- Mar 21, 2004
- 13,576
- 6
- 76
I devised methods for testing for it other then "look and see if you catch it". I was getting micro-stuter on a single GPU (both 8800GTS 512 and 4850) in mass effect on an E8400 OC. I tracked it down to CPU speed, and I found a way to document and test microstutter.
I used fraps to measure exact frame times, then I created another column with the formule to get the exact time to render each frame.
I copied the data only, sorted them, and then searched for exactly the same numbers shown (like, 41.32ms to render a frame)...
I found that on occasion those high ms to render a frame were preceded and succeeded by low ms. 41ms = 23FPS.
An example was
Frame 661: 10ms
frame 662: 41ms
frame 663: 15ms
frame 662 is micro stutter in action. it drops from 100fps to 23fps and then goes up to 66fps.
These btw were on 720x480 resolution, with the 4850 running at 20% utilization and the E8400 @3.6ghz @100% utilization.
This prompted me to upgrade my CPU from an E8400 @3.6ghz OC, to Q6600 @3.0ghz oc with undervolt. (I got higher OCs with over volting, but come on, huge OC with undervolt? that is just awesome!)
Anyways, I am getting off track. The point is that I think with this method it IS possible to test for micro stutter. And I think we should. I wonder if we should only look at how many ms it takes to render each frame (and look for largest ms figure), or if the time it takes to render the frames before and after it matter as well.
I used fraps to measure exact frame times, then I created another column with the formule to get the exact time to render each frame.
I copied the data only, sorted them, and then searched for exactly the same numbers shown (like, 41.32ms to render a frame)...
I found that on occasion those high ms to render a frame were preceded and succeeded by low ms. 41ms = 23FPS.
An example was
Frame 661: 10ms
frame 662: 41ms
frame 663: 15ms
frame 662 is micro stutter in action. it drops from 100fps to 23fps and then goes up to 66fps.
These btw were on 720x480 resolution, with the 4850 running at 20% utilization and the E8400 @3.6ghz @100% utilization.
This prompted me to upgrade my CPU from an E8400 @3.6ghz OC, to Q6600 @3.0ghz oc with undervolt. (I got higher OCs with over volting, but come on, huge OC with undervolt? that is just awesome!)
Anyways, I am getting off track. The point is that I think with this method it IS possible to test for micro stutter. And I think we should. I wonder if we should only look at how many ms it takes to render each frame (and look for largest ms figure), or if the time it takes to render the frames before and after it matter as well.
