ATI HD 4870 X2 QuadCF Unwrapped @ Rage3D

[Lupine]

Santa delivers us a pair of ATI HD 4870 X2 bound together with sugar and spice and nearly everything nice. Join us as we peel back the wrapping and take a look at some rather toasty topics including microstuttering, GPGPU and the new Catalyst 8.12 driver set in our final review of 2008!

With some help from a lazy reindeer, we've broken into Santa's workshop for a gander at his new HD 4870 X2 Quad Crossfire equipped sleigh. Will the twin turbo sled render the reindeer obsolete, or will the Stutter Grinch bring the Christmas Eve flight crashing back to earth?

ATI HD 4870 X2 QuadCF Unwrapped

[Sound_Card]

As always enjoy reading Alex's material.

[caveman-jim]

Hybrid SLI or Hybrid Crossfire are very poor as performance enhancers

Hybrid AFR is a performance solution for low end GPU's such as integrated chipsets to be used with single high-end cards, to provide high performance on demand with reduced energy and heat costs for normal use. I don't believe I have seen them marketed as more than that, though I could be wrong.

[shadow001]

I read the article and it was quite informative,especially since i am a quad crossfire user(pair of HD 4870 X2's),and generally satisfied with the performance i get overall,but here's a question that may completely change the performance characteristics of Multi GPU setups.


There's a company out there called lucid,and they built a sort of intelligent PCI-e bridge chip that can distribute graphics workloads between multiple GPU's in ways that far exceed standard AFR/SFR or tiling techniques,and it's called the Hydra chip.


As it turns out,it seems that it performs so well,that even Intel is an investor in the company,and that they plan on adding it to their next revision of the Core i7 smackover motherboard,stated right here in fact.


http://www.theinquirer.net/inquirer/...d-hydra-on-x58


And here's a link to give a better idea on how it works.


http://www.theinquirer.net/inquirer/...ulti-gpus-easy


As good as the perfomance has been overall,i think there's a lot of speed still not being exploited in this setup,especially since it's still managed thru software,but what if a moherboard has this onboard,and it's this doing the workload distribuition,could the performance scale much better?

[train2win]

Great article, there is a big discussion in the drivers section regarding the microstutter. I was under the impression that certain settings in fc2 went over the 512mb frame buffer thus causing the stutter.

Shadow1, I want to see the benchmarks of that Lucid chip before passing judgement. It would be interesting to see if they could increase performance with that card.

[shadow001]

Originally Posted by train2win

Great article, there is a big discussion in the drivers section regarding the microstutter. I was under the impression that certain settings in fc2 went over the 512mb frame buffer thus causing the stutter.

Shadow1, I want to see the benchmarks of that Lucid chip before passing judgement. It would be interesting to see if they could increase performance with that card.

You're not the only one,since i know that ATI already went on statement that if they can get to about 2.5x the performance of a single GPU with Quad crossfire thru constant driver improvements,they're be pretty pleased with the results overall....


But what if,and i know this may be a stretch,that this lucid chip may allow to get overall performance much closer to 4x the performance of a single GPU,since there are 4 GPU's in a quad crossfire setup afterall,the performance increase would be substancial to say the least,and you'd get to see what all 4 GPU's can really do when working together....

[train2win]

I just can't see how they could do it if the two big elephants in the room couldn't do it. Nv and ATI just do graphics...I'd figure they would have found a way first but I could be wrong. What I don't get is why some games get it while others don't. There are certain games that run great on my rig and a few that have microstutters if I crank the up the settings. Is it a coding thing?

[shadow001]

Originally Posted by train2win

I just can't see how they could do it if the two big elephants in the room couldn't do it. Nv and ATI just do graphics...I'd figure they would have found a way first but I could be wrong. What I don't get is why some games get it while others don't. There are certain games that run great on my rig and a few that have microstutters if I crank the up the settings. Is it a coding thing?

Well if you read the article i linked to,you'll see that the way it works goes beyond AFR/SFR or tiling in terms of distributing the work between the GPU's,the chip seems to measure the amount of time needed for a particular amount of work for each GPU and constantly readjusts that to make the render times even between all the GPU's present.


