Frame Rating Dissected: Full Details on Capture-based Graphics Performance Testing

[Lonbjerg]

Wow.Lonbjerk that sure made your last zillion posts look stupid lolol:biggrin:
Show me the bit where 7970 doesn't own the 680..what?....you can't?
Ouch.

???

 

[showb1z]

Pcper has shown very clearly what the issue with vsync is, it causes higher variance, it definitely causes input latency and it also causes huge steps in the frame rate. If you play a game with this constant 60/30 jumping its horrible. You need to leave a lot of performance and set your graphics up for the very worst so you never end up swapping back and forth.

Adaptive Vsync.
Though he made a pretty glaring mistake in his settings. He's running adaptive/smooth vsync (half refresh rate) instead of normal. Which is why those tests are only running at 30fps. Strange he didn't catch that.

 

[GaiaHunter]

Even AMD fanbois should like the AMD driver team being held to a higher standard, as it helps AMD fanbois get more out of their cards. It's a win for everybody to use better metrics than the outdated frames-per-second metric.

The fun part is that fraps is shown in a pretty good light.
[H] using fraps numbers and charts plus adding their own impression on what settings are playable have been doing a more solid job for years now.

Also notice how tomshardware found nothing wrong with the 7878 crossfire setup.

 

[willomz]

They do not state anywhere that "maybe the nvidia provided tool is not able to read the competetitor's cards correctly". Which should really appear in any scientific conclusion. I mean is does seem rather silly than nvidia would go out of their way to make their software work for other drivers, especially when this software was really only developed in their labs to work on their gpu's.

This completely bypasses the drivers:

"Rather than attempting to tap into the video drivers themselves – a process inherently fraught with problems if you’re intending to do it in a vendor-neutral manner that works across all video cards – through FCAT NVIDIA can do true frame analysis, capturing individual frames and looking at them to determine when a buffer swap occurred, and in turn using that to measure the frame interval."

This works purely on what is displayed to the screen, i.e, exactly what the user would see.

 

[bononos]

The article said that AMD fixed stuttering in dx9 quickly - I though a general fix for both dx9/10+ is still in the works. When did AMD fix the issue in dx9 in a single stroke?

 

[willomz]

The fun part is that fraps is shown in a pretty good light.
[H] using fraps numbers and charts plus adding their own impression on what settings are playable have been doing a more solid job for years now.

Also notice how tomshardware found nothing wrong with the 7878 crossfire setup.

Mostly yes, FRAPS does tend to be pretty accurate. Not always though:

http://www.tomshardware.co.uk/graphics-card-benchmarking-frame-rate,review-32658-12.html

The minimum in FRAPs is 38fps the minimum in FCAT is 26fps and that's not just a short downwards spike either. The graph shows a 20-30fps discrepancy for large parts of the benchmark.

Tomb raider is probably the most extreme example, but it's not the only one.

The Tom's report is flawed however, there are no FRAPS figures for the 660 Ti and it's hard to say that the 7870 and 660 Ti are 'closely-priced'.

 

[Ibra]

Wow.Lonbjerk that sure made your last zillion posts look stupid lolol:biggrin:
Show me the bit where 7970 doesn't own the 680..what?....you can't?
Ouch.

 

[Goatsecks]

Even AMD fanbois should like the AMD driver team being held to a higher standard, as it helps AMD fanbois get more out of their cards. It's a win for everybody to use better metrics than the outdated frames-per-second metric.

Exactly, this is a good day; we are now equipped with a new metric, and repeatable results. The quality of a users experience is only going to improve (for either amd or nvidia users) as a result.

Bring on the driver updates.

 

[Silverforce11]

Pcper has shown very clearly what the issue with vsync is, it causes higher variance, it definitely causes input latency and it also causes huge steps in the frame rate. If you play a game with this constant 60/30 jumping its horrible. You need to leave a lot of performance and set your graphics up for the very worst so you never end up swapping back and forth.

Crossfire is just broken, has been since the day the 7970 was released and I am so glad this level of test has finally been done.

They are completely bad for not even mentioning that AMD cards get adaptive vsync and frame metering as well, just load up Radeon Pro. Takes a few seconds to set a profile for your game.

There's no 60/30 jumps, only for lazy ppl who dont use 3rd party tools.. i dont know many enthusiasts who dont at least use afterburner or variants. Radeon pro has an in-game OSD as well. Get with the times.

 

[Keysplayr]

Lazy people. That don't use 3rd party tools.
Did you sound this out in your mind first before committing it to a forum post?
At any rate, may I quote you and use this in my sig?
AMD should have a form of adaptive vsync and frame metering right in its drivers. Who is lazy and lets others do the work?

 

[Keysplayr]

 

[ICDP]

Lazy people. That don't use 3rd party tools.
Did you sound this out in your mind first before committing it to a forum post?
At any rate, may I quote you and use this in my sig?
AMD should have a form of adaptive vsync and frame metering right in its drivers. Who is lazy and lets others do the work?

Totally agree with this. I do advocate the use of RadeonPro but this does not excuse AMD for not including at least some essential options that Nvidia give us as default. I have said in the past that AMD needs to introduce far more features in their basic driver pack.

DirectX Triple Buffering
Flip queue adjustment (render ahead)
Equivalent of Nvidias Adaptive Vsync
FPS cap would be nice

 

[AdamK47]

Pcper has shown very clearly what the issue with vsync is, it causes higher variance, it definitely causes input latency and it also causes huge steps in the frame rate. If you play a game with this constant 60/30 jumping its horrible. You need to leave a lot of performance and set your graphics up for the very worst so you never end up swapping back and forth.

Crossfire is just broken, has been since the day the 7970 was released and I am so glad this level of test has finally been done.

This post has shown very clearly that you are unaware of the benifits of vsync when triple buffering is enabled. No 60/30 jumps when used.

http://www.anandtech.com/show/2794

 

[sontin]

Triple-Buffering adds a another frame to the input lag. In combination with AFR you have at least 3 frames ahead than with a single GPU and without Vsync.

 

[ICDP]

Triple-Buffering adds a another frame to the input lag. In combination with AFR you have at least 3 frames ahead than with a single GPU and without Vsync.

Is this not where max render ahead or flip queue size comes in? I notice a definite decrease in input lag when I adjust the flip queue size to 1.

 

[SirPauly]

It's wonderful to see more investigations!

Now with tools like this that go beyond just raw frame-rate translates into resources and focus from the IHV's on smoother gaming for their sku's and platforms.

 

[Cloudfire777]

wow, how horrible the AMD cards stutter. Nvidia cards have a straight line, AMD cards have like this ocean of latency.
The videos was truly great, showing the difference.

 

[willomz]

wow, how horrible the AMD cards stutter. Nvidia cards have a straight line, AMD cards have like this ocean of latency.
The videos was truly great, showing the difference.

Generally speaking the single card AMD solutions are pretty consistent. It is crossfire that has some pretty dire problems in certain games, but in other titles it can match SLI smoothness.

 

[UaVaj]

dissappointed. no titan sli results?