Support NeoGAF

[Slavik81]

Unity is seems really behind the times IMO. That engine is basically perfect for Vulkan, why not?

Then again, this is the engine that does not even support Fullscreen modes at runtime..

They explained how difficult it is to transition to the DX12/Vulkan way of doing things if you're already heavily dependant on the DX11 way of doing things. They've done DX12, but it's not really been any improvement yet because they're not structured to take advantage of it.

I wouldn't worry about a lack of Unity support in Vulkan for a little while at least. A lot of their work on DX12 support should help if/when they add Vulkan support.

 

[Greenzxy]

Still waiting for Nvidia talks to come online.

 

[show me your skeleton]

SIGGRAPH expands out to Special Interest Group GRAPHics. There's a SIG for a lot of fields in CS. It's organized by the ACM.

ah awesome, cheers!

 

[vpance]

This new? Killing Monsters: Behind the Scenes of the Witcher 3: Wild Hunt

Japanese article but there's a bunch of screens and clips at the link.

 

[blu]

Not a scientific presentation, but still highly relevant and interesting: a Vulkan / OpenGL ES / OpenGL BOF video is up.

I care mostly about the Vulkan part, which starts here: https://youtu.be/faYDPjI2zhU?t=1h11m20s
In the big picture, they confirmed a 2015 release, and that conformance testing is in great shape, which should be music to the ears of everyone who ever fought different OpenGL driver's "interpretation" of the API.

I don't think conformance tests (or absence thereof) have been the main issue with the state of GL(ES) compliance. When major vendors' stacks have "specialized support" for specific titles you know you have your API compliance on the background, and other priorities on the foreground. We can only hope Vulkan to be somewhat "immune" to those pitfalls by virtue of being a much leaner API and not having much of a state.

On an unrelated note, why is it that respectful organisations cannot produce adequate videos of their conference talks? The slides on this one are readable only in the 720p stream version, which makes it more or less unwatchable on mobiles. Rule of thumb when doing talks with slides: picture-in-picture with slides on the background, speaker on the small foreground. When you're unable to do that, at the very least place the camera face-on with the projection plane.

 

[LCGeek]

I personally also found some of the specifics about windowing system integration interesting, but that was probably not too exciting for most. The Why Vulkan is great part of the presentation should be easily digestible though.

Also, there was confirmation that Vulkan development will be fully integrated and supported in the Android NDK, which was to be expected after the initial Android announcement but is still good to hear.

This has me shit grinning. If vulkan is good and certain devices keep getting more powerful in a few years handheld stuff will be very appealing to me again.

Thanks for the link my head is mush and don't need too technical stuff right now.

 

[Chev]

On an unrelated note, why is it that respectful organisations cannot produce adequate videos of their conference talks?

They do as long as you have two hundred bucks to spare. So naturally it's not in their best interest to just give those to you for free.

 

[JonathanNemo]

They do as long as you have two hundred bucks to spare. So naturally it's not in their best interest to just give those to you for free.

I'm pretty sure Khronos' BoF sessions aren't on the Encore DVD. Before we get all rah rah established organizations are evil and bad.

And I'm speaking as someone that bought the Encore package for the on-demand archive of all past (digitally archived) SIGGRAPH sessions. It's pretty groovy.


Honestly the audio levels in the room could have been better, and the video quality had its problems. I could probably bring it up and ask what could be done in the future. Much of that part of the conference was put together and run by Khronos Group volunteers, very friendly folks at that.

Come to think of it, there were audio problems in other parts of SIGGRAPH, but I think those were handled by different folks. Mostly around the bigger stages in the west hall of LA Convention Center.


(and if I hadn't made it already clear, Khronos Group != ACM SIGGRAPH. They different people, folks.)

 

[Deviousx]

Still waiting on those Global illumination papers on Quantum Break. I want to know more about that amazing lighting. :/

 

[Dictator93]

This new? Killing Monsters: Behind the Scenes of the Witcher 3: Wild Hunt

Japanese article but there's a bunch of screens and clips at the link.

Yeah that is new. I hope they release the slides and videos! Just added this link to the OP though!

