[MfA]

Can't you force supertiling with a 3870x2?

[AlexV]

Originally Posted by MfA

Can't you force supertiling with a 3870x2?

Not really.

[Arnold Beckenbauer]

Originally Posted by Morgoth the Dark Enemy

Not really.

With AI off?

[AlexV]

Originally Posted by Arnold Beckenbauer

With AI off?

With AI off everything is off, from optimizations to Crossfire itself, so its hard to discern what's going on. I think SuperTiling died a horrible death at some poing, between driver revisions...it was initially available in Vista drivers, but disappeared at some point. I may be wrong though and it would be nice if one of the ATi guys cleared this up.

[Albuquerque]

I thought supertiling still existed for those games that were already "tweaked and accounted for" in the drivers. Even without it, I think the CFAA modes still do something entirely unrelated to AFR; I can't exactly say why I believe that though.

[AlexV]

Originally Posted by Albuquerque

I thought supertiling still existed for those games that were already "tweaked and accounted for" in the drivers. Even without it, I think the CFAA modes still do something entirely unrelated to AFR; I can't exactly say why I believe that though.

Nope. I've seen no evidence of supertiling still being used for any title at this moment. CFAA modes don't do something unrelated to AFR...what you're thinking about is SuperAA, which only exists in the form of 16X AA for the 2900CF and upwards(including the 3870X2 here), that mode basically has both GPUs render at 8X MSAA and the master chip combines the samples in order to get 16X AA-that is indeed a different rendering mode, as detailed by ATi in their documentation.

Both SuperTiling and Scissor seem to be MIA, at least in my experience.

[chavvdarrr]

Originally Posted by Morgoth the Dark Enemy

Both SuperTiling and Scissor seem to be MIA, at least in my experience.

Which is a shame AFR sux cause of lags

[Albuquerque]

Hmm, well if that's the case, then I'm not really disappointed. I've been playing a large number of games over the last four weeks ownership of my pair of 3870's, and I've yet to have any complaints about "lag".

I just assumed it was because most of my games are "older" and thus would already have optimized profiles for scissor / supertile. Guess I was wrong That's ok though, I'm still loving the performance with all the AA and AF thrown on.

[AlexV]

Originally Posted by chavvdarrr

Which is a shame AFR sux cause of lags

It depends, like everything else in life. It's not as bad as it's made out to be...it can be horrid for some ppl. As I said, it depends. There are worse things about AFR then perceived lag...

[MfA]

Such as? The stuttering should only be present in highly GPU limited games and even then it should be easy to fix in the driver.

PS. can NVIDIA still do SFR with their modern drivers?

[AlexV]

Originally Posted by MfA

Such as? The stuttering should only be present in highly GPU limited games and even then it should be easy to fix in the driver.

PS. can NVIDIA still do SFR with their modern drivers?

Such as the fact that it seems to be the first things developers bork at some point or another, by deciding that they absolutely need this or that technique that doesn't play nice with it. Such as data duplication. Those are about 2 examples.

But since it's advantages far outweigh its disadvantages, I guess it'll be here to stay.

[Albuquerque]

Originally Posted by MfA

Such as? The stuttering should only be present in highly GPU limited games and even then it should be easy to fix in the driver.

PS. can NVIDIA still do SFR with their modern drivers?

Stuttering? I guess I haven't seen that yet. More descriptive term perhaps? Or a way to force it?

And I'm sure NV can still do split-frame rendering just like ATI can, the question is do they?

[ChrisRay]

SFR is supported if you force it. But only like 2 games that Nvidia has profiles for actually use it.

[MfA]

Originally Posted by Morgoth the Dark Enemy

Such as the fact that it seems to be the first things developers bork at some point or another, by deciding that they absolutely need this or that technique that doesn't play nice with it.

For every such technique there is an AFR compatible alternative for which the only downside is that it uses more memory. They'll adjust.

[AlexV]

Originally Posted by MfA

For every such technique there is an AFR compatible alternative for which the only downside is that it uses more memory. They'll adjust.

Sure. Anytime soon?Check Crysis, the jewel of technology?That screwed AFR from the start, and is still screwing multi-GPUs.

I'm not saying it's not fixable. What I'm wondering is how much time devs will actually devout to it. I'm hoping that with ATi's current push for AFR and nV seeming to push it as well things will change.

[Blacklash]

X-Bit's new review of the X2-

"The new card won six out of the nineteen tests, at least at high resolutions...

