Overclock.net banner

[Tweaktown] AMD's dual GPU version of the Radeon R9 Fury X will be out this year

4K views 51 replies 30 participants last post by  F3ERS 2 ASH3S 
#1 ·
#4 ·
One fury x and a dual for tri fire would be the route I would be thinking.
 
#6 ·
Quote:
Originally Posted by Apexii22 View Post

A shame its only 4gb vram, unless dx12 lets each card stack, then tri-fire would be awesome.
DX12 does alloow for stacking vRAM, however it is up to the developer of the game to use that option.. it is not done automatically but programatically (sp) (I haven't had my coffe yet)
 
#7 ·
Why November prediction? Any word of cooling solution?
 
#8 ·
Quote:
Originally Posted by F3ERS 2 ASH3S View Post

DX12 does alloow for stacking vRAM, however it is up to the developer of the game to use that option.. it is not done automatically but programatically (sp) (I haven't had my coffe yet)
Technically it isn't "VRAM stacking" so much as it is how multiple video cards will draw a screen. In a two card system, one card will draw the odd lines, the other card will draw the even lines. Thus a card does not need to draw the entire screen image, only half of it, thus it will only need half as much memory.

It's a fine hair to split, but still, just wanted to put that out there.
 
#9 ·
Quote:
Originally Posted by 47 Knucklehead View Post

Technically it isn't "VRAM stacking" so much as it is how multiple video cards will draw a screen. In a two card system, one card will draw the odd lines, the other card will draw the even lines. Thus a card does not need to draw the entire screen image, only half of it, thus it will only need half as much memory.

It's a fine hair to split, but still, just wanted to put that out there.
From what you said .. half correct, yes it splits it, but what you explained was tradition as to how it has.. the other way is to actually split the screen differently, thus each GPU controls a different part of the screen itself and then can utilize the full VRAM for that GPU available.. as opposed to the interlacing rendering that requires the same information to be loaded on both GPUs to keep in sync, Thus removing the force of having to have the GPUs synced allows for more VRAM to be used.. maybe VRAM stacking isn't 100% accurate however the way that it is done does the same thing as long as it is coded appropriately.

Source
 
#11 ·
Quote:
Originally Posted by denman View Post

Dropping my GTX 690 for the Fury X2 unless Nvidia surprises me with a GTX 990 before then. Fury X2 and a new MG279Q will be nice together.
If the Ares Fury x2 comes out in a reasonable amount of time and doesn't cost 3 virgins and a golden goose, I'm grabbing it. So far all signs point to no Lightning Fury X, and that has me bummed.
 
#12 ·
Quote:
Originally Posted by F3ERS 2 ASH3S View Post

Quote:
Originally Posted by 47 Knucklehead View Post

Technically it isn't "VRAM stacking" so much as it is how multiple video cards will draw a screen. In a two card system, one card will draw the odd lines, the other card will draw the even lines. Thus a card does not need to draw the entire screen image, only half of it, thus it will only need half as much memory.

It's a fine hair to split, but still, just wanted to put that out there.
From what you said .. half correct, yes it splits it, but what you explained was tradition as to how it has.. the other way is to actually split the screen differently, thus each GPU controls a different part of the screen itself and then can utilize the full VRAM for that GPU available.. as opposed to the interlacing rendering that requires the same information to be loaded on both GPUs to keep in sync, Thus removing the force of having to have the GPUs synced allows for more VRAM to be used.. maybe VRAM stacking isn't 100% accurate however the way that it is done does the same thing as long as it is coded appropriately.

Source
That's not how SFR works. There's no drawing lines. In both cases of AFR and SFR whole screens are drawn. The difference between AFR and SFR is that in SFR the whole screen is split in proportionate halves between the available gpus. See Civ:BE for an example of SFR. And SFR btw is intended for tight frametimes and high minimum fps over high FPS so not all games will lend themselves to SFR.
 
#14 ·
Quote:
Originally Posted by Ganf View Post

If the Ares Fury x2 comes out in a reasonable amount of time and doesn't cost 3 virgins and a golden goose, I'm grabbing it. So far all signs point to no Lightning Fury X, and that has me bummed.
I'm just hoping EK comes out with a full block for the Fury X2 shortly after release. Have two 240mm Alphacool Monsta radiators for just my 4790k and the Fury X2.
 
