Videocard Benchmarks + Frametimes

[LOLPhilips]

Hi everyone..

Benchmarks with Frametimes have been out now for a few months, but I have not seen them talked about on the forums alot though..

Whats a Frametime ?? Well we all know Fps (the frames per second) But a Frametime is that one second benchmarked and looked at to see how consistent those frames are thrown out in that second..

Example: your card scores 60fps on battlefield this means 1sec=1000ms 1000/60=16.66667 which means a frame every 16.7 ms..

These kind of benchmarks are really interesting cause it can show you cards that run faster on fps but score worse on frametimes..

A good example was a lot of complaints of radeon users just after the launch of farcry 3.. where they'd score higher fps with then there nvidia counterparts but had a lot of stuttering (this was caused because the frametime scores were far worse..

I'll provide a link to a article about the 7870 against the 660.. where looking at Fps the 7870 looks to be a clear winner but scores worse on some games at the frametimes

http://uk.hardware.info/reviews/4041/amd-radeon-hd-7870-ghz-edition-vs-nvidia-geforce-gtx-660-frametimes-review

This is just an example and I'm not saying amd always score lower or whatever (cause they dont) But it is still quite interresting and a good read..

These tests can be found with most current GPU's and might sway your choice if you're looking to buy right now, so I encourage everyone to take a look and maybe do some research yourselfs..

srry for my english.. And I put this here and not at gpu's since this is a fairly new method and feels like its more like technews to me..

[LOLPhilips]

For those interrested in a better explanation and more frametime benches with differents cards

http://techreport.com/review/21516/inside-the-second-a-new-look-at-game-benchmarking <--- Good explanation

http://us.hardware.info/reviews/4135/amd-radeon-hd-7790-review-incl-frametimes <--- Benches of the new 7790

[eshbop]

In a lot of ways these numbers are more important than FPS. PCperspective has done some pretty intersting analysis in this regard. They developed a way to measure the number of frames displayed on the screen at once (which is related to the tearing you see when v sync is off ie more than 1 frame being displayed at once). A 60hertz monitor would mean 60 measurements per second. The problem with an FPS measurement is that each frame is measured as an equal, where in reality a frame may only be a sliver of the screen therefore providing no perceivable benefit to what is being displayed (but it is still counted as 1 extra FPS). PCPer calculated the frames being displayed as small slivers (under a certain threshold then) then recalculated FPS results. They found that in Crossfire there were many of these insignificant frames, and when they took this into account the FPS difference (leaving out these slivers of a frame as they add nothing to the experience) between a single 7970 and 7970 Crossfire was something like 10%. SLI was better at it, but still nowhere near as good as FPS measurements suggest.

The short story is that, yeah SLI/Crossfire setups may pump out more FPS than the single GPU solution, but if a large number of these frames are only taking up a tiny portion of the monitor there is no perceivable frame rate improvements and the extra FPS is essentially useless. Which pretty much goes without saying if the FPS produced is above the monitors refresh rate.

Obviously with v-sync on this isn't a problem as frames are locked to the refresh rate of the monitor hence only 1 frame being displayed per refresh rate cycle. But there are still some not so optimal frame time results. Point is, more than just FPS should be used as a means to determine how GPUs compare between models in terms of 3D rendering, even more so when SLI/Crossfire is involved.