You'll see a picture there that even distributes particular objects to render to a given GPU,and constantly modifies the workload to each GPU in real time depending on what's being rendered...It basically goes beyond alternate frme rendering,split frame rendering or tiling,wich has to be managed by the CPU to make matters worse.


It might even help with render to texture operations,wich have traditionally never scaled at all with SLI or crossfire,since with the current way of doing things,both GPU's(or more),are forced to do the same work under those types of operations,not to mention that in can also reduce CPU driver overhead since the CPU doesn't have to manage it anymore...It's dedicated hardware.

[Blín D'ñero]

Very well compiled and written article, thanks!
I missed the resolution used in each game paragraph (then found one on the last page), and the monitor used in the system setup.
But again thanks for the article. Gaming, with my single 4870X2 i'm still sticking to my 1600x1200 Eizo without regret. Only in Photoshop i wish i had a 26"widescreen.

[AlexV]

Originally Posted by BlindNero

I missed the resolution used in each game paragraph (then found one on the last page), and the monitor used in the system setup.

The table from the last page makes reference only to the settings used for the 8.11 vs 8.12 graphs exactly above. For the bulk of the article, all tests were done at 1920x1200.

Oh, and Merry Christmas everyone!

[VW_Factor]

Originally Posted by caveman-jim

Hybrid AFR is a performance solution for low end GPU's such as integrated chipsets to be used with single high-end cards, to provide high performance on demand with reduced energy and heat costs for normal use. I don't believe I have seen them marketed as more than that, though I could be wrong.

You are correct. It wasnt exactly meant to increase gaming performance, more than its capability to reduce power consumption. Which even today runs as a concern for some enthusiasts.

[acroig]

Originally Posted by AlexV

Oh, and Merry Christmas everyone!

Likewise and a prosperous 2009 to you and yours.

[caveman-jim]

When reading the following:

let's consider how normal, sequential, single-GPU rendering works:

*

the GPU takes its time to render the frame - we'll call this time/frame from now on
*

the GPU presents the rendered frame
*

the cycle repeats itself

This is obviously simplified since we're ignoring page flipping, aka double buffering, and its impact. However, the main idea should be rather clear: rendering is work-present-work-present in a constant cycle...
...Any attentive reader will notice that adding coarse frame-level parallelism introduces the risk that micro-stuttering will exist, since you get (at least) one extra frame ready and available within the same time/frame interval but can't ensure that this rhythm is maintained. Why? Because the (n+1)th frame (n=number of GPUs) will require more time/frame due to the reasons described above.

I found it easier to think of 'time/frame' as 'frame render time', and 'present' as 'display rendered frame'. It helped me understand the point made more easily.

[SIrPauly]

I'm reading this so far and it's great to see someone raise micro-stuttering and attempting to discuss it and explain it. Thank you!

[overcast]

Seriously considering getting rid of my second 4870x2, even at 2500x1600 , it's just not worth it.

[Lupine]

Originally Posted by overcast

Seriously considering getting rid of my second 4870x2, even at 2500x1600 , it's just not worth it.

What issues are clouding your experience?

[overcast]

Originally Posted by Lupine

What issues are clouding your experience?

Biggest issues are especially in Far Cry 2 right now. Quad I7 3.2GHZ and Quadfire 4870X2 and 6GB RAM should be pretty pleasureable experience even at the highest settings. Constant slowdowns , hitches are driving me nuts. As far as the rest of the games are concerned, a single 4870X2 will max everything out anyhow. Left 4 Dead, Half Life 2, Dead Space. The scaling with another card is just pretty poor, clearly seen in these benchmarks and many others.

[shadow001]

Originally Posted by overcast

Biggest issues are especially in Far Cry 2 right now. Quad I7 3.2GHZ and Quadfire 4870X2 and 6GB RAM should be pretty pleasureable experience even at the highest settings. Constant slowdowns , hitches are driving me nuts. As far as the rest of the games are concerned, a single 4870X2 will max everything out anyhow. Left 4 Dead, Half Life 2, Dead Space. The scaling with another card is just pretty poor, clearly seen in these benchmarks and many others.

