Support NeoGAF

ScepticMatt · Mar 5, 2015

Road to VR live blogs:

AMD
http://www.roadtovr.com/amd-on-low-...vr-and-graphics-applications-live-blog-330pm/

LiquidVR SDK

-Latest Data Latch (latency reduction, better parallelism and performance)
-Affinity Multi-GPU (latency reduction, better CPU performance)
-Direct to Display
-"Racing the beam" during rolling shutter for 1ms motion-to-photon latency
-Working with Oculus/Valve

NVIDIA
http://www.roadtovr.com/vr-direct-h...mproving-the-vr-experience-live-blog-2pm-pst/

VR Direct

-Late Latching constants
-Driver-level asynchronous time warp, priority over "normal" rendering
-VR SLI (lower latency)
-"can't expect to see perfect 2x scaling here because it depends on the overlap of rendering between frames. If you see 40-50% increase in performance with SLI, you're doing well"
-lower latency GPU distortion shader, future compatibility with Oculus SDK
-"hot out of the oven", "will be in flux for awhile"

Valve
http://www.roadtovr.com/valve-talks-advanced-vr-rendering-live-blog-5pm-pst/

Advanced VR rendering

-HTC Vive specs (90Hz, 110 FoV, 1080x1200 per eye, room-scale tracking)
-instancing to double geometry setup (half API calls, improved cache coherency)
-DX11 Multi-GPU extensions: nearly doubled framerate on AMD, have yet to test NVIDIA implementation but will soon.
-Low persistence global display, panel only lit for 2ms after raster scan finished
-"running start" for lower latency and better parallelism
-"Aliasing is your enemy". 4xMSAA minimum, Valve use 8xMSAA. Jittered SSAA is the best
-New method for filtering normal maps.
-Roughness mip maps
-Tangent-space axis aligned anisotropic lightning, "It's so critical that you have good specular in VR", "We think AA specular in VR looks so good that we want to make sure everyone knows."
-Geometric specular anti-aliasing "hacky math"
-More efficient mip map encoding using a hemi-octahedron
-"Noise is your friend." Gradients are awful in VR and they pop, banding is horrible. Every pixel should have noise. Absolutely do this.
-Stencil out areas that you can't see through the lens. 15% performance improvement
-Warp mesh, only warp rendered area (15% performance cost reduction)

ashecitism · Mar 5, 2015

Oh, was just thinking of doing something like this. Thanks!

mhayze · Mar 5, 2015

ScepticMatt said:
Road to VR live blogs:
AMD
-"Racing the beam" during rolling shutter for 1ms motion-to-photon latency
NVIDIA
-Driver-level asynchronous time warp, priority over "normal" rendering
Valve
-HTC Vive specs (90Hz, 110 FoV, 1080x1200 per eye, room-scale tracking)
-instancing to double geometry setup (half API calls, improved cache coherency)
-Low persistence global display, panel only lit for 2ms after raster scan finished
-"running start" for lower latency and better parallelism
-Stencil out areas that you can't see through the lens. 15% performance improvement
-Warp mesh, only warp rendered area (15% performance cost reduction)

Interesting stuff across the board.

The resolution is one of the things that caught my attention, since that is not an even divisor of any common display resolution. But the last two items in bold make sense - due to the desired shape of the projected display and probably the optics, parts of the display are masked out - 15% as indicated. Chances are this may be a 2560x1440 display with some masking (i.e. a split 1280x1440 masked down to 1080x1200 - so about 100px all around per eye) or it could just be some new display resolution?

Sounds like AMD has a solid head start in this stuff, but I'm also very interested to see what Nvidia does here (at a driver/software level).

ashecitism · Mar 5, 2015

btw what's jittered SSAA like?

CTLance · Mar 5, 2015

Thanks for the thread.

mhayze said:
Sounds like AMD has a solid head start in this stuff, but I'm also very interested to see what Nvidia does here (at a driver/software level).

So it wasn't just my bias. Phew. Seems like team green is either keeping their cards incredibly close to their chests... or they got caught on the wrong foot with this VR thing. How odd.

ScepticMatt · Mar 5, 2015

ashecitism said:
btw what's jittered SSAA like?

see my older post on this: http://www.neogaf.com/forum/showpost.php?p=98325467&postcount=44

In essence, you randomize the sample positions over time, changing aliasing into "noise".

left: random (poisson), right: poisson disk sampling. Those examples are without SSAA, with it you increase the number of "dots".

