Microstuttering tests on ATI Radeon HD 4870 X2

[STEvil]

I think these guys need to see a good video representation of "microstutter".

Click here then click the youtube link in cadaveca's post. http://www.xtremesystems.org/forums/...&postcount=156

I prefer to cause it micropausing still as it seems to look like a timing error to me.

[cadaveca]

Um, STEvil...As Sampsa has shown, a timing error is EXACTLY what Microstutter is all about...the timing when frames are displayed onscreen, and when they are rendered.

Sometimes it's because of the engine, because of needed geometry data for frame 2, that has to wait for frame1 to complete.

Sometimes it's becuase the workload balancing in not right...one gpu ends up doing the majority of work...

Sometimes it's becuase memory usage is high, and incoming data to be buffered in the framebuffer on the main card gets delayed...

Sometimes it's becuase of the driver trying to deal with one of the above three problems.




I saw your post on the hydra, STEvil, and had to laugh. At least you seem aware that this has been a problem since the stinky old article you linked...

[SEA]

I think these guys need to see a good video representation of "microstutter".

Click here then click the youtube link in cadaveca's post. http://www.xtremesystems.org/forums/...&postcount=156

I prefer to cause it micropausing still as it seems to look like a timing error to me.

Just keep in mind that FRAPS'ing game to movie adds stutter itself with
frequency = game fps - capturing fps.

And i just checked and reproduced that kind of stutter. It is particularly well seen if game fps and movie fps are close.
Off course game itself went smoothly.

[RaZz!]

nonetheless, the stuttering is clearly visible in this video. especially when he's walking straight forward. it's like the player gets stuck every now and then.

thanks for the video. never saw microstuttering "in action"

[cadaveca]

Video was w/ QX9650 @ 4ghz, 2x HD3870x2's, 1280x1024 res, high details.

Also, 1680x1050, same settings, no stutter...more work for gpu, yet less stutter? Crysis is a special situation, as is GRiD.

[L7R]

I think these guys need to see a good video representation of "microstutter".

Click here then click the youtube link in cadaveca's post. http://www.xtremesystems.org/forums/...&postcount=156

I prefer to cause it micropausing still as it seems to look like a timing error to me.

That doesn't look like microstuttering like I see it in crysis.
To me best way to see it is start the game and murd...eliminate first 3 kore... enemies and stand in the front of the red flare and move your (or strafe) mouse from left to right. That's it. Fps goes from 10fps minimum upto 110maximum senselesly and the game feels very "stuttery", not jumpy. But in reality that minimum fps is actually the real one, not that average one. If your minimum fps is much more than that, 40 for example, you might even notice it because that 40 is actually pretty smooth. If I use single card I get about 30fps but it feels much smoother than in 45-60fps with CF enabled.

Next place where it's clearly evident is in animation sequence where Aztec is hanging in a tree. Very stuttery even fps counter says 40 average, but again it's that minimum you are actually seeing.

[Aberration]

I must say, you guys are really selling me on the GTX280...

[cadaveca]

L7R, that's not stutter, as you describe. Stutter is when the actual framerate, and what is displayed is different. AVG framerate has no bearing in this measurement.

For example, in that video, FPS remain fairly consistent. Notice the lack of AI, and only scripted events...I purposely chose that area as I knew that performance penalties from extra rendering duties would not be the issue.

As I walk forward in the video, at the end, less geometry data is displayed, and even when the gpu does less work, the stutter is evident...when minimum framerate is increasing.

It's not as simple as just checking FPS counters to find stutter...FPS counter will tell you how many frames are rendered, but not how many are displayed. It's this difference, between rendered FPS, and displayed FPS, that we are talknig about when refering to "Microstutter".



And, Aberration, nV cards are prone to the same problems, but because thier inter-gpu communication is different, it manifests itself at different times, even completely diffferently, than what ATI does.

[safan80]

I must say, you guys are really selling me on the GTX280...

it's not worth it compared to the 4870x2.

[Eastcoasthandle]

You can mimic microstuttering by playing a game then record it in Fraps. Any kind of i/o activity while gaming can give the impression of microstuttering. Therefore, it's source is not bound to a video card. Other variables can come into play that can result in the stutter you see on the screen. These issues are but not limited to:
-drivers
-tight dram timing/subtiming
-unoptimzed game or a game optimized for a speciic hardware
-PCIe at x16 and x4 when dual GPUs are used
-i/o operations runnning in the background (already discussed)
-processes running in the background
-Chipset timing
-vregs of the video card getting to hot (little do people know those need active cooling)
- power play (anyone remember that problem from 3800 series)
-Vista's super prefetching the first 5 minutes (or so) when Vista starts up
-etc

Now I have to ask another question. Did Sampas use the AFR-FriendlyD3D.exe trick to make sure that the cards were scaling properly when using Grid.exe? As there is no mention of what drivers were used to conduct these tests.

[cadaveca]

grid works fine after 8.6/4800 hotfix. Stutter is almost gone.

good list too...you know what's what.

i don't use fraps in any microstutter yesting, for the reason you mentioned.

[Aberration]

L7R, that's not stutter, as you describe. Stutter is when the actual framerate, and what is displayed is different. AVG framerate has no bearing in this measurement.

For example, in that video, FPS remain fairly consistent. Notice the lack of AI, and only scripted events...I purposely chose that area as I knew that performance penalties from extra rendering duties would not be the issue.

As I walk forward in the video, at the end, less geometry data is displayed, and even when the gpu does less work, the stutter is evident...when minimum framerate is increasing.