 

[didevol]

Yeah that is new. I hope they release the slides and videos! Just added this link to the OP though!

It was a talk at siggraph, i have notes on it. I'm exhausted, but going through the notes tonight. Ill post them up if you guys want.

 

[tuxfool]

It was a talk at siggraph, i have notes on it. I'm exhausted, but going through the notes tonight. Ill post them up if you guys want.

Yes, please.

 

[Durante]

I don't think conformance tests (or absence thereof) have been the main issue with the state of GL(ES) compliance. When major vendors' stacks have "specialized support" for specific titles you know you have your API compliance on the background, and other priorities on the foreground. We can only hope Vulkan to be somewhat "immune" to those pitfalls by virtue of being a much leaner API and not having much of a state.

Having a baseline of fully tested conformant behaviour you can depend on is certainly a significant improvement. Of course it doesn't solve every problem by itself, but together with other changes which should result in a simplifications of the driver implementation and fewer opportunities for error (e.g. SPIR-V input instead of text) I expect good things.

 

[kyser73]

So I understood about 1/3 of the MM presentation (and that largely thanks to GFX threads on gaf making me go an look stuff up), and now I can't wait for PGW to actually see Dreams in operation.

Also blew my mind that the stuff they used in the PS4 reveal (the puppet band) is actually using a renderer that has since been discarded! Amazing R&D work! Was also impressed by how generally humble Alex' presentation was, especially the stop press comment about DICE.

 

[KKRT00]

Btw anyone was on Siggraph and was able to see this?
http://crytek.com/news/can-it-run-v...in-forces-to-create-virtual-reality-benchmark

-------
Really interested in this presentation
http://delivery.acm.org/10.1145/278..._=1439710656_3002964ceedbee94fab26d37b1c2252e

----
Interesting bits from abstract page of new Tomb Raider:
http://delivery.acm.org/10.1145/280..._=1439711050_1c629277a6180dc061449a8e24b8bc49

- Volumetric lighting is voxel based and is using async compute
- snow deformation tech is agnostic to the mesh and is using compute shader and tessellation
- they have their own AO technique that its better than HBAO

Ps. New Deus Ex is also using async compute.

===

Really big and detailed presentation about:
Real-time Many-Lights Management and Shadows with Clustered Shading
Its about Just Cause 3 lighting, new shadow mapping technique and clustered shading used in practice.
http://delivery.acm.org/10.1145/280..._=1439711522_55671f14ead3a1255b5eac3624f5d64c

 

[Kezen]

Btw any one was on Siggraph and was able to this?
http://crytek.com/news/can-it-run-v...in-forces-to-create-virtual-reality-benchmark

-------
Really interested in this presentation
http://delivery.acm.org/10.1145/278..._=1439710656_3002964ceedbee94fab26d37b1c2252e

----
Interesting bits from abstract page of new Tomb Raider:
http://delivery.acm.org/10.1145/280..._=1439711050_1c629277a6180dc061449a8e24b8bc49

- Volumetric lighting is voxel based and is using async compute
- snow deformation tech is agnostic to the mesh and is using compute shader and tessellation
- they have their own AO technique that its better than HBAO

Ps. New Deus Ex is also using async compute.

===

Really big and detailed presentation about:
Real-time Many-Lights Management and Shadows with Clustered Shading
Its about Just Cause 3 lighting, new shadow mapping technique and clustered shading used in practice.
http://delivery.acm.org/10.1145/280..._=1439711522_55671f14ead3a1255b5eac3624f5d64c

I can't access any of your links.
EDIT : fixed now.

It would be great if the PC version could ship with a DX12 renderer in order to use async compute. I assume the technical features are still possible with dx11 but performance is left on the table.

 

[Javin98]

Still waiting for Guerrilla's presentation on the volumetric cloud system in Horizon. Or is it already out?

 

[KKRT00]

Still waiting for Guerrilla's presentation on the volumetric cloud system in Horizon. Or is it already out?

Are we sure it will be on siggraph? I have not seen anything related to that on paper list. I could miss it though.

 

[dr_rus]

I can't access any of your links.
EDIT : fixed now.