No AA, 4x MSAA/SSAA (points are at the same location if they are not jittered)

The Llama · Mar 5, 2015

CTLance said:
Thanks for the thread.

So it wasn't just my bias. Phew. Seems like team green is either keeping their cards incredibly close to their chests... or they got caught on the wrong foot with this VR thing. How odd.

There have been rumors AMD has been ahead on VR for a while now. IIRC someone on here (who I forget) posted that they were working closely with Valve on this stuff and nVidia wasn't, or something like that.

Gumbie · Mar 5, 2015

Nice OP, thanks for all the info! It's interesting that AMD mentions they are already working with Oculus/Valve and Valve already has the Multi-GPU extensions working with AMD. Wonder why Nvidia seems to be behind in the VR game?

Sciz · Mar 5, 2015

-DX11 Multi-GPU extensions: nearly doubled framerate on AMD, have yet to test NVIDIA implementation but will soon.

-"can't expect to see perfect 2x scaling here because it depends on the overlap of rendering between frames. If you see 40-50% increase in performance with SLI, you're doing well"

You're doing it wrong, Nvidia.

ashecitism · Mar 5, 2015

ScepticMatt said:
see my older post on this: http://www.neogaf.com/forum/showpost.php?p=98325467&postcount=44

In essence, you randomize the sample positions over time, changing aliasing into "noise".

left: random (poisson), right: poisson disk sampling. Those examples are without SSAA, with it you increase the number of "dots".

No AA, 4x MSAA/SSAA (points are at the same location if they are not jittered)

Thanks!

Seiru · Mar 5, 2015

I'm going to be incredibly nervous if I end up having to buy an AMD card for my VR rig. I've been burned by them so many times in the past.

Durante · Mar 5, 2015

ScepticMatt said:
Road to VR live blogs:

AMD
[...]
NVIDIA

Awesome. Never expected things to progress so quickly on the vendor side when I backed the Oculus KS back in 2012.
Also, I wouldn't really interpret too much into any perceived performance difference at this point. What's important is that both are actively working on improving VR rendering.

ScepticMatt said:
-"Aliasing is your enemy". 4xMSAA minimum, Valve use 8xMSAA. Jittered SSAA is the best

Couldn't agree more, on DK1 and 2 HL2 without AA and with 8xSGSSAA is really a massively different experience.

Objectively Bad Opinion · Mar 5, 2015

Gumbie said:
Nice OP, thanks for all the info! It's interesting that AMD mentions they are already working with Oculus/Valve and Valve already has the Multi-GPU extensions working with AMD. Wonder why Nvidia seems to be behind in the VR game?

A little while ago I saw a post from an Oculus dev saying they hadn't heard a peep from Nvidia after their VR Direct unveiling last year, but that they had been working with AMD and were very pleased with the results.

Nvidia gonna Nvidia. Probably just playing their cards close to their chest as usual.

Though Crossfire clowning SLI so soundly seems off to me, I never thought of the two technologies as so different that a gap this wide would be possible. But it's early still, so.

Dr. Kaos · Mar 8, 2015

Good news. Ati seems to be ahead on this, but we'll see.

I do think that VR deserves its own hardware-level optimizations as well. If Ati/nVidia/intel have anything in the planning stages, they cannot announce it or some customers will hold off on buying a new GPU, which costs them sales.

Objectively Bad Opinion · Mar 8, 2015

Nvidia's slides (and speaker notes!) here: http://www.reedbeta.com/talks/

Durante · Mar 8, 2015

Objectively Bad Opinion said:
Nvidia's slides (and speaker notes!) here: http://www.reedbeta.com/talks/

Thanks!

VR DSR. I'll need two Titans.

syko de4d · Mar 8, 2015

Sounds like my future PC will run two GPUs

If VR really gets a tons of buyers AMD and Nvidia will sell so much GPUs. They are the real winners between the battle of Oculus and Valve

Durante · Mar 8, 2015

Finished reading the NV slides. I like that they already have high priority context support not just for DirectX but also for Android, I bt some people at Oculus wish Samsung used Tegra

Also, they seem to think that upsampling frame rate with async timewarp is a bad idea, though it's one of the Morpheus operating modes.

Interesting stuff about the dedicated copy engine for SLI. Also VR SLI drivers for DX11 are already available under NDA.

Felix Lighter · Mar 8, 2015

Do we know what machines they were using for the demos?

