Go Back   Hardware Canucks > HARDWARE > Video Cards

    
Reply
 
LinkBack Thread Tools Display Modes
  #31 (permalink)  
Old August 20, 2008, 01:07 PM
DK2 DK2 is offline
Allstar
F@H
 
Join Date: Apr 2008
Location: GTA
Posts: 821
Default

I kind of meant either the gpu or communications (CF/SLI) getting behind between cards.
I was wondering though, wouldn’t having double or triple frame buffering turned on
hide the problem. I don’t have an ATI card, but I know Nvidia
usually has the frame buffering turned off by default, or used to.
While buffer #1 gets built, buffer #3 gets sent to the monitor, while buffer #2 is the holding
buffer between the two, so when the frames skew, one is always ready to be sent.
Reply With Quote
  #32 (permalink)  
Old August 20, 2008, 01:13 PM
Banned
F@H
 
Join Date: Jun 2007
Location: Edmonton
Posts: 1,628
Default

triple-buffering(a framebuffer of 3 frames, rather than the standard two) doesn't affect microstutter. Microstutter is caused by the frames never able to make it to the primary card, adn the driver dropping frames from the slave cards framebuffer. triple buffering will allow the driver greater selection in which frames to drop, but doesn't help get the frames to the primary gpu's compositing engine...the bandwidth limitiation of the crossfire interconnect is the issue. Enter teh XSP, with 5Gb, over 1GB...
Reply With Quote
  #33 (permalink)  
Old August 20, 2008, 01:27 PM
DK2 DK2 is offline
Allstar
F@H
 
Join Date: Apr 2008
Location: GTA
Posts: 821
Default

Since you put it that way. It sounds like the only solution would be to
have a more powerful primary card to handle the extra load and extra bandwidth for the CF connection.
Much simpler would be to just improve the drivers synchronization and arbitration algorithms.
There must be something in hardware or the compatibility of communications protocols that prevents them from doing anything about it.
Just dropping a frame isn’t a very good solution.
Reply With Quote
  #34 (permalink)  
Old August 20, 2008, 01:47 PM
Cheator's Avatar
Hall Of Fame
 
Join Date: Mar 2007
Location: Ottawa, Canada
Posts: 1,093
Default

Quote:
Originally Posted by cadaveca View Post
I never said I'm the only one who understands..clearly AMD does...they've even admitted that the PLX chip in 3870x2 wasa major source of this problem, due to a lack of bandwidth, and have both used the new PLX chip, as well as offering the XSP.

I'm dealing only with your "stutter under 16x/4x Crossfire", which is not microstutter. Microstutter is not an issue fixed purely by pci-e bandwidth, but stutter from 16x/4x can be fixed, by using a board with 2x16 electrical slots.

This is the only issue I have with your claims. using 16x/4x for crossfire is nothing more than a bad system config, and builder error.


Maybe you should enlighten me... you clarified how?
Omg come on man!! I clarified this in a post afterwards! I was mentioning a P5k-E in the first post to imply that there were other problems, and clarified that I had both chipset bottleneck and CF bottleneck... see? You even replied to it! I know you read some of the post... PLEASE read it again... jeez...
__________________
Optimus: Antec P180, Asus P5Q-E, E8400, 4 GB RAM, Asus 4850, Seagate 320gb, Antec TC II

Teletraan I: Asus P5K-E/WIFI, E4500, 2 GB RAM, ATI 3450, Sapphire Theatrix, 2.25TB storage, Antec SP2 400w

Galvatron: Asus M2NPV-VM, 3800+ X2, 2GB RAM, NV8500GT, Seasonic 500w

Tracks: Acer AS5920, C2D 2.1ghz 45nm, 3GB DDR2, NV8600m GT, 250gb hdd

Hound: Gigabyte GA-MA74GM-S, AMD 4400+ X2, 1 GB RAM, 2900XT, 500w Seasonic

My Heatware!
| Got RMA? Tell us your experience! | VOIP Help? PM ME!
Reply With Quote
  #35 (permalink)  
Old August 20, 2008, 01:50 PM
Banned
F@H
 
Join Date: Jun 2007
Location: Edmonton
Posts: 1,628
Default

Quote:
Originally Posted by DK2 View Post
Since you put it that way. It sounds like the only solution would be to
have a more powerful primary card to handle the extra load and extra bandwidth for the CF connection.
this would be akin to leaving the slave card in 2D clocks, and it didn't work very well.
Quote:
Originally Posted by DK2 View Post
Much simpler would be to just improve the drivers synchronization and arbitration algorithms.
This is the current solution, but it leads to the mis-matched frametimes that Cheator refers too..again, not the problem ,but a problem caused by the attempted solution.


Quote:
Originally Posted by DK2 View Post
There must be something in hardware or the compatibility of communications protocols that prevents them from doing anything about it.


In this regard I have no info. It could be more than possible that there is a patent holding them back, I do not know for sure. I don't currently have any contacts @ AMD that I can speak with on the issue, so I've been left "flying blind".
Quote:
Originally Posted by DK2 View Post
Just dropping a frame isn’t a very good solution.
Whne you can only push so many frames across the interconnect, there's not much else for choices. But with the 4870x2...there is another choice, in singlecard config...the XSP. Used in conjunction with the already present CBI(Crossfire Bridge Interconnect), total bandwidth between the cards is 6Gb. plus the pci-e interconnect. But we are still left with the issue of differring latencies across each interconnect.