It would be great if the PC version could ship with a DX12 renderer in order to use async compute. I assume the technical features are still possible with dx11 but performance is left on the table.

I think you have to link to pages and not PDFs themselves:
Rendering the World of Mirror’s EdgeTM
Labs R&D: Rendering Techniques in Rise of the Tomb Raider
Real-time many-light management and shadows with clustered shading

A couple more of interest:
Labs R&D: rendering Deus Ex: mankind divided
Accumulative anti-aliasing
Building the city of glass in mirror's edge™
Doing R&D for open worlds

 

[Javin98]

Are we sure it will be on siggraph? I have not seen anything related to that on paper list. I could miss it though.

Yeah, it was definitely on the list. Just look it up on the website.

 

[Kezen]

I think you have to link to pages and not PDFs themselves:
Rendering the World of Mirror’s EdgeTM
Labs R&D: Rendering Techniques in Rise of the Tomb Raider
Real-time many-light management and shadows with clustered shading

Thanks for that. Any idea when will the slides will be made available ?

 

[ashecitism]

Nvidia talks are up.

 

[Durante]

Accumulative anti-aliasing

Oh! That's actually very straightforward and smart.

Too bad most engines do deferred shading these days so it's not too widely applicable.

 

[charsace]

Btw any one was on Siggraph and was able to this?
http://crytek.com/news/can-it-run-v...in-forces-to-create-virtual-reality-benchmark

-------
Really interested in this presentation
http://delivery.acm.org/10.1145/278..._=1439710656_3002964ceedbee94fab26d37b1c2252e

----
Interesting bits from abstract page of new Tomb Raider:
http://delivery.acm.org/10.1145/280..._=1439711050_1c629277a6180dc061449a8e24b8bc49

- Volumetric lighting is voxel based and is using async compute
- snow deformation tech is agnostic to the mesh and is using compute shader and tessellation
- they have their own AO technique that its better than HBAO

Ps. New Deus Ex is also using async compute.

===

Really big and detailed presentation about:
Real-time Many-Lights Management and Shadows with Clustered Shading
Its about Just Cause 3 lighting, new shadow mapping technique and clustered shading used in practice.
http://delivery.acm.org/10.1145/280..._=1439711522_55671f14ead3a1255b5eac3624f5d64c

Not surprised that CD is pushing tech. They have been amazing tech wise since Soul Reaver.

 

[Noobcraft]

Oh! That's actually very straightforward and smart.

Too bad most engines do deferred shading these days so it's not too widely applicable.

Perhaps Forza Horizon 3 could benefit from it. The no AA through transparent objects would be a problem though.

 

[Durante]

Nvidia talks are up.

Just watched "Vulkan on NVIDIA GPUs". Very interesting that they allow mixed OpenGL/Vulkan applications to ease the transition. It's a bit like a OpenGL equivalent to MS' 11on12 feature.

Also, it looks like Vulkan drivers from NV will be ready as soon as the spec is released later this year.

 

[JonathanNemo]

Btw any one was on Siggraph and was able to this?
http://crytek.com/news/can-it-run-v...in-forces-to-create-virtual-reality-benchmark

I didn't remember seeing that on the schedule, and checking the Augmented/Virtual Reality track for 2015, it doesn't seem like Crytek or Basemark had any talks, BoFs or panels on VR. No presence on the exhibition floor either.

Announcements like these tend to be press releases timed with the event, sometimes with private meetings with journalists. But they're not presented as, say, flashy keynotes with the CEO doing product demos before a live crowd.

 

[M1ch04c4n0]

Very interesting, as usual. To this day I still regret having a paper accepted there and no pushing more to get the resources to attend.

 

[JonathanNemo]

Just watched "Vulkan on NVIDIA GPUs". Very interesting that they allow mixed OpenGL/Vulkan applications to ease the transition. It's a bit like a OpenGL equivalent to MS' 11on12 feature.

Also, it looks like Vulkan drivers from NV will be ready as soon as the spec is released later this year.