Well it is a very powerfull setup overall,and hard to leverage all that speed on a consistent basis,as it's a given that some games are going to be more CPU demanding than others,while some,due to the way they're coded,just don't scale well with multiGPU setups while using software to distribute the workload between each GPU....It has it's limits.


I'm in the same boat here as well,but i'm still satisfied with the overall performance on my penryn running at 3.7 Ghz,and before i hop on a new motherboard,i want to see those new motherboards that will use the Lucid Hydra chip,and see if a hardware solution further helps with scaling on multiple GPU's,than simply relying on drivers.


Seems Intel will have it on their next revision of their smackover board btw...

[overcast]

Originally Posted by shadow001

Well it is a very powerfull setup overall,and hard to leverage all that speed on a consistent basis,as it's a given that some games are going to be more CPU demanding than others,while some,due to the way they're coded,just don't scale well with multiGPU setups while using software to distribute the workload between each GPU....It has it's limits.


I'm in the same boat here as well,but i'm still satisfied with the overall performance on my penryn running at 3.7 Ghz,and before i hop on a new motherboard,i want to see those new motherboards that will use the Lucid Hydra chip,and see if a hardware solution further helps with scaling on multiple GPU's,than simply relying on drivers.


Seems Intel will have it on their next revision of their smackover board btw...

Yeh I'm definitely keeping my eye on that Hydra Chip. Though it almost sounds too good to be true. If they achieve ANY type of linear scaling it will be monumental , even if it is only 30% per GPU added. Anything is better than what we have now. Hopefully this means moving from away from subjective to objective performance.

[shadow001]

Originally Posted by overcast

Yeh I'm definitely keeping my eye on that Hydra Chip. Though it almost sounds too good to be true. If they achieve ANY type of linear scaling it will be monumental , even if it is only 30% per GPU added. Anything is better than what we have now. Hopefully this means moving from away from subjective to objective performance.

Well,it does go beyond the usual AFR/SFR or Tiling schemes used with drivers to split the workload up between multiple GPU's,as it distributes and calculates the actual rendering time for a given workload for a specific frame to each and every GPU present in the system,and dynamically changes that workload on a frame by frame basis as well,until the rendering times between each GPU are the same for every frame needed to be rendered,and that's something no driver will ever do....



So the queston to ask is,how much better can it make the cards perform over using a software solution,and if the extra boost in performance is worth the added cost of adding the Hydra chip on the motherboard,making the necessary changes to the board itself,and make sure that there's no compatibility/stability issues to deal with.


They're even gone on the record that you can mix different cards from the same brand,though i'm pretty certain it would have to be from the same Direct X generation,and that the Hydra chip has the ability to vary the workload by making the faster card do more work overall,but the slower card still contributes to increase performance no matter what,so it may allow some pretty wierd combo's to work together...


Would be wicked to try my pair of HD4870 X2's,with an older HD 3870 X2 i have laying around,wich is currently just collecting dust.....6 GPU mayhem anyone,

[Treeckcold57]

It would be fun if how about try to mix the 4870X2 with 3870X2. An 3870X2 should do some physics while 4870X2 rendering the graphic.

[shadow001]

Originally Posted by Treeckcold57

It would be fun if how about try to mix the 4870X2 with 3870X2. An 3870X2 should do some physics while 4870X2 rendering the graphic.

Is there any physics support planned with future catalyst driver releases,similar to what Nvidia already does?.

[Treeckcold57]

i think AMD says Havok is dead until DX11 that would come with physics API. But Lucid company said that you can get an ATI 4800 series card with any nvidia card that support physics through lucid hardware.

[shadow001]

Originally Posted by Treeckcold57

i think AMD says Havok is dead until DX11 that would come with physics API. But Lucid company said that you can get an ATI 4800 series card with any nvidia card that support physics through lucid hardware.

True,i heard that as well,but since i already have that HD 3870 X2 lying around,i'd rather use that for Physics instead of buying yet another card,not to mention having to deal with potential issues with Vista by having both ATI and Nvidia drivers installed in the same system,which can be it's own set of headaches right there....


I thought that ATI had made a deal with Nvidia,regarding ageia Physics support in future drivers,as well as Intel with Havok,that way they'd be covered no matter wich one came out the winner.....Assuming it actually takes off in the first place of course.