Single-GPU CPU Frame Latency

Go read the article again. Look at their fps charts. They weren't hitting anything close to a reliable 60fps in the first place. If the fps is bouncing around all over from 20-60fps, then of course they aren't going to notice latency. Throughout their entire article there is already a far more serious frame rate issue in play that vastly overshadows the issue of latency.

Latency matters when you are reliably hitting 60fps. Then you've got long frames jumping in to interrupt your smooth motion, which again, is noticeable and disrupting to the experience.

In short, Toms Hardware does exactly what Toms Hardware has been known for for the last decade: A half-ass job with unreliable conclusions.

Ooh! I've got the analogy:

Looking for frame latency inside of a wildly fluctuating 20-60fps render is like looking for a spec of dust on the lens of a noisy, low-light photo. The image is so noisy that the extra 1% of noise from the dust isn't enough to be noticeable. That same spec of dust on the lens of a well-lit photo that is sharp and noise-free? Blindingly obvious and it destroys the image. A dropped frame in a sea of dropped frames? Impossible to find. A dropped frame in an otherwise stable render? Again, blindingly obvious as it disrupts the illusion of smooth motion.

macks, I understand that you are trying to draw a narrative where frame latency doesn't matter. We've also discussed what your system is currently running. In effect, you are looking at that noisy, low-light photo and repeatedly declaring that because you can't find the dust in that image it is therefore impossible for anybody to find dust on the lens in any image ever. It's speaking from a place of ignorance. Being ignorant on a subject is fine, especially if it isn't willful. Pushing that ignorant view as fact though? No, that is not okay.

Tom's Hardware got this one wrong by missing the obvious. Finding incorrect views that match your own is not how we learn.

There is no mention of any CPU in your post. The entire content of your post is a re-post of what some nub over at Toms thought. It sure looks to be mirroring what you've been repeatedly telling us around here about how frame latency doesn't matter.

You'll also notice that that 1100T hangs out right around the Phenom II X4 in all of the graphs, whether it's the almost useless FPS graph, or the poorly constructed Latency graph. Again, it would not be a worthwhile upgrade over your Phenom II X4.

Also too, you'll notice that for their frame latency graphs they cut off at the 95% mark. In effect they are saying "It's perfectly okay to drop 3 frames every single second". Three frames in a row is 1/20th of a second, which is very obvious judder. Three frames spread out is more manageable, but not ideal. Really, you want to drop less than a frame a second. 95% is a bad place to make the cut-off. You'll notice that Tom's competition makes the cut-off at 99%. You'll also notice that the vast majority of difference between CPUs shows up in that last 4%. This is a bad article, through and through.

Indeed dlj is jaded on just about every single post/topic and constantly tells everyone just how wrong they are, don't take it personal, he just spends far too much time reading wiki.

We have to take it easy on him though, imagine how complicated life must be having the IQ of Stephen Hawking and knowing everything about everything.


Jason