.

kithylin wrote:

Actually it -IS- perfectly smooth and glorious in games, -IF- you have a fast enough CPU to keep up with the extra overhead of managing both cards. That's the key. When you pair SLI with a slower cpu then you get hitches and stalls and such and it's not very fun. But with decent fast chips it's just smooth and you never notice anything but high frames.

That's not exactly what I'm referring to. Basically, with how alternate frame rendering works (SLI), there is an added delay with frame delivery. This causes the reported frame rate to not actually be as smooth as it would be if a single GPU was rendering. There have been some tests done to measure this in recent years.

For example
http://techreport.com/news/21625/multi-gpu-mi … ptured-on-video

DirectX 12 multi-GPU has options other than AFR that might make this issue go away. If any games bother to develop for it.

.

Btw - R600 CF is really far from "smooth". That thing is ultimate microstuttering generator. 😁
Also no issues with Sandy Bridge stability here... running one in my home server since January 2012 more or less 24/7 in lowend H61 board and not a single crash.

TheAbandonwareGuy wrote:

Avoid 680i/780i motherboards if you can. There OK for light usage and testing but they like to die quickly if you ask them for moderate or worse daily use.

Yeah these things get HOT. The chipset is actively cooled but still manages to run around 70C. One could convincingly run these daily but only after replacing the northbridge cooler with something massive and keeping the board in a case with very good airflow. These boards are also made in that cursed timeframe when everyone switched those early ROHS solder balls witch crack under moderate heat.

havli wrote:

Btw - R600 CF is really far from "smooth". That thing is ultimate microstuttering generator. 😁
Also no issues with Sandy Bridge stability here... running one in my home server since January 2012 more or less 24/7 in lowend H61 board and not a single crash.

That is very driver and game dependent. Using the latest drivers there's barely any microstutter.

kanecvr wrote:

That is very driver and game dependent. Using the latest drivers there's barely any microstutter.

Well, I find this hard to believe... as my experience speak otherwise. Pretty much any game where CF has positive scalling also is plagued by stuttering. For example these few images (sure, this was measured using catalyst 11.x... but I seriously doubt AMD would do any tweaks beyond this point, R600 was long dead after 2011).

Bioshock 1600x1200

FEAR 1600x1200

S.T.A.L.K.E.R.: Shadow of Chernobyl 1600x1200

..

Perhaps for you it looks smooth, but in reality it simply isn't. Yes, fps counter shows twice as big number and that's it... CF cost twice as much, generates twice as much heat, more noise and in the end the actual "smoothness" is almost the same as with single GPU of the same type.

Of course this applies for SLI as well.

havli wrote:

Perhaps for you it looks smooth, but in reality it simply isn't. Yes, fps counter shows twice as big number and that's it... CF cost twice as much, generates twice as much heat, more noise and in the end the actual "smoothness" is almost the same as with single GPU of the same type.

Of course this applies for SLI as well.

It depends on a lot of things. What motherboard you're using, what video card drivers, how fast the HT bus is running (or whatever tech the board uses to move data from the CPU or northbridge to the video cards and so on. In any case the stuttering was not noticeable. I'm usually sensitive to stuttering - even if I can't perceive it I get eye strain - but that did no happen on my 2900xt cf rig back in the day, nor on the LGA2011 rig I had a few years ago (It ran 2xR9 280x in CF). I did however notice slight stuttering on my first Ivy bridge machine - Z77 MSI board, i5 3350p, 2x HD 7950.

I split the micro stutter discussion of the 939 thread.

..

Just noticed the "no stuttering with fast CPU" opinion. Well, I can't really say how AFR works in CPU limited situations (never bothered to test it)... but at 100% GPU load there is no reason for it to disappear. More CPU power simply allow both GPUs to produce as many frames as possible. But hardly can affect frames distribution in time, that depends on GPU driver and possibly hardware implementation of SLI/CF.
like:
------frame 1 (GPU1)---frame 2 (GPU2)-----------------------------------------------------------------------frame 3 (GPU1)---frame 4 (GPU2)
------frame 1 (GPU1)-------------------------frame 2 (GPU2)-------------------------frame 3 (GPU1)---------------------------frame 4 (GPU2)

The R600 stuttering above was measured using i5 2500k @ 4.5 GHz + Gigabyte P67 board. For sure that is more than enough performance to get 2x R600 running at full power.
For those interested, here is complete set of my GPU benchmarks (frametimes included). Everything was running on the mentioned 2500k OC. As the results suggest, not only 2900 XT CF is more or less useless, but also HD 4870 CF, GF6 SLI, GF7 SLI and GF8/9 SLI... although these are slightly better. 😵
http://www.hw-museum.cz/data/Prubehy_fps_benchmark_VGA3.rar

Maybe modern GPUs and DX11 games are better in this matter (no personal experience), although I wouldn't be very optimistic https://www.pcper.com/reviews/Graphics-Cards/ … nd-Theft-Auto-V

havli wrote:

Just noticed the "no stuttering with fast CPU" opinion. Well, I can't really say how AFR works in CPU limited situations (never […]
Show full quote

Just noticed the "no stuttering with fast CPU" opinion. Well, I can't really say how AFR works in CPU limited situations (never bothered to test it)... but at 100% GPU load there is no reason for it to disappear. More CPU power simply allow both GPUs to produce as many frames as possible. But hardly can affect frames distribution in time, that depends on GPU driver and possibly hardware implementation of SLI/CF.
like:
------frame 1 (GPU1)---frame 2 (GPU2)-----------------------------------------------------------------------frame 3 (GPU1)---frame 4 (GPU2)
------frame 1 (GPU1)-------------------------frame 2 (GPU2)-------------------------frame 3 (GPU1)---------------------------frame 4 (GPU2)