#15 ·
Quote:
Originally Posted by tsm106 View Post

That's not how SFR works. There's no drawing lines. In both cases of AFR and SFR whole screens are drawn. The difference between AFR and SFR is that in SFR the whole screen is split in proportionate halves between the available gpus. See Civ:BE for an example of SFR. And SFR btw is intended for tight frametimes and high minimum fps over high FPS so not all games will lend themselves to SFR.
Thats what I said.. you are arguing what I said LOL
rolleyes.gif

Quote:
You: In a two card system, one card will draw the odd lines, the other card will draw the even lines. Thus a card does not need to draw the entire screen image
Me: half correct, yes it splits it, but what you explained was tradition as to how it has.. the other way is to actually split the screen differently,
So yeah you just took what I sid reworded it.. contradicted yourself then took what I said as the right way just making yourself look like you're right when I had actually clarified and corrected what you had said.. .bunch o lolz
rolleyes.gif
 
#16 ·
Quote:
Originally Posted by tsm106 View Post

And SFR btw is intended for tight frametimes and high minimum fps over high FPS so not all games will lend themselves to SFR.
SFR is actually older than AFR and was used long before frame interval was a real concern. The original 3DFX SLI (scanline interleave) is a type of SFR and many games used SFR during the first several years of NVIDIA's SLI and ATI's Crossfire.

SFR was almost completely supplanted by AFR because AFR scales better in raw frame rate and benchmarks. However, as you point out, AFR brings with it frame timing issues. AFR also depends on GPUs of near identical performance for good scaling and introduces latency due to the need for a large queue of pre-rendered frames.
 
#18 ·
Quote:
Originally Posted by F3ERS 2 ASH3S View Post

Quote:
Originally Posted by tsm106 View Post

That's not how SFR works. There's no drawing lines. In both cases of AFR and SFR whole screens are drawn. The difference between AFR and SFR is that in SFR the whole screen is split in proportionate halves between the available gpus. See Civ:BE for an example of SFR. And SFR btw is intended for tight frametimes and high minimum fps over high FPS so not all games will lend themselves to SFR.
Thats what I said.. you are arguing what I said LOL
rolleyes.gif

Quote:
You: In a two card system, one card will draw the odd lines, the other card will draw the even lines. Thus a card does not need to draw the entire screen image
Me: half correct, yes it splits it, but what you explained was tradition as to how it has.. the other way is to actually split the screen differently,
So yeah you just took what I sid reworded it.. contradicted yourself then took what I said as the right way just making yourself look like you're right when I had actually clarified and corrected what you had said.. .bunch o lolz
rolleyes.gif
lol wut? Where did you explain that in a simple manner? What is this force and full vram gibberish? Again there is no drawing lines odd or even lol. The whole screen is drawn in AFR and in SFR the screen is divided by the number of gpus.
doh.gif
 
#19 ·
Quote:
Originally Posted by tsm106 View Post

lol wut? Where did you explain that in a simple manner? What is this force and full vram gibberish? Again there is no drawing lines odd or even lol. The whole screen is drawn in AFR and in SFR the screen is divided by the number of gpus.
doh.gif
What I mant was splits the screen, with DX12 instead of having to load the same ram info to both cards to keep synced, it allows for each card to take a portion of the screen (not interlacing but sections IE top and bottom of the screen) and that is where the boost of the RAM comes from.. so it allows the VRAM to be utilized better as opposed to having to load redundant information to keep it synced.
 
#22 ·
Three main ways to do SFR that I am aware of:

- Scan-line interleave. Old 3DFX method where each card draws alternating scanlines. Thus if you were playing a game at 800x600, each card would render an 800x300 image, which would be interleaved.

- A literal split frame. The driver looks at each incoming frame, guesstimates where the entire frame would need to be split into two (or more) parts so that each part takes roughly the same time to render, then gives each part to a different GPU to render before compositing them all together and displaying the finished frame. Early implimentations seem to have split the frame at the same point regardless of what was being displayed, which tended to scale very poorly as it was likely each GPU would get sections that required very different times to render, making the others wait before a frame was ready.

