I'll post this in Delfino for now and see if this goes anywhere. I've ran my processor through
this and my A6-3400M quad-core overclocked to 2.4GHz gets a total of 37GFLOPS. In my estimation running Zelda Windwaker full speed (give or take 2-3fps) needs at least a 37GFLOP processor. If we can come up with some minimum Flops for games to run well, we can give other people a better idea of if their processor can run a game well or not.
Must I point out the obvious?
GFLOPs != game performance
NV: GFLOPS doesn't mean anything on its own, but when a processor is known to run a game at full speed, you can then use that GFLOP number to estimate what is needed to run the game.
That makes no sense.
What you're basically saying is that a cpu that reaches the same GFLOPs in this benchmark will achieve the same performance in that game. Which is of course 100% false.
GFLOPs is a synthetic measurement of floating point heavy code. It does not reflect the performance of an application using heavily mixed code (a.k.a. the vast majority of all applications including dolphin). For example bulldozer is faster than sandy bridge in synthetic GFLOPs, MIPS, and IOPS tests.
(01-28-2013, 08:23 AM)NaturalViolence Wrote: [ -> ]What you're basically saying is that a cpu that reaches the same GFLOPs in this benchmark will achieve the same performance in that game.
Granted this is not the best way to measure, but yes, generally. Run that test with your CPU and tell me the number you get and your CPU clock speed.
When I get home I will but just keep in mind that these results carry no weight. No matter how hard people may try their is no magic number that can be used as a comparison to determine a games performance without actually running the game.
Floating point benchmarks have an extremely loose relationship with real world performance of mixed code, or even applications that actually use a lot of floating point vector arithmetic. An architecture (like bulldozer for example) that handles this type of code well will require a higher "GFLOPs rating" to reach the same game performance as another cpu. Thus the comparison falls apart.
I was also going to use the WW benchmark to verify this a bit. On the first screenshot I got 106% on my 2.4GHz 37GFLOPS processor. I'd expect you to get around 85 GFLOPS at 3.7GHz based on the 241% speed of your first shot. Again, not an exact science, but generally.
I get 95GFLOPS average with a i5-2500K at 4.5GHz and 106GFLOPS average at 5GHz.
With my laptop, which has an i5-2410M at 2.9GHz, I get 60GFLOPS.
(01-28-2013, 08:55 AM)NaturalViolence Wrote: [ -> ]When I get home I will but just keep in mind that these results carry no weight. No matter how hard people may try their is no magic number that can be used as a comparison to determine a games performance without actually running the game.
Floating point benchmarks have an extremely loose relationship with real world performance of mixed code, or even applications that actually use a lot of floating point vector arithmetic. An architecture (like bulldozer for example) that handles this type of code well will require a higher "GFLOPs rating" to reach the same game performance as another cpu. Thus the comparison falls apart.
So wouldn't this be more accurate if used only to compare processors of the same architecture?
If they're using the same architecture , all have 3+ cores, and have the same amount of cache, then yes. But in that case clock rate would be just as effective for a comparison.