Objectively Bad Opinion · Mar 8, 2015

Durante said:
Finished reading the NV slides. I like that they already have high priority context support not just for DirectX but also for Android, I bt some people at Oculus wish Samsung used Tegra

Why use an entire context just for timewarp though. I don't understand.

Nvidia GPUs don't support asynchronous compute?

Bl@de · Mar 8, 2015

Any chance there will be a Youtube upload of these talks?

Zeyphersan · Mar 8, 2015

So I'm starting to think planning my build around a mITX board with room for only one GPU is a bad idea, since the world of VR is going to revolve around two?

jediyoshi · Mar 8, 2015

Is the Oculus SDK dropping DX9/DX10 something people didn't freak out about?

brain_stew · Mar 8, 2015

jediyoshi said:
Is the Oculus SDK dropping DX9/DX10 something people didn't freak out about?

Honestly, all VR development should be focused towards DX12 and Vulcan. DX9 and DX10 class hardware is not performant enough nor is the API.

viveks86 · Mar 8, 2015

Glad to see AMD/Nvidia stepping in to optimize for VR already! I guess I'm doing SLI after all. Thinking 980 SLI or 980 Ti SLI (depending on when that gets released and price) .

kami_sama · Mar 8, 2015

brain_stew said:
Honestly, all VR development should be focused towards DX12 and Vulcan. DX9 and DX10 class hardware is not performant enough nor is the API.

Also, I think they will drop DX11 soon enough. Lots of cards support already most if not all of DX12 features, I think.

ThoseDeafMutes · Mar 8, 2015

viveks86 said:
Glad to see AMD/Nvidia stepping in to optimize for VR already! I guess I'm doing SLI after all. Thinking 980 SLI or 980 Ti SLI (depending on when that gets released and price) .

There is no 980ti. Titan X is a month away, the 1000 series or whatever they rename it as is probably 5-8 months away

Prophet Steve · Mar 8, 2015

kami_sama said:
Also, I think they will drop DX11 soon enough. Lots of cards support already most if not all of DX12 features, I think.

Sounds like a weird thing to do. DX11 is still in active development. DX12 is for when programmers want to do more low-level programming and DX11 for the way we do things now which is "easier".

viveks86 · Mar 8, 2015

ThoseDeafMutes said:
There is no 980ti. Titan X is a month away, the 1000 series or whatever they rename it as is probably 5-8 months away

I meant the one that is rumored. It has been talked about since last year. Chances are it will get announced at GTC (unless their rumored production issues push it out to 2016). So I'm gonna wait till GTC to see if it is a real thing. If not, then 980 SLI it is. Pretty sure Titan X is gonna be ridiculously priced like all Titans, so that's not on the cards for me, unless it's price/performance ratio is in line why the 980s.

mrklaw · Mar 8, 2015

I want to see both AMD and nvidia start selling more dual GPU cards. Maybe market them as 'VR edition' or something like that. SLI is clearly of benefit for VR, but plenty of people will not want to buy two separate GPUs, and some simoly won't be able to use two GPUs if they have SFF builds with only single GPU motherboards.

Quaz51 · Mar 8, 2015

mhayze said:
a split 1280x1440 masked down to 1080x1200 - so about 100px all around per eye) or it could just be some new display resolution?

no it's 1080x1920 display masked down to 1080x1200

Xyber · Mar 8, 2015

viveks86 said:
I meant the one that is rumored. It has been talked about since last year. Chances are it will get announced at GTC (unless their rumored production issues push it out to 2016). So I'm gonna wait till GTC to see if it is a real thing. If not, then 980 SLI it is. Pretty sure Titan X is gonna be ridiculously priced like all Titans, so that's not on the cards for me, unless it's price/performance ratio is in line why the 980s.

The card people have been talking about will be on the full Maxwell chip, so it won't be a 980Ti since 980 is already using the full 204 chip and there's no extra stuff to enable. The card you are thinking about (unless you are actually thinking about the Pascal cards, but those are not coming for quite some time) will be the 1080 or whatever it will be called.

The Shogun · Mar 8, 2015

This is hella exciting as well as interesting stuff. VR is coming baby, and its great seeing so much focus on it with dev's happening at GDC. Hopefully this gets the creative juices flowing and we are going to see a big ramp up in terms of games in development for VR.

Also, one thing I noticed, its mentions that these are Vive dev kit specs. Nowhere does it say final consumer model.