- Tiling/supertiling. Each frame is split into many small segments called tiles and each GPU renders alternating tiles. Easier to balance loads this way.
Quote:
Originally Posted by F3ERS 2 ASH3S View Post

What I mant was splits the screen, with DX12 instead of having to load the same ram info to both cards to keep synced, it allows for each card to take a portion of the screen (not interlacing but sections IE top and bottom of the screen) and that is where the boost of the RAM comes from.. so it allows the VRAM to be utilized better as opposed to having to load redundant information to keep it synced.
SFR only reduces frame buffer size needed, all the assets needed to render each frame must still be mirrored, so VRAM savings tends to be minimal.

DX12's new method is not SFR, it's something far more.
 
#23 ·
Quote:
Originally Posted by Blameless View Post

Three main ways to do SFR that I am aware of:

- Scan-line interleave. Old 3DFX method where each card draws alternating scanlines. Thus if you were playing a game at 800x600, each card would render an 800x300 image, which would be interleaved.

- A literal split frame. The driver looks at each incoming frame, guesstimates where the entire frame would need to be split into two (or more) parts so that each part takes roughly the same time to render, then gives each part to a different GPU to render before compositing them all together and displaying the finished frame. Early implimentations seem to have split the frame at the same point regardless of what was being displayed, which tended to scale very poorly as it was likely each GPU would get sections that required very different times to render, making the others wait before a frame was ready.

- Tiling/supertiling. Each frame is split into many small segments called tiles and each GPU renders alternating tiles. Easier to balance loads this way.
SFR only reduces frame buffer size needed, all the assets needed to render each frame must still be mirrored, so VRAM savings tends to be minimal.

DX12's new method is not SFR, it's something far more.
I think the example that I used was what threw off what I was saying,, so poor example.. but that is why I tried correcting the example, so instead of splitting the whole frames to the gpus, now it splits the screen so the frames are smaller and each gpu renders a different area of the screen per frame.

in fact thinking about it more it was.. cause the DX11 frames one gpu renders one frame the other GPU renders the next or in a similar way,,

as where DX12 splits the actual render per frame if used

so now that the entire frame isn't needing to be rendered on both GPUs only sections it allows more ram to be utilized as there is less overhead from the redundant storage
 
#24 ·
Quote:
Originally Posted by Blameless View Post

Three main ways to do SFR that I am aware of:

- Scan-line interleave. Old 3DFX method where each card draws alternating scanlines. Thus if you were playing a game at 800x600, each card would render an 800x300 image, which would be interleaved.

- A literal split frame. The driver looks at each incoming frame, guesstimates where the entire frame would need to be split into two (or more) parts so that each part takes roughly the same time to render, then gives each part to a different GPU to render before compositing them all together and displaying the finished frame. Early implimentations seem to have split the frame at the same point regardless of what was being displayed, which tended to scale very poorly as it was likely each GPU would get sections that required very different times to render, making the others wait before a frame was ready.

- Tiling/supertiling. Each frame is split into many small segments called tiles and each GPU renders alternating tiles. Easier to balance loads this way.
SFR only reduces frame buffer size needed, all the assets needed to render each frame must still be mirrored, so VRAM savings tends to be minimal.

DX12's new method is not SFR, it's something far more.
http://blogs.msdn.com/b/directx/archive/2015/05/01/directx-12-multiadapter-lighting-up-dormant-silicon-and-making-it-work-for-you.aspx
 
#25 ·
Maybe they're waiting for dual-link 8-hi HBM to become a reality so they can put two 8GB cores on this card.
 
#26 ·
Quote:
Originally Posted by denman View Post

I'm just hoping EK comes out with a full block for the Fury X2 shortly after release. Have two 240mm Alphacool Monsta radiators for just my 4790k and the Fury X2.
If the Ares releases quickly after the X2's launch EK isn't too much of a worry. The 290x2 Ares was a nice full cover copper block with a single slot bracket already installed.

So much Smex.... So much premium price tag...
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top