Cheator, I think you are missing that you are showing an issue, but I delve futher, into what causes the issue you refer to. It's not a matter of saying the same thing differently, it's a matter of looking deeper into the issue. Again, the mismatched frame times are not the source of the Microstutter, it is merely a side-effect of one of the attempted solutions.

Last edited by cadaveca; August 20, 2008 at 01:56 PM. Reason: response to Cheator
Reply With Quote
  #36 (permalink)  
Old August 20, 2008, 02:13 PM
Cheator's Avatar
Hall Of Fame
 
Join Date: Mar 2007
Location: Ottawa, Canada
Posts: 1,093
Default

What are you talking about now? You were hung up on my P35 limiting my crossfire, now you're hung up on semantics?
__________________
Optimus: Antec P180, Asus P5Q-E, E8400, 4 GB RAM, Asus 4850, Seagate 320gb, Antec TC II

Teletraan I: Asus P5K-E/WIFI, E4500, 2 GB RAM, ATI 3450, Sapphire Theatrix, 2.25TB storage, Antec SP2 400w

Galvatron: Asus M2NPV-VM, 3800+ X2, 2GB RAM, NV8500GT, Seasonic 500w

Tracks: Acer AS5920, C2D 2.1ghz 45nm, 3GB DDR2, NV8600m GT, 250gb hdd

Hound: Gigabyte GA-MA74GM-S, AMD 4400+ X2, 1 GB RAM, 2900XT, 500w Seasonic

My Heatware!
| Got RMA? Tell us your experience! | VOIP Help? PM ME!
Reply With Quote
  #37 (permalink)  
Old August 20, 2008, 02:26 PM
Banned
F@H
 
Join Date: Jun 2007
Location: Edmonton
Posts: 1,628
Default

no, just the inaccuracy of your statements. Cause and effect are not one and the same. You called ME out, as inaccurate, i have merely shown how and why your statements are inaccurate themselves.

No emotion involved.

Quote:
Originally Posted by Cheator View Post
Its not about the frame rate, its about the length of time between frames being displayed.
Not Microstutter

Quote:
Originally Posted by Cheator View Post
Microstuttering is a problem, and I witnessed it somewhat because of a bottleneck in the P5K-E.
Not microstutter..you experienced something else.
Quote:
Originally Posted by Cheator View Post
It occurs on games that usually bottleneck cards and is not restricted to crossfire.
Not microstutter, and definately is restricted to multi-gpu.

Quote:
Originally Posted by Cheator View Post
And yes it has every bit to do with the length of time between frames displayed, you just described what I said in a different way.
Nope, wrong.

Quote:
Originally Posted by Cheator View Post
you're not the only person on the planet that understands what microstuttering is...
Maybe not, but you don't tget it, at least not yet, given your statements. Shall I continue?


Last edited by cadaveca; August 20, 2008 at 02:36 PM.
Reply With Quote
  #38 (permalink)  
Old August 20, 2008, 03:12 PM
Cheator's Avatar
Hall Of Fame
 
Join Date: Mar 2007
Location: Ottawa, Canada
Posts: 1,093
Default

What can I say? You're taking the vague detail in my posts and claiming that its not microstuttering. When I clarify, you don't read. So there is no point in you and I talking anymore. you're not listening, and I don't care for what you have to say since you're talking about something else.

GPU "Microstuttering" FAQ - [H]ard|Forum

Quote:
When running two Graphics Processing units (GPUs) in tandem, via Crossfire or SLI, in Alternate Frame Rendering (AFR) mode, the two GPUs will produce frames asyncronously (for lack of a better term). Microstuttering can be expressed one way as your computer experiancing, in extreme rapid sucession, a high FPS, followed by a low FPS, followed by a high, then low, and so on.
This is what I was experiencing.

Quote:
How does microstuttering impact me in game?
Microstuttering can make playing what fraps is calling a 60fps game feel identical to playing a 30 fps game (literally, this is a potential true case mathmatically).
This happned to me with Crysis especially. I chalked it up to P35 limitations, upgraded to the P45 and still experienced the issue. Your claim of changing mobos is sound, especially when considering the limitations of P35. The p45 did not solve the problem, nor did an x48 I demoed.

You assumed (for some reason) I hadn't upgraded my board to fix the issue.

Quote:
In my example lets go with a game running at 50 FPS. A frame is built and displayed by GPU "A". Exactly 20 miliseconds (ms) later GPU "B" must have completed building and displaying its frame. Exactly 20 ms after that GPU "A" must have finished building and displaying the next frame, and so on. Each frame must be displayed in exactly 20 ms after the previous one.
so basically its about the length of time between frames being displayed. Or, to be just as general

Quote:
Originally Posted by Cheator
Its not about the frame rate, its about the length of time between frames being displayed

So where is the problem? That site agrees with me. So does this one. And reading many forum posts I seem to be smack dab on the money.