On the other hand, it was either equal to or slower than the single-chip ATI Radeon HD 3870 in seven out of the nineteen tests..."

http://www.xbitlabs.com/articles/vid..._19.html#sect0

Just like I've been saying from my experience with the X2; inconsistent, unpredictable and not a new champ.

[mczak]

Originally Posted by Morgoth the Dark Enemy

It depends, like everything else in life. It's not as bad as it's made out to be...it can be horrid for some ppl. As I said, it depends. There are worse things about AFR then perceived lag...

But AFR has its limit. Most notably it doesn't scale. 2 cards (chips) is about the limit which is useful, so if the future is multichip they better figure out a different method.
Maybe that's the reason CrossfireX isn't ready.
By doesn't scale I don't mean it wouldn't spit out more frames per second, but it's useless since input lag won't be improved compared to a single card. And if there's no perfect scaling with multiple cards (and when does this happen...) lag actually gets worse. Even with perfect scaling lag just gets too big for interactive applications (if you could squeeze out 40 fps with a quad-afr setup you'd have 100ms lag) - sure it's not a problem if your application runs at 400fps with 4 cards, but you wouldn't need 4 cards to run it fast enough in the first place in this case...

[fellix]

Originally Posted by Blacklash

http://www.xbitlabs.com/articles/video/display/ati-radeon-hd3870-x2_19.html#sect0

I just can't get it, why Xbit always conduct their testing with driver Alpha-test AA enabled?

[Albuquerque]

Originally Posted by Morgoth the Dark Enemy

Sure. Anytime soon?Check Crysis, the jewel of technology?That screwed AFR from the start, and is still screwing multi-GPUs.

This doesn't mirror my experience at all. I own the full Crysis game, and I own two 3870's. Crossfire improves my DX9 performance by approximately 85% across the board; DX10 performance is spottier but gets about a 65% gain.

Crossfire on my system is the difference between "barely playable at 1680x1050 using high settings" and "entirely enjoyable at 1680x1050 using tweaked ultra-high settings". We're talking average framerates of better than 40FPS, and minimum framerates in the high 20's -- versus averages of low 30's and minimums in the teens or even single digits.

I think you're getting bad information.

[MfA]

Originally Posted by mczak

But AFR has its limit. Most notably it doesn't scale. 2 cards (chips) is about the limit which is useful, so if the future is multichip they better figure out a different method.

It's not the future, not the near future anyway. It has always been a way to generate hype and little else.

The only alternative to AFR which they could implement cheaply and transparently enough is sort last (ie. divide render calls across the multiple GPUs and composite the z-/frame-buffers at the end).

[AlexV]

Originally Posted by Albuquerque

This doesn't mirror my experience at all. I own the full Crysis game, and I own two 3870's. Crossfire improves my DX9 performance by approximately 85% across the board; DX10 performance is spottier but gets about a 65% gain.

Crossfire on my system is the difference between "barely playable at 1680x1050 using high settings" and "entirely enjoyable at 1680x1050 using tweaked ultra-high settings". We're talking average framerates of better than 40FPS, and minimum framerates in the high 20's -- versus averages of low 30's and minimums in the teens or even single digits.

I think you're getting bad information.

From my eyes, the software I use and my hardware?Gee, I must be really misinformed there. Crysis is spotty at best...I don't know how every now and then someone seems to get huge fps at huge resolutions with moderate HW. Even SLi barely scales in most reviews, and when it scales it doesn't scale as well as everyone hoped, given the fact that Crysis was expected to be very pixel bound, a scenario where scaling should be at its best.

And considering I'm using a QX9770 at 4.2 in tandem with a pair of 3870X2s(no Quad drivers for me yet though), I'm not sure that it's really the HW that is lacking.

[Hubert]

Originally Posted by Hubert

I noticed stuttering in The Witcher ... what was fluid with an X1800XT, at 25 fps, seems stuttering at 45 fps using a 3870X2, Catalyst 8.1. Trade Quarter map was always stuttering but Temple Quarter was not ... and yet I have the feeling that X1800 XT was more fluid while having less fps.

Stuttering goes away if I deactivate AA. So it is something else. I am a bit disappointed, as I sold my X1800XT ... it provided better quality gaming in The Witcher. And you freaked me out with this microstuttering thing. Now I'll have to read all the posts of this thread.

[Skinner]

Originally Posted by Blacklash

X-Bit's new review of the X2-

"The new card won six out of the nineteen tests, at least at high resolutions...

On the other hand, it was either equal to or slower than the single-chip ATI Radeon HD 3870 in seven out of the nineteen tests..."

http://www.xbitlabs.com/articles/vid..._19.html#sect0

Just like I've been saying from my experience with the X2; inconsistent, unpredictable and not a new champ.

