[ Guru 3d ] AMD Teases PCs with Radeon Fury X2 Dual FIJI GPUs

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Feb 19, 2009
10,457
10
76
Your image seems do describe exactly what I'm saying. Each gpu renders one eye (aka 1080x1200, one half of a full frame), gpu2 sends its half to gpu1, the halves are combined, then the latest rotational data is polled from the headset, timewarp is applied, and the finished frame is sent to the headset.

I have seen nothing that indicates oculus/vive can operate how you are describing.

It's not half of a full frame.

Each GPU renders the entire frame, in the PoV of that eye.
 

Mahigan

Senior member
Aug 22, 2015
573
0
0
Your image seems do describe exactly what I'm saying. Each gpu renders one eye (aka 1080x1200, one half of a full frame), gpu2 sends its half to gpu1, the halves are combined, then the latest rotational data is polled from the headset, timewarp is applied, and the finished frame is sent to the headset.

I have seen nothing that indicates oculus/vive can operate how you are describing.
You're right, I just checked with a VR guy and the image is combined by the compositing engine found in all GCN GPUs. So there's no overhead unless you start to scale past 4 GPUs apparently.
 
Feb 19, 2009
10,457
10
76
You're right, I just checked with a VR guy and the image is combined by the compositing engine found in all GCN GPUs. So there's no overhead unless you start to scale past 4 GPUs apparently.

It's combined for the Async Timewarp step AFAIK if the user has moved their head, the frame needs to be updated on the fly before displaying, to reduce motion to photon latency.

All this is still paper, without major games to demonstrate the technique.
 

Mahigan

Senior member
Aug 22, 2015
573
0
0
It's combined for the Async Timewarp step AFAIK if the user has moved their head, the frame needs to be updated on the fly before displaying, to reduce motion to photon latency.

All this is still paper, without major games to demonstrate the technique.
I guess that's why those Falcon Northwest*Tiki Fury X2 machines are headed into the hands of VR developers.
 

pj-

Senior member
May 5, 2015
481
249
116
It's not half of a full frame.

Each GPU renders the entire frame, in the PoV of that eye.

Yes, but what is actually going out over the wire is 90 2160x1200 frames per second. Not 180 1080x1200 frames per second.
 

pj-

Senior member
May 5, 2015
481
249
116
It's combined for the Async Timewarp step AFAIK if the user has moved their head, the frame needs to be updated on the fly before displaying, to reduce motion to photon latency.

All this is still paper, without major games to demonstrate the technique.

There is a difference between asynchronous timewarp and timewarp. What this image shows is normal timewarp. The frame finishes on time and is warped to account for any rotational movement that's occurred since rendering began.

Asynchronous timewarp is used when the GPU doesn't finish the frame in time and instead re-uses the last frame with a reprojection based on updated rotation data from the HMD. Gear VR has async timewarp, but the PC HMDs do not as of now.

Timewarp has been a part of the oculus SDK for a long time and works as expected. Vive chose not to use timewarp as it allows them to draw only the circular area of pixels needed for display (it gives about 17% increased performance vs drawing the entire 1080x1200 region for each eye). Timewarp requires drawing the full frame so if the image needs to be moved, it's not pulling in black pixels.

I sometimes forget the difference and use the terms interchangeably even though they aren't quite the same.
 

Timmah!

Golden Member
Jul 24, 2010
1,418
630
136
The card is slowly turning to vaporware. Those PCB photos have been shown like half year ago.
 

maddie

Diamond Member
Jul 18, 2010
4,738
4,667
136
The card is slowly turning to vaporware. Those PCB photos have been shown like half year ago.
I think that Gemini [Fury X2] will be used as a booster for Polaris sales in the VR market. AMD is probably winding down 28nm production and this card would never have been a significant seller at best.

Consider it a technology demo for the use of dual GPU VR. A very high visibility marketing tool. Remember, Koduri claims that Polaris will allow the entry for VR to be a lot more affordable, suggesting the 380 price range instead of the 390. If true, then buyers might go for 2 cards instead of 1 after seeing multi-GPU VR in action.
 

Bacon1

Diamond Member
Feb 14, 2016
3,430
1,018
91
The card is slowly turning to vaporware. Those PCB photos have been shown like half year ago.

They've said they are waiting for Rift / Vive to launch as thats what they were designed for.

They just showed them off a few days ago.
 

Timmah!

Golden Member
Jul 24, 2010
1,418
630
136
I think that Gemini [Fury X2] will be used as a booster for Polaris sales in the VR market. AMD is probably winding down 28nm production and this card would never have been a significant seller at best.

Consider it a technology demo for the use of dual GPU VR. A very high visibility marketing tool. Remember, Koduri claims that Polaris will allow the entry for VR to be a lot more affordable, suggesting the 380 price range instead of the 390. If true, then buyers might go for 2 cards instead of 1 after seeing multi-GPU VR in action.

Are those Polaris 10/11 cards expected to be at least as fast as current FuryX? At the Radeon 390 price level? Why would anyone buy Fury X2 over 2 Polaris cards then? I mean except the miniscule number of customers (like me for example) who would want 2 cards but have space just for one... Or does this mean Fury X2 is going to be priced about the same as 2 390s?
 

maddie

Diamond Member
Jul 18, 2010
4,738
4,667
136
Are those Polaris 10/11 cards expected to be at least as fast as current FuryX? At the Radeon 390 price level? Why would anyone buy Fury X2 over 2 Polaris cards then? I mean except the miniscule number of customers (like me for example) who would want 2 cards but have space just for one... Or does this mean Fury X2 is going to be priced about the same as 2 390s?
Those are questions I'm sure many here would love to have answered.

Polaris 10, no way = Fury X

Polaris 11? We might know more next week. Journalists did say that the two versions they saw in January were tiny and large.

Fury X @ $ Radeon 390. ???? will happen eventually, hopefully soon.

At this point I don't believe AMD expects to sell a lot of Fury X2 cards, so they are trying to maximize the mindshare gains through them with VR
 

Timmah!

Golden Member
Jul 24, 2010
1,418
630
136
Yeah, thanks for response. We shall see next week. If only Nvidia would drop some more info about Pascal too.
 

swilli89

Golden Member
Mar 23, 2010
1,558
1,181
136
Would love to AMD come out and drop a bombshell: Fury X2 $699.

They'd have no problem at that price as it would be the undisputed king for a fairly long time at a great price. GP104 and Polaris 11 won't touch it.
 

crisium

Platinum Member
Aug 19, 2001
2,643
615
136
I have no doubt it will be at $699 or cheaper sometime this year after next gen launches: see 295x2 at $650 against the $550 GTX 980. I'd be surprised to see it that cheap at launch though. I'm just hopping for no more than $999, because at least that is still cheaper than 2x 980, let alone 980 Ti.
 

maddie

Diamond Member
Jul 18, 2010
4,738
4,667
136
Would love to AMD come out and drop a bombshell: Fury X2 $699.

They'd have no problem at that price as it would be the undisputed king for a fairly long time at a great price. GP104 and Polaris 11 won't touch it.
You know what? I still tend to think that Polaris 10 & 11 are GDDR5 & HBM2 versions and that we will have more than 2 GPU die sizes. Next week might reveal more info.

I don't see it making $ worthwhile for AMD to sell Fury X2 that cheap.