It's not as simple as just checking FPS counters to find stutter...FPS counter will tell you how many frames are rendered, but not how many are displayed. It's this difference, between rendered FPS, and displayed FPS, that we are talknig about when refering to "Microstutter".



And, Aberration, nV cards are prone to the same problems, but because thier inter-gpu communication is different, it manifests itself at different times, even completely diffferently, than what ATI does.

No I mean a single 280

[Aberration]

it's not worth it compared to the 4870x2.

How do we know that for certain?

There have only been a few very limited tests. I wont be in the market for a few months due to travel, so I will wait until it is all tested very well.

[safan80]

How do we know that for certain?

There have only been a few very limited tests. I wont be in the market for a few months due to travel, so I will wait until it is all tested very well.

I own a GTX280 and I've seen and played with a 4870. that said with how well 4870 Xfire works which has already been tested and the 4870x2 being close if not better than 4870 in Xfire I'm confident that the 4870x2 can easily smash the GTX 280. The last ATI card I owned was a 9800 pro and it was nice but it had problems. ATI has seriously come a long way in the past year and I think people need to put their old feelings about ATI behind them because the only way nvidia can beat the 4870x2 is with a new card.

The biggest problem at the moment with the GTX 280 is the high temp issues and I've already RMAed my card once and I'm stuck with yet another GTX 280 that has 105C load temps.

[bedlamite]

Why some ppl say that Vsync doesnt repair microstuttering?
I mean if Vsync in CF/SLi would be possible with tripple buffering, everything should be fine.
So where is something going wrong?

[safan80]

Why some ppl say that Vsync doesnt repair microstuttering?
I mean if Vsync in CF/SLi would be possible with tripple buffering, everything should be fine.
So where is something going wrong?

because in the past gen of SLI/CF the problem is in the hardware.

[bedlamite]

I understand it's in the hardware.
But Vsync somehow "goes around" it.
Yes/no?

[jdhaan]

Sideport is still there on single gpu's...there is no longer ringstops for that. SO it should be the same for Crossfire via mobo pci-e, or crossfire via vga board pci-e...either way you are no longer using memory resources for multi-gpu communication, and using the side port instead. R700 excels due to lower latency due to less physical distance between the cards, as well the PLX chip should offer at least 16-lanes of pci-e 2.0 between the cards in combination with the side port, whereas some mobo's will not give the same connectivity.


there's moer issues at play in "microstutter" than anyone is really aware of, other than maybe myself and those that built the cards...and even then I doubt they are putting all that much work into it...you can only fix what's broken, you can't fix something not broken!

What I understood, is that the sideport is only used in the 4870X2, not while using crossfire.. 4870 crossfire uses PCI Express 2.0 instead of 1.0 crossfire-bridges.

[Seraphiel]

I understand it's in the hardware.
But Vsync somehow "goes around" it.
Yes/no?

Yes

It goes around it in most titles in my experience with SLI. However, Vsync is handled differently in Vista and XP AFAIK, which may complicate things, and there are those games that don't support Vsync on some setups.

EDIT: My upcoming post about my own microstutter findings are temp delayed, because of downloading the FFCM8Final mod. The downloading of it complicates the last tests and give strange results. Will make a seperate thread in the computer disc forum, and 10 games will be tested in various resolutions and settings. I am only half done until now.

[cadaveca]

I understand it's in the hardware.
But Vsync somehow "goes around" it.
Yes/no?

No, it doesn't, because then the input lag makes apps unplayable. I mentioned this in the past too...it gets so bad that in FEAR, for example, I can dominate any server when playing with a specific friend. We just have our tactics down, blah blah blah... Anyway, enabling v-sync, I'll be at the bottom of the server...no matter what. It's hard to react to something when it's already happened, and you see it after the fact...I mean..reaction time means NOTHING then.

It's a lose-lose situation. Especially when the multi-gpu target audience is what it is at the moment.

jdhann, IF they have disabled the sideport in single-gpu configurations, they aren't very smart. I don't understand how you tihnk this would make ANY sense at all...

Why would they purposely gimp the gpu, when even the heatdump is unimportant?

I mean, with Hector, anything is possible...but really now.

[safan80]

I understand it's in the hardware.
But Vsync somehow "goes around" it.
Yes/no?

Vsync made no difference for me when I ran SLI in the 6 and 7 series. it didn't help my 9800gx2 either.

[Seraphiel]

No, it doesn't, because then the input lag makes apps unplayable. I mentioned this in the past too...it gets so bad that in FEAR, for example, I can dominate any server when playing with a specific friend. We just have our tactics down, blah blah blah... Anyway, enabling v-sync, I'll be at the bottom of the server...no matter what. It's hard to react to something when it's already happened, and you see it after the fact...I mean..reaction time means NOTHING then.

Does it affect the single player experience of F.E.A.R.? Can you play single player games and not notice any problems with v-sync enabled?

[cadaveca]

Some people may not notice it, but I do, obviously. Then there are situations such as in that video where ANYONE can tell...you can't broadly define the problem, unfortuantely, and this is why it's such a bugger.


Heh. I mean, it could be bad enough to cause seizures, I suppose. It's very disorienting, and for some gmaes, it means they've been shelved. Stupid 30-inch monitor is the bane of my computing experience, I tell ya.

[postumus]

HAHA same here my 30inch monitor means i always have to have the fastest single gpu card around, aka the gtx 280 for now =P

[postumus]

oh and I realize they may have improved microstutter but it was readily apparent on the 3870x2, I'm not sure I want to splurge on a r700 if it will still have the same issues, even if it's only under certain games/drivers etc.