CPU speed benchmark is not even close

[Habeed]

My CPU : i7 930 overclocked to 4.08 ghz. Yet, the silly little benchmark gives me "ratings" in the orange range. No, there's nothing wrong with my computer, in actual games I get sim speeds that are at the top of the pack. The only player that does better than me in most games is running an i7 2600k overclocked to 4.8 ghz.

What's the "benchmark" even measuring? How come I don't hear my CPU fans spin up when it runs?

[Duck_42]

Habeed,

It's a pretty basic benchmark. It's only going to use one core and it only runs for 5 seconds, so I doubt your CPU fan would spin up. Regarding the rating, it WILL give you a considerably worse rating if you tab out or switch to another window while the benchmark is running (nothing I can do about that). To rerun the benchmark, hit the button next to Observe. If it still gives you a bad rating after that, let me know.

[Ionic]

Does the same for me even when I re-run the benchmark, I play in windowed mode with an i7 and usually have the best sim in game.

[Habeed]

Same. It's not even remotely close at all to my in game performance. I mean, I could forgive it if say some players got a "50" and I got a "75" on your scoring scale, even though my machine might be faster than the fellas with the "50" in game. But I'm getting 350 or more, it's TOTALLY wrong. Fix your code please.

Though...I also play in Windowed mode. hmm...

[BRNKoINSANITY]

This is a running theme. Back when this was only on the phantom games one other person and I tried it out in windowed, tabbed, and in regular fullscreen. He has an absolute beast of a pc and I have a decent one. We both got the best score in full, worse in windowed, and horrible rating when tabbed to a different window. It was bad to the point of his rig scoring lower than known laggers when in windowed mode.

Honestly this test needs tweaked quite a bit to really be useful.

[Duck_42]

@Habib,

I play in windowed mode as well. I didn't notice much difference in the scores between windowed and full screen on mine. Without having any idea of what's "wrong" with the code, there's no way I can fix it. All the benchmark knows is that your PC runs it slower. One thing I looked at in the initial testing was validating that the scores match game performance. From my experience, they do. There may be some exceptions though.

@Brink,
I've said it before, tabbing out before the benchmark runs will mess up your score. There's no way to fix that. Don't tab out until it's done.

@Everyone,

PM me when I'm online if you want to set up a quick test game (compare CPU scores, then spawn a bunch of units and watch netstats) to validate (or invalidate) the CPU score. I'm not chasing ghosts in the machine until I actually see them.

[Flynn]

Is this a new feature? If so, where do I find it?

[Ze_PilOt]

Maybe you are using some kind of speed reduction thingy on your CPU, making it slower when it's not on heavy load.

As the test run for 5 seconds, it's not enough to "unleashed" it, and so your score is bad.

[Rienzilla]

What about a different approach?

The goal of the CPU speed benchmark is to prevent games going at a slow simspeed, because some players have a pc that can't handle the size of game they're playing, right? Why don't we measure exactly what we want to know?

I would periodically measure the amount of units in a game, and the actual max simspeed for a player during every game, and store that in the database. You can normalize that that into a performance number or curve (I haven't extensively thought about a good normalization, but examples: 12000 units with +10 simspeed is 100/100, 1 unit with -10 simspeed is 0/100. Or a curve: simspeed at 100, 500, 1000, 2000, 4000 etc, which you can extrapolate and then integrate to get a meaningful number). Then, remember just the last 10 or 100 measurements for every unit count and average that, so that a players that buys a new pc, or disables his seti@home or whatever eats his processing power, improves his cpu rating. Maybe we can even somehow get a unique pc-identifier so a user that plays on different pc's is also tracked accurately.

So, approximately
- During game: measure and store tuples [unitcount, maxsimspeed] in database for unitcount = 1; unitcount < 12000; unitcount+=100) (Zep, is this possible as far as you know?)

To calculate actual cpu preformance index:
- Server side: average the last 10? maxsimspeed values for every unitcount from the previous for loop. Results in a set of tuples S = [unitcount, avgmaxsimspeed]

and then either:
- fit a curve C over set S
- calculate integral I over unitcount = 0 to 12.000 of C

or: publish 3 numbers for small (~200), medium (~1000), and large (~4000) units.
--
Rien

[Pathogenic]

Yeah, I think the CPU speed indicator was a bad idea from the start. This thread only reinforces my opinion of that.