The R600 stuttering above was measured using i5 2500k @ 4.5 GHz + Gigabyte P67 board. For sure that is more than enough performance to get 2x R600 running at full power.
For those interested, here is complete set of my GPU benchmarks (frametimes included). Everything was running on the mentioned 2500k OC. As the results suggest, not only 2900 XT CF is more or less useless, but also HD 4870 CF, GF6 SLI, GF7 SLI and GF8/9 SLI... although these are slightly better. 😵
http://www.hw-museum.cz/data/Prubehy_fps_benchmark_VGA3.rar

Maybe modern GPUs and DX11 games are better in this matter (no personal experience), although I wouldn't be very optimistic https://www.pcper.com/reviews/Graphics-Cards/ … nd-Theft-Auto-V

And there's your problem. I bet your P67 board has one 16x slot (16x electrically) and a second 16x slot that is wired for 4x, it will cause stutternign because the 2900XT is a rather early PCI-E card, and won't take advantage of fancy things like PCI-E 2.0 - witch means the 4x slot is most likely starving the second video card. I'll bet my left nut on it. Even if the card was running at PCI-E 2.0 speeds (witch it's not, it only supports PCI-E 1.0) there would be a slightly noticeable latency because of the different slot speeds. If you could somehow set both slots to 4x the stuttering would go away.

Is this your board? http://www.gigabyte.com.ro/products/product-p … spx?pid=3882#sp - if so, the second slot is indeed 4x.

Now I tested a set of Radeon 4870 cards in crossfire a couple of months ago, and they were really smooth on my P35 based Asus P5kWS-64, but when I stuck them in my AMD 760 based biostar (1x16x and 1x4x) to use with a faster Phenom II CPU, I noticed stuttering not present on the intel system - even tough overall framerates were higher... Not only different PCI-E bus speeds make a difference, but different cards from different vendors will.

...

kanecvr wrote:

And there's your problem. I bet your P67 board has one 16x slot (16x electrically) and a second 16x slot that is wired for 4x, it will cause stutternign because the 2900XT is a rather early PCI-E card, and won't take advantage of fancy things like PCI-E 2.0 - witch means the 4x slot is most likely starving the second video card. I'll bet my left nut on it. Even if the card was running at PCI-E 2.0 speeds (witch it's not, it only supports PCI-E 1.0) there would be a slightly noticeable latency because of the different slot speeds. If you could somehow set both slots to 4x the stuttering would go away.

Is this your board? http://www.gigabyte.com.ro/products/product-p … spx?pid=3882#sp - if so, the second slot is indeed 4x.

Now I tested a set of Radeon 4870 cards in crossfire a couple of months ago, and they were really smooth on my P35 based Asus P5kWS-64, but when I stuck them in my AMD 760 based biostar (1x16x and 1x4x) to use with a faster Phenom II CPU, I noticed stuttering not present on the intel system - even tough overall framerates were higher... Not only different PCI-E bus speeds make a difference, but different cards from different vendors will.

Then you bet wrong 😁 This is my board http://www.gigabyte.com/products/product-page … spx?pid=3759#ov so of course it supports SLI and CF 8+8.

I can try 2900 XT again on SB-E as CF 16+16... but I seriously doubt it could solve such extreme stuttering.

I think when you guys say you are seeing stuttering, you are seeing something really bad. AFR simply not working properly for some reason.

But in general AFR causes latency that is unavoidable in all situations. Even if you think you aren't seeing it. This simply means that the frame rate it provides is not as fluid as if a single GPU were rendering alone. You will need more AFR frame rate for the same fluidity as a single GPU setup at the same frame rate.

From what I understand, it is part of the reason for the disappearance of 3-4 way SLI. Besides efficiency being horrible, the latency is even worse as well. It was only useful for benchmarking.

Well I have yet to see this "extreme" stuttering you speak of except for some rare cases. I gamed for 1.5 years on my 2900xt CF setup and was extremely happy with it - even more so with my R9 280X setup - and this coming from a person who spent a fortune on a GTX 1070 because I consider anything under 55 fps unplayable - especially large variations in framerate.

* I did notice tearing - sometimes considerable tearing in my SLi and CF setups - but no stuttering (ever) unless the CPU was underpowered or I was using mismatched PCI-E slots. Tearing can be fixed by fiddling with nvidia control panel / catalyst control center options.

In the process of doing a couple of sli builds though very behind at this time but I once had a XFX 780i and it was a royal pain in the ass to "get right". Nforce4, 680i, and 780i boards are ok but you have to sadly earn it or you will get a lot grief especially with the micro shutter thanks to the shit pci-e latency issues with boards equipped the infamous N200 pci-e bridge.

Nforce4 (Intel) extreme heat may as well cook meat but pretty solid after cooling is upgraded (had to drill a large all copper socket A Thermalright cooler to fix the temps)
680i, great overclocking except the cooling is still very poor and neglect kills boards
780i, basically a 680i with a N200 bridge so thermal hell but potentially amazing overclocking if you can replace the cooler altogether preferably with a water block
790i, the best all around SLI chipset for 775 except they are costly, hard to find, and wish that I had one

If any of you have a few copies of "hyper sli" which is a registry hack for older drivers you can get SLI working on a wider verity of boards with AMD and Intel chipsets. There is a hack for modern drivers but haven't tested it.

..