Memhorder :
slimething :
All you nVidia guys need to do is post a video running BF4 on 64 player Shanghai server 1920 x 1080p, all Ultra settings and 4x msaa etc. The video will show the FPS, graph and video info at the top, and also verify the video settings. Cap max FPS at 75.
Then in the console type 'perfoverlay.framefilelogenable 1' to capture the data while playing. Before stopping the recording, turn it off or simply quit the game.
Download http://www.bytemedev.com/bf4-fta/ , import the data and post the results.
Like this:
And here's the video of my 280x running Mantle, capped at 75 FPS.
https://www.youtube.com/watch?v=NxRMNmyTxKo&feature=youtu.be
Forget the outdated benchmarks, it's real world user results that matter. The 280x stomps the 770 and I'll bet the 780 can't beat it.
Ok Results. 780 stock clocks. CPU is stock as well. Very interesting.
[/URL][/img]
In order to do a proper test, it should be done on a "full" server (64 player) and the complete round if possible to attain a better average, or say a minimum 15 minutes. Those 5 minute tests in "benchmarks" are nearly completely meaningless because 5 minutes is an insufficient amount of time for this type of test IMO (and experience in doing stats for a living). If you read some of them the testers even admit as much.
For Shanghai I use =[PT]=Total Mayhem | 24/7 Siege | Fast Spawn | High Tickets as during the evening EST it is usually full
For Metro/Op Locker I use BanZore.com 24/7 Hardcore Metro / Op Locker #1200 tickets
The procedure I use is:
1) Join server, wait until the map is fully loaded
2) In console, type perfoverlay.frametimelogenable 1
3) Play for a minimum 15 minutes, try to mix it up on map locations with as much action as possible. NO SNIPING
4) Before the round ends, type perfoverlay.frametimelogenable 0
Use BF4 Frametime Analyzer. If you see some crazy FPS highs and lows, it can skew the results. Use the smoothing option to see if it changes the averages much. If it does, the test is void right out of the gate. I make it a habit of deleting 5 (5,000 ms) seconds from the beginning and end anyway.
Cautions:
1) Do not start recording frame times until you are actually spawned on the map.
2) Do not allow the game to end without disabling frametimelog first. If the game ends or is quit before disabling, edit the data to eliminate the last 30 seconds (30,000ms) or so. Otherwise you end up with wild low/high FPS numbers.
3) Recording the game obviously skews the results. ALWAYS start recording first, then begin capturing frames and ALWAYS stop capturing before stopping recording.
I'm sure there are more ways to refine the results to test for validity, but I didn't write the app so can't say with a high degree of certainty what they'd recommend, However, I just don't think the myriads of benchmarks we see coming out of _____________ (fill in the blank) can be relied upon as much as people do.