[citation][nom]ibjeepr[/nom]Again, another good point. Don't supposed I could get you to give me a quick vocab lesson on what the break down is on the adopted names for the stutter types?[/citation]
Sure, I'll give some examples.
Each number written in each line below is a frame and each dash (-) is a moment of time between frames being delivered.
With a single GPU, it's likely to look something like this:
1---2---3--4--5----6---7---8----9--10
Overall, the frames are delivered fairly consistently. There are a few minor variances, but overall, it's very smooth. It can get bad to the point where it's more like this in bad situations:
1---2----3------4------5------6------7
and so on, but it's not a repeating pattern. This may happen when you run into a particularly intensive scene and your graphics card is having a little trouble coping with the increased work load (such as a big explosion) or if some other issue causes the GPU to be unable to cope with the workload IE abnormal throttling or a CPU bottle-neck.
As you probably know, most modern multi-GPU setups have the GPUs in the array each do their own frames individually and take turns delivering frames for the whole display configuration. Micro-stuttering is caused when they get out of sync. Let's say that GPU A gets odd numbered frames and GPU B gets even numbered frames. Ideally, you'd want the frame delivery *map* to look like this:
A1---B2---A3---B4---A5---B6---A7
Well, you'd want something close to that like what you had with the first line for a single GPU at the top of this post. Unfortunately, two GPUs often get stuck in a pattern more like this:
A1-B2-----A3-B4-----A5-B6-----A7
Comparing this to the line above it, you can see that FPS is the same, but the experience would be far from the same because the bottom line here has the GPU B delivering a frame imperceptibly close to GPU A's last frame, forcing GPU A to wait a relatively long time to the next frame that can be very perceptible. This issue is more prominent on slower GPUs where the GPUs aren't fast enough to recover from such a pattern after getting stuck (among other issues) and is also much more prominent on dual-GPU solutions than on triple and especially quad-GPU solutions. This is because it's much less likely for three or four GPUs to get stuck in the same pattern than it is for two GPUs and even if four GPUs do get stuck, they have a lot of time to re-time themselves. They have more time because of the poor performance scaling with such setups. With four GPUs only having the performance of say three GPUs, they have an ~33% higher amount of time for each GPU to render a frame. When you're rushing, you're more prone to mistakes applies to GPUs as well as if does to humans
😉
Variable sub-second FPS (IDK if that's exactly a normal term, but it's literal meaning applies just fine anyway) is kinda like an intermediate between the repetitiveness of micro-stutter and the near randomness of what the second line I made shows. With in each second, but not with every single frame, the frame rate can vary along semi-repetitive but longer cycle patterns. Instead of being every frame, you'll get more of an oscillation every few to every few dozen frames. For example, for the first third of a second, you might have a about 15 frames (45FPS), the next third may be about 10 frames (30FPS), and back again. Like micro-stuttering, this type of stutter is also often related to dual-GPU solutions and can be alleviated by higher0end GPUs or having more than tow GPUs, but also like the simple inconsistent frame rate exhibited in the first two lines at the top of my post, this can affect single GPU solutions too.
Here are some graphs to demonstrate:
first example from lines at top of my comment:
The Radeon 7950 here is clearly shown to have had significant consistency issues in this title in the past, but it is not having sub-second issues nor micro-stuttering. According to the article, this was due to having an issue with tessellation performance. This graph is over the course of several dozen minutes or something like that IIRC.
Variable sub-second FPS:
Here we can see that the GTX 560 had a few big hiccups in sub-second performance even as a single GPU setup and two Radeon 6870s had major problems with this type of stutter in Crossfire. This graph's points are smaller than a second but larger than individual frames.
Micro-stuttering:
Here is an excellent example of real micro-stutter. The average FPS is somewhere around 45FPS, but it'd actually look worse than stutter-free 30FPS. This graph's points are individual frames and how much time passed between their delivery as measured in milliseconds.
They're up to something sneaky. I bet it's driving Jen-Hsun Huang crazy.
Facebook
Twitter
Instagram