FCAT nonsense...& just plain dumb 3d-card reviews

Ah yes, no WaltC topic is complete without an nv30 detour.

None of this gibberish has any bearing on the fact that AMD has a well known issue with stunted frames.

Your original wall of text is barking up the wrong tree, and none of this idle speculation about Nvidia, HC and Skymtl does anything to change that.

 

Why don't you simply separate measured facts from opinion? It makes things much easier that way. It's fine not to like HC for being biased one way of the other (I've never visited their site and you may well be right), but the measurement method of FCAT is hard to dispute.

These kind of runt frames should never ever happen in working system. It's really a case of extreme malfunction in their video source switcher. Remember: in some cases it switches from frame N to frame N+1, back to frame N and then back to N+1!

There's nothing wrong in pointing this out. It's a bug. AMD will fix it.

 

And obviously it works as seen in the comparison video of Crossfire running Crysis 3.

 

So has anyone measured the actual maximum-lag that is coming from implementing smoothing algorithms?
Is it in the range of one or two frames?
A simple 3d program with flying-balls running from left to right with different predefined speeds with time stamps should do the trick I guess.

That way we would see the actual time difference between two displayed frames. And use the distance traveled from the flying-balls to aid with recovering the timeline data. Perhaps we could also gain some insight about completely dropped frames by the frame smoothing algorithm.

Lag is important for gamers - more important in some simulation software I guess.

And anyway I would be glad to find out more about the affects of such a smoothing algorithm, either nVidias implementation or AMDs.

Second, I am not deep into DVI (I am not even sure that all DVI implementations arew the same with the one used in FCAT), but would n’t DP show different results?

 

Nobody has measured it. But earlier discussions here show that you don't need add lag at all times to smooth it out. Just adding a one-time single delay to a steady state situation can fix the imbalance. In practice, nothing is steady state, and you'll need to add a little bit of delay to keep things balanced, but my belief is that it can be done with very minor delay spread over multiple frame. E.g. 1/10th of a frame lag. Definitely not multiple frames. How to measure this is something else entirely...

See here and follow-ups.

I wouldn't worry about that part: GPUs don't have large on-chip memories (ignoring L2 cache), which means that pixels are requested on-demand to be sent to the output. I'm sure there's some amount of buffering to make things easier for the MC and probably a couple of line buffers for output filtering, but that'd be a delay of uS, not mS.

So given that, there should be virtually no difference in delay between DVI and DP. Both send pixels in a digital way. Just the encoding is different.

 

My opinion is multi-GPU set ups are not worth the heartburn.

It was much more fun then. All the 9500pros you could turn into 9700, the same on the other side. That stuff was fun. It seemed the cadence of progress was quicker then as well. I was an AMD fan myself since they were an underdog and Intel was the evil empire back then. Now I wish we could just have some decent competition in all aspects. Intel is dominating AMD in CPUs, and while AMD and Nvidia are close on GPUs I worry that it is basically a losing proposition for them as tablets and smart phones etc.. are adopted. I was bummed about the merger and I am bummed that AMD has the console wins b/c I like at least one to go to each.

As far as the NV30 that was a hilarious time, but not something I got upset about. It was entertaining to me. You take care too it is good to see people back.

 

Thank you for your answer.

I guess someone will eventually set up an experiment to measure all the above.

 

Looks like some devs are taking notice of Microsstuttering


ps: I dont understand the logic of this

 

That probably not really microstuttering, but simple multi-GPU optimizations. In this instances I would guess they have some frame-to-frame storage of the foliage details that would degrade the mGPU rendering performance, so they cut those down to speed the operation up.

What title is that?

 

its the incredible adventures of van helsing
only £10 on steam
http://store.steampowered.com/app/215530/

 

Maybe it's just the language barrier, but are you suggesting in the first part that a reviewer should cherry pick their tests in advance just to let a product shine? If so, where's your rant about all the incompetent GTX Titan reviews where no one really seemed to show how utterly important the addition of an extra 3 GiB of framebuffer for that product was.

Regarding 1 vs. 2 GiB HD 7850's I'm sorry to disappoint you (if no one has done so yet), but 2 GiB of framebuffer memory was the standard configuration with which HD 7850 launched:
http://www.pcgameshardware.de/Grafi...m-Test-Schnell-und-sparsam-dank-28-nm-870528/

 

Do you have link for this? (not that I wouldn't believe there's reviewers that are self-admitted fans of company x, but the link would still make claims relevant)

 

I had to register just to add my two cents to this.

This recent FCAT craze is definitely a net loss to review usefulness, but not because the data is in any way rigged by the nVidia tool. It's absolutely true that nVidia has a habit of playing dirty (such as disabling PhysX when a non-nVidia card is present, or crippling CPU-based PhysX), but it's just paranoia to claim that the FCAT data is tainted.

The problem is that the reviewers are completely wrong-headed in how they're interpreting the data.

The concept of a "runt frame" not counting towards the frame rate is nothing short of idiotic. When you don't sync updates to the screen refresh, *ALL* frames are "runt frames". No screen is ever going to be one distinct rendered frame. Every screen update will be a combination of several different rendered frames, which is why you get tearing when you radically change the view without vsync - the seams between different frames don't mesh.

The simple fact is, the absolutely maximum "practical" or "actual" frame rate is the refresh rate of the monitor. You will always be rendering frames that don't make it to the screen, to some extent or other, if your frame time is lower than the refresh interval, because the frame buffer is overwritten with different frames during the screen draw.

The only difference between the even frame time spacing of nVidia and uneven spacing of AMD is where the seams in screen tearing will appear. So if you have three sequential frames - A, B, and C - generated inside one refresh interval, the difference is in what portion of each frame actually makes it to the screen. With even spacing, it'd be 33% A, 33% B, and 33% C. Mix it up, so it's 30% A, 10% B, and 60% C, and what do you get? Just different seam positions. And what if poor frame B is dropped entirely? Then the screen will be some percentage of A and some percentage of C instead, and you lose absolutely nothing of consequence.

As soon as you turn on vsync, the entire picture changes. Every single frame will be complete, and if the frame rate is at least as high as the refresh rate, time between frames will be identical at all times. And contrary to popular opinion, you don't get "input lag", unless you happen to be playing Dead Space.

 

I do think they should stop testing with vsync off. Who the hell plays with vsync off?

 

All competition players?

If people are willing to suffer TN LCDs for a few ms of latency reduction, there's no point in enabling VSYNC, which will, on average, guarantee a latency increase of 8ms.

 

You're right in a way, yes. But taken to the extreme, you definitely loose smoothness with renderframes that are not displayed or only displayed for a few lines on the display. That's why the effect in Multi-GPU setups is called microstuttering, which can get really annoying, reducing the value of the higher Fps rate.

Now, to which degree this value is reduced and from what no. of lines a frame should be called a runt frame and not counted toward "perceived smoothness" surely is debatable and will certainly differ from one person to the next.

Me personally, I'm a vsync-off gamer, but I can't stand texture shimmering or edge aliasing, so I turn on supersample-AA whenever I get the chance. Other people don't care about anti-aliasing, but always play vsynced.

 

I do!