Microstuttering tests on ATI Radeon HD 4870 X2

[Vapor]

for the 3rd time..

is there ms @ 2 x 4870x2 ?

No....none of the review sites got 2 cards for this preview.

[Xoulz]

I am surprised many people are not going crazy over this news worthy story....


It seems Micro-stutter might be a thing of the past.





.

[NapalmV5]

No....none of the review sites got 2 cards for this preview.

i rechecked the pic @

http://plaza.fi/muropaketti/artikkel...4870-x2-r700,1

for a moment there i thought its 2x 4870x2

[v_rr]

No....none of the review sites got 2 cards for this preview.

They did:
http://enthusiast.hardocp.com/articl...50aHVzaWFzdA==

[tombman]

@SAMPSA:

No matter what you do, but when testing microstuttering (MS), you have to test at the same FPS LEVEL!! Best @ ~30fps. Adjust all settings so that all cards are @ ~30fps when you test MS.

There is NO sense in testing card A @ 30fps and card B @ 50fps only to find out that card B @ 50fps has less MS than card A.
Usually higher fps result in LESS MS, that is just my experience.

MS is most annoying when playing at ~30fps, because SLI and CF Setups are mostly there to make settings playable that were unplayable with just single gpus.

If you have MS @ 50fps, who cares? Because 50fps WITH MS still feels very playable, similar to 35fps with a single card. But when you get MS @ 30fps, it feels like 20fps- and THAT'S WHERE IT HURTS.

So PLEASE test all single and multi gpu setups @ 30fps- only then can we can find out whether CF really shows better MS handling than SLI or not.

[saaya]

very nice! maybe you should use a different way of graphing it though as it looks like fps per time, its not so easy to read.

so the 4800 series fixes the micro stuttering issue?
how about gtx280 sli and 8800gt sli?

[Sampsa]

Update:

Here are charts showing the difference between frametimes with ATI Radeon HD 4870 X2, 4870 CrossFireX and single 4870 in Race Driver: GRID:



And GeForce 9800 GX2 and GeForce 9800 GX2 in single GPU mode:




Still waiting input from ATI for my questions about microstuttering.

[Shintai]

Having trouble reading a graph?

Does the 4870X2 hold a benefit over 4870 CF? Yes or no?

[Sampsa]

very nice! maybe you should use a different way of graphing it though as it looks like fps per time, its not so easy to read.
?

Graph clearly describes what it is displaying and there is no mention about FPS? It shows how much time between rendering frames variates.

[Sampsa]

Does the 4870X2 hold a benefit over 4870 CF?

Is it really that HARD to read the graph

As you can see 4870X2 and 4870 CF show almost equal variation between times frames are being rendered.

[Solus Corvus]

Does the 4870X2 hold a benefit over 4870 CF? Yes or no?

Clearly no, they appear to perform about the same. What a hollow observation on your part though, since neither appear to have MS. That's definitely an advantage over last generation hardware that...did, at least in more cases. Don't worry, I'm sure you can still find plenty of other bad things about multi-GPU. I'll help you find some if you like.

[Shintai]

Clearly no, they appear to perform about the same. What a hollow observation on your part though, since neither appear to have MS. That's definitely an advantage over last generation hardware that...did, at least in more cases. Don't worry, I'm sure you can still find plenty of other bad things about multi-GPU. I'll help you find some if you like.

I dont need to. They do that fine by themselves.

But in this case it was about the PLX chip and CF sideport. They offer no advantage compared to regular CF.

Is it really that HARD to read the graph

As you can see 4870X2 and 4870 CF show almost equal variation between times frames are being rendered.

Oh I know. You missed the previous part.

So 4870X2 card doesnt hold any "microshutter" advantage.

[Chickenfeed]

As you can see 4870X2 and 4870 CF show almost equal variation between times frames are being rendered.

Graphs are great, thanks for the hard work

I do find it strange however that a single 4870 seems to have larger gaps in the frametimes, its as high as 5ms between the 15th and 16th frame

As far as the sideport showing no benifits, I doubt AMD implemented it for our health... I say wait until finalized samples and drivers reach the reviewers and a wider variety of tests are ran before you jump on it.

[Solus Corvus]

But in this case it was about the PLX chip and CF sideport. They offer no advantage compared to regular CF.

No argument there. That's certainly the way things appear now and I don't expect that to change. But nor do they seem to confer a disadvantage either - at least as far as performance is concerned.

If anything it simply adds some to the already great power draw and heat dissipation. Certainly a big disadvantage for some, though not much of a concern of mine.

[Sampsa]

As far as the sideport showing no benifits, I doubt AMD implemented it for our health... I say wait until finalized samples and drivers reach the reviewers and a wider variety of tests are ran before you jump on it.

As far as I know there aren't anything offical information told about crossfire sideport yet, but I'm hoping to hear some details about it from AMD.

My wild guess is that RV770 graphics chip actually has two interfaces:

One is for normal PCI Express 2.0 bus on motherboard which is being used with Radeon HD 4870 and 4850 graphics cards.