Is it possible Valve is going to aim slightly higher in terms of FPS / resolution in the consumer model at the end of the year? I can't help but wonder if they will at least improve FPS since stuff like Morpheus now is 120fps.

Gumbie · Mar 8, 2015

I mostly buy nvidia cards but seeing how amd has been working closely with oculus and valve is making me consider the Radeon 300 series for my next card.

Tetranet · Mar 8, 2015

-"Noise is your friend." Gradients are awful in VR and they pop, banding is horrible. Every pixel should have noise. Absolutely do this.

Huh, fascinating. This can prove to be quite useful in masking such shortcomings.

brain_stew · Mar 8, 2015

Prophet Steve said:
Sounds like a weird thing to do. DX11 is still in active development. DX12 is for when programmers want to do more low-level programming and DX11 for the way we do things now which is "easier".

DX11 has no place in VR longterm. If you want to create a good VR experience then you need to be obsessive about latency, frame times and performance. DX12 and Vulcan are critical to the success of VR on the PC.

Prophet Steve · Mar 9, 2015

brain_stew said:
DX11 has no place in VR longterm. If you want to create a good VR experience then you need to be obsessive about latency, frame times and performance. DX12 and Vulcan are critical to the success of VR on the PC.

Ah, I understood it wrong, I thought he meant the development and support for DX11 altogether, not only for VR. Makes sense then.

viveks86 · Mar 9, 2015

Xyber said:
The card people have been talking about will be on the full Maxwell chip, so it won't be a 980Ti since 980 is already using the full 204 chip and there's no extra stuff to enable. The card you are thinking about (unless you are actually thinking about the Pascal cards, but those are not coming for quite some time) will be the 1080 or whatever it will be called.

I see… well, I'm basically waiting for some flagship card that's not a Titan to be announced at GTC. If there are none, I'm settling for a 980 SLI. Unless the Titan X costs $1000 AND is faster than 980 SLI (unlikely).

ashecitism · Mar 9, 2015

It's a safe bet that no other card will be announced at GTC, Jen-Hsun will just expand on the Titan X, and maybe talk again about Pascal.

Durante · Mar 9, 2015

There could be Maxwell Teslas

viveks86 · Mar 9, 2015

ashecitism said:
It's a safe bet that no other card will be announced at GTC, Jen-Hsun will just expand on the Titan X, and maybe talk again about Pascal.

Don't break my heart

gofreak · Mar 10, 2015

Full Valve slides: http://alex.vlachos.com/graphics/Alex_Vlachos_Advanced_VR_Rendering_GDC2015.pdf

raphier · Mar 10, 2015

gofreak said:
Full Valve slides: http://alex.vlachos.com/graphics/Alex_Vlachos_Advanced_VR_Rendering_GDC2015.pdf

has a couple of high quality pictures of Aperture Demo aka. Source 2.

gofreak · Mar 10, 2015

raphier said:
has a couple of high quality pictures of Aperture Demo aka. Source 2.

Well, higher quality I guess, but not native res:

Seanspeed · Mar 10, 2015

Its going to look different in VR anyway.

gofreak · Mar 10, 2015

Seanspeed said:
Its going to look different in VR anyway.

True. I'm not sure if the heavy-ish noise in the second one is just the image or the added noise for VR they mention.

Durante · Mar 10, 2015

This is a good one to quote regarding expected fidelity and HW requirements for VR.

KKRT00 · Mar 10, 2015

Makes wonder what res and framerate the Crytek demo was running on Titan X.
Source 2 shots look like something that is doable in those framerates and res, but this?

http://images.eurogamer.net/2013/articles//a/1/7/4/1/3/2/9/dino.jpeg.jpg

https://www.youtube.com/watch?v=m0Dt5x1ShvY

Durante · Mar 10, 2015

Support NeoGAF

VR Rendering summary for AMD/NVIDIA/Valve tech talks

Member

Member

Member

Member

Member

Member

Member

Member

Member

Member

Banned

Member

Member

Banned

Member

Member

Member

Member

Member

Member

Member

Banned

Member

Member

Member

Member

Member

Member

Member

MrArseFace

Member

Member

Banned

Member

Member

Member

Member

Member

Member

Member

Member

GAF's Bob Woodward

Banned

GAF's Bob Woodward

Banned

GAF's Bob Woodward

Member

Member

Member

Member

Similar threads