There is no point in continuing this. You'll probably read 4 or 5 words from this post and assume something else.
__________________
Optimus: Antec P180, Asus P5Q-E, E8400, 4 GB RAM, Asus 4850, Seagate 320gb, Antec TC II

Teletraan I: Asus P5K-E/WIFI, E4500, 2 GB RAM, ATI 3450, Sapphire Theatrix, 2.25TB storage, Antec SP2 400w

Galvatron: Asus M2NPV-VM, 3800+ X2, 2GB RAM, NV8500GT, Seasonic 500w

Tracks: Acer AS5920, C2D 2.1ghz 45nm, 3GB DDR2, NV8600m GT, 250gb hdd

Hound: Gigabyte GA-MA74GM-S, AMD 4400+ X2, 1 GB RAM, 2900XT, 500w Seasonic

My Heatware!
| Got RMA? Tell us your experience! | VOIP Help? PM ME!
Reply With Quote
  #39 (permalink)  
Old August 20, 2008, 03:44 PM
Banned
F@H
 
Join Date: Jun 2007
Location: Edmonton
Posts: 1,628
Default

No, I read it all, as I have the other forums. I've also mentioned elsewhere that the majority of the sites reporting on the issue aren't reporting it accurately. You're jsut following them, so I don't blame you. But unless YOU have done the testing, how do you know they are accurate?

Now, here's the issue with all those other threads on this, that I have intentionally stayed out of...you cannot get Microstutter from apps that have framerates less than what exceeds the bandwidth of the interconnect. The problem of loss of sync-d frames(oi, I'm getting tired of repeating this) is due to how the driver deals with the framebuffer of the slave cards, and which frames get pushed out along the CBI. PCGH called the lack of sync, Microstutter. And again, although they mention the lack of sync, they do not explain why. There's the difference.

Of course, you're aware then, that AFR is NOT the only way to render? And that the other rendering modes have stutter as well? So what happens in those cases?

HARDOCP...lol...lol...lol...they aren't "hard" enough, just yet.

Quote:
If I notice microstuttering, can I minimize/eliminate it?
Quote:
Yes. By running the game at a setting where your graphics cards are able to output more than the monitors refresh rate (that is, the maximum FPS the monitors are capible of; the pixels on your screen can only change so fast) microstuttering is eliminated completely.
Hmmm..my stutter above 100 FPS must be my imagination then. It's ok, I'm a ninja.

I can pull up many times that both sites have been wrong, so don't take them so seriously. Really. If you speak to some of the the more adept members here, I'm sure they'll have lots to say about both of those sites...


Here's HARDOCP's source:

PCGH: Micro stuttering on multi GPU solutions - XtremeSystems Forums

Note my missing posts...I'm quoted, so some remain, but for reasons I cannot go into, I had to remove them.


Issue...poor performance in comparison between singlecard to multi-card.
Issue...poor performance is due to lack of sync of frames
Issue...Lack of sync is caused by lack of bandwidth of CBI that links framebuffers of each card.

The three issues lead to the real problem, and your mind is stuck in the middle. Please unstick it and blame the real problem.:rtfm:

Just so you're aware, here's me almost exactly 3 years ago, when I begun my forays into causes of graphical stutter:

XtremeSystems Forums - View Single Post - Dothan vs San Deigo!!! Title Bout!!

Quote:
Originally Posted by cadaveca View Post
depends on the game. BF2, for example...x850xtpe @ 620/620..yes, there was stutter. 7800GTX...still there, but it did not take 5 minutes for it to go away.
Now, this tells me that the problem is because textures have not been fetched appropriately, and seeing how the TWIWMTBP logo is on the back of the box of my BF2 case, i'd say that the app's programming was at fault...some people will remember the valve source engine doing kinda the same to nvidia, with water features.
Once the map has been loaded a few times on either card, there is no lag. I also find that the stutter lasts as long on my 9800xt based cards as it does with the x850xtpe...again, pointing to texture fetches.

but hey, whadda i know? :stick:

Last edited by cadaveca; August 20, 2008 at 03:53 PM. Reason: gotta stop taking so long to post!
Reply With Quote
  #40 (permalink)  
Old August 21, 2008, 03:27 PM
kev.nam's Avatar
Top Prospect
F@H
 
Join Date: Aug 2008
Location: Greater Toronto Area
Posts: 55
Default

Ahem. We are CANADIANS. We don't flame on forums, we are better than that :)
__________________
Main Rig: E8400 @ 3600mhz @ 1.248V | Asus P5Q Vanilla | OCZ 2x2GB 800mhz 4-4-4-12 | eVGA GTX 260 Core 216 55nm | WD 500GB AAKS | Corsair HX520 | OCZ Vendetta 2 | Antec Three Hundred | Samsung 2443BW
HTPC: E5200 @ 3150mhz Passive Thermaltake Big Typhoon VX | EVGA 780i SLI | Asus 9600GSO 512mb | OCZ Platinum 2GB | Seagate 500GB + WD 1TB RE3 | OCZ GamexStream 600W | LG70 47" TV 120hz

Reply With Quote
Reply


Thread Tools
Display Modes