Second is crossfire sideport which is optimized for PLX gen 2 PEX8647 PCI Express switch and used with two GPU Gemini configurations with Radeon HD 4850 X2 ja 4870 X2 graphics cards.

AMD/ATI has been 100 % sure to produce two GPU configuration since the beginning of RV770 designing. Maybe they figured out, why not to create separate optimized interface for PLX switch which would offer some optimized benefits compared to normal PCI Express interface on motherboard.

Nothing above is based on facts, just my own 2 cents

[Hornet331]

@SAMPSA:

No matter what you do, but when testing microstuttering (MS), you have to test at the same FPS LEVEL!! Best @ ~30fps. Adjust all settings so that all cards are @ ~30fps when you test MS.

There is NO sense in testing card A @ 30fps and card B @ 50fps only to find out that card B @ 50fps has less MS than card A.
Usually higher fps result in LESS MS, that is just my experience.

MS is most annoying when playing at ~30fps, because SLI and CF Setups are mostly there to make settings playable that were unplayable with just single gpus.

If you have MS @ 50fps, who cares? Because 50fps WITH MS still feels very playable, similar to 35fps with a single card. But when you get MS @ 30fps, it feels like 20fps- and THAT'S WHERE IT HURTS.

So PLEASE test all single and multi gpu setups @ 30fps- only then can we can find out whether CF really shows better MS handling than SLI or not.

QFE!

[Iconyu]

As far as I know there aren't anything offical information told about crossfire sideport yet, but I'm hoping to hear some details about it from AMD.

My wild guess is that RV770 graphics chip actually has two interfaces:

One is for normal PCI Express 2.0 bus on motherboard which is being used with Radeon HD 4870 and 4850 graphics cards.

Second is crossfire sideport which is optimized for PLX gen 2 PEX8647 PCI Express switch and used with two GPU Gemini configurations with Radeon HD 4850 X2 ja 4870 X2 graphics cards.

AMD/ATI has been 100 % sure to produce two GPU configuration since the beginning of RV770 designing. Maybe they figured out, why not to create separate optimized interface for PLX switch which would offer some optimized benefits compared to normal PCI Express interface on motherboard.

Nothing above is based on facts, just my own 2 cents

ATI have been tweaking drivers so they avoid talking over the PCI bus, and via the cross fire channel. If the normal 4780's use the side port as well they won't vary very much from the 4780x2 results.

The only way to find out for sure would be to run the 4870's on a 1.1PCIe set up, as the 3870x2's PLX chip is only 1.1 you'll be able to see if that's the cause of micro stutter in this case, if 1.1 doesn't make a difference than all the 4780's are using the side port/driver tweaks.

Have you tried SLI (with separate cards) on a older mobo, or tried forcing the mobo to PCIe 1.0/1.1 and recording results?

[kromosto]

Update:

Here are charts showing the difference between frametimes with ATI Radeon HD 4870 X2, 4870 CrossFireX and single 4870 in Race Driver: GRID:

so is really 2 gpus are enabled i grid. because below graph show

x2 or cf has something like 57 to 66 fps variation
single card has 50 to 66 fps variation.

this is confusing isnt it

[Xoulz]

Is it really that HARD to read the graph

As you can see 4870X2 and 4870 CF show almost equal variation between times frames are being rendered.

LOL..!

I think they are just playing around. Your graph clearly illustrates what is going on.

Thanks.





.

[Chickenfeed]

Were some of you absent when then made you do graphs in school

@ kro

If you read all the graph's header you will see the information represents ONLY 30 separate frames. The lines for each card indicate how much time passes between each frame. The smaller the difference, the smoother the image update. Simple enough? Clearly in Grid all the tested 4870s exhibit consistent render times thus theres no perceivable stuttering.

[Cooper]

TBH I'm a bit confused about the graphs as well.

Sampsa if it's possible can you add some comments on the methodology and what we are looking for in the graphs and provide the theoretical ideal graph as reference.
Would be really usefull

[kromosto]

Were some of you absent when then made you do graphs in school

@ kro

If you read all the graph's header you will see the information represents ONLY 30 separate frames. The lines for each card indicate how much time passes between each frame. The smaller the difference, the smoother the image update. Simple enough? Clearly all the tested 4870s exhibit consistent render times thus theres no perceivable stuttering.

ok let me simplify my question and again ask it.

graph shows that for single 4870 it took something like 20ms to 15ms to show frames on screen. for 4870x2 it took something like 18ms to 15ms to show frames on the screen. i think we are agreed on this as it is shown on the screen.

so fps of the single 4870 is between:

1/0.020 and 1/0.015 this means 50fps to 66fps

for 4870x2:

1/0.018 to 1/0.015 means 55 to 66fps

what i think is this gap between 4870 and 4870x2 should be much more. and i am asking why it is like this?

[Cooper]

kromosto

This may indicate that CF does not work in GRID

[kromosto]

kromosto

This may indicate that CF does not work in GRID

yes that is what i am thinking. asking it to be sure.

[Chickenfeed]

Oh sorry about that didn't realize you were working the math there, I see what your getting at now.

As far as CF not working in Grid, that would be odd as the one preview shows the 4870s owning that game ( both the 4870s and even more so the X2 )