COD4 (water on LOW!!! there is a perf. BUG no one mentions ???!!!) and Oblivion are way faster then this in my experience with (the X2) CF.
They are among those games which are scaling very nicely in CF (Bioshock, GOW dx9, Stalker, FEAR, Blacksite (with FEAR.exe), Fable, DreamFall, Timeshift, Splintercell DA, Doom3 engines and Hl2 engines are the rest in my list).

But there are still some left, like Hellgate dx10, GOWdx10, UT3 (low minfps/severe fps drops in some maps) Gothic3, Two Worlds and Hitman BM which don't do much in CF for example.

Crysis is improving but, still has stuttering and fps drops (constant data reading/caching?)

I am one of those who think Hardopc testsstyles are needed to get the whole picture, play a game, see how it cache the data, do a fast turn arround and see the difference, see how it deals with the involvement of heavy AI calls to the CPU, which stays hidden in static tests.

[Albuquerque]

Originally Posted by Morgoth the Dark Enemy

From my eyes, the software I use and my hardware?Gee, I must be really misinformed there.

Dunno, that or you have absolutely terrible luck with hardware and software. I can recreate the vast improvement in performance on both XP32 and Vista64; it's not OS dependant. I'm on a GA-X38-DS4 and an E8400 at 4.3Ghz, and since Crysis still isn't using quad cores, I'm pretty sure that means your performance is capable of being around the same as mine.

So, I have two operating systems where CF works as-advertised, and you have a self-proclaimed uber-rig that sucks balls. Sorry for your luck but your experience isn't the bottom line of crossfire capabilities.

[AlexV]

Originally Posted by Albuquerque

Dunno, that or you have absolutely terrible luck with hardware and software. I can recreate the vast improvement in performance on both XP32 and Vista64; it's not OS dependant. I'm on a GA-X38-DS4 and an E8400 at 4.3Ghz, and since Crysis still isn't using quad cores, I'm pretty sure that means your performance is capable of being around the same as mine.

So, I have two operating systems where CF works as-advertised, and you have a self-proclaimed uber-rig that sucks balls. Sorry for your luck but your experience isn't the bottom line of crossfire capabilities.

Guess me and all of the reviewers have rigs that suck balls and a seeming inability to get this game to scale. Oh well, thank the lord for forums...without em, only us unlucky sobs would be around, but with forums, trust the truth to come out.

Probably the whole bunch of other guys with both CF and SLi which complain about poor performance are also non-existant. Because you have two operating systems where CF works as advertised. YOU. I am still getting things right...YOU-doesn't YOU=1 data point. Doesn't:
me+
http://www.techreport.com/articles.x/14161/7+

http://www.bit-tech.net/hardware/200...n_hd_3870_x2/7+

http://www.hardocp.com/article.html?...50aHVzaWFzdA==+

http://www.xbitlabs.com/articles/vid..._11.html#sect0+

http://www.digit-life.com/articles3/...0-3-part3.html+

http://www.driverheaven.net/reviews/...iew/crysis.php+

probably quite a few others=something maybe a tad bit more then 1?Maybe, just maybe?There are cases where it scales, but it's inconsistent. It doesn't always scale as expected considering the fact that IT SHOULD BE PRIMARILY PIXEL BOUND, AND THAT'S THE BEST CASE SCENARIO FOR MULTI-GPUs. I won't talk about the ludicrous claim of average framerates over 40 and lows over twenty, considering that 3 Ultras in SLi don't seem to get that WHEN USING VERY HIGH IN DX10. Sorry, tweaked settings don't tell me anything unless properly detailed, so I'll assume you're talking about Very High there.

So nV is working on scaling, AMD are working on scaling, there seems to be a number of bugs with 2D clocks coming in in CF mode in Crysis at certain points, and with the 64-bit version of the darned game not scaling whilst the 32-bit one seeme to scale, and yet you get an 85% improvement in DX9(holy schmoli, that's a lot, even IHVs are quite happy to suggest 70% scaling to be expected generally with AFR) and a 65% one in DX10(yikes, also close to the magic 70% number). You also get a doubling of minimums, when a side-effect of using multi-GPUs is that due to the need of uploading data to both GPUs there'll be points where framerates plummet for short periods, and thus minimums are almost always lower in that scenario(granted, maybe Crysis is an exception, although it happens there as well on some maps at certain points, but let's ignore that fact). All I can say is:good for you.