frame % 32 (so resource frames etc. are visible):  More of these here: https://dl.dropboxusercontent.com/u/204076271/springprof/simframe.zip
+9 / -0
|