During the BoFs, Nvidia indicated that their Vulkan driver will be available for download at https://developer.nvidia.com/opengl-driver

I'm not super happy that they're keeping their monolithic OpenGL driver side by side with their Vulkan driver, but it's probably a massive technical challenge at this point to rewrite the GL driver as a sequence of Vulkan commands and maintain all the bloat that they've accumulated to get their GL driver working with the best performance possible for various popular apps and games.

 

[Tripolygon]

I can't access any of your links.
EDIT : fixed now.

It would be great if the PC version could ship with a DX12 renderer in order to use async compute. I assume the technical features are still possible with dx11 but performance is left on the table.

Asynchronous Compute seems to be the theme of most these papers i've read from siggraph. Sony seemed to have bet big on this since they designed the PS4 to be very good at it.

 

[Kezen]

Asynchronous Compute seems to be the theme of most these papers i've read from siggraph. AMD seemed to have bet big on this since they designed the GCN to be very good at it.

Fixed.

AMD designed GCN, not Sony. This is AMD you should be thankful to, they have made async compute one of GCN's core pillars.

 

[tuxfool]

Asynchronous Compute seems to be the theme of most these papers i've read from siggraph. Sony seemed to have bet big on this since they designed the PS4 to be very good at it.

More like AMD bet big on it and Sony ran with it.

e: beaten.

 

[Foutrologue]

 

[Kezen]

overused gif

?

 

[vpance]

AMD designed GCN, not Sony. This is AMD you should be thankful to, they have made async compute one of GCN's core pillars.

More like AMD bet big on it and Sony ran with it.

Sony had non trivial input into the design and influenced GCN changes, with added ACE being one of them. This was confirmed by an AMD employee IIRC.

 

[Kezen]

Sony had non trivial input into the design and influenced GCN changes, with added ACE being one of them. This was confirmed by an AMD employee IIRC.

I can't find any proof of that. Any source of yours ?

That said they requested 8 ACES for a reason so it's rather clear they take compute seriously, just like AMD.

 

[tuxfool]

Sony had non trivial input into the design and influenced GCN changes, with added ACE being one of them. This was confirmed by an AMD employee IIRC.

The extra ACE was something AMD was considering anyway. The known feature Sony contributed was selective cache line replacement. AFAIK that is still in any GCN1.1+ gpu (not that it is all that useful for dx11 applications).

 

[vpance]

I can't find any proof of that. Any source of yours ?

That said they requested 8 ACES for a reason so it's rather clear they take compute seriously, just like AMD.

Found it. Guy works at AMD

Since Cerny mentioned it I'll comment on the volatile flag. I didn't know about it until he mentioned it, but looked it up and it really is new and driven by Sony. The extended ACE's are one of the custom features I'm referring to. Also, just because something exists in the GCN doc that was public for a bit doesn't mean it wasn't influenced by the console customers. That's one positive about working with consoles. Their engineers have specific requirements and really bang on the hardware and suggest new things that result in big or small improvements.

https://forum.beyond3d.com/threads/playstation-4-codename-orbis-technical-hardware-investigation-news-and-rumours.53602/page-117#post-1698936

 

[Kezen]

Found it. Guy works at AMD



https://forum.beyond3d.com/threads/playstation-4-codename-orbis-technical-hardware-investigation-news-and-rumours.53602/page-117#post-1698936

Interesting. I'll retract my previous comment then.

I hope Nvidia are going to do whatever they need not to let compute-heavy games run badly on their Kepler/Maxwell hardware.

 

[tuxfool]

Interesting. I'll retract my previous comment then.

I hope Nvidia are going to do whatever they need not to let compute-heavy games run badly on their Kepler/Maxwell hardware.

Maxwell is fine with compute. Kepler not so much. They still don't have anything equivalent to ACE in Maxwell but instead something that does something similar. Maybe they have fewer issues with scheduling on account that Nvidia warps are only 32 elements (vs 64 element wavefronts on AMD).

 

[Kezen]

Maxwell is fine with compute. Kepler not so much. They still don't have anything equivalent to ACE in Maxwell but instead something that does something similar.

I thought GCN 1.1 and above had significantly superior compute performance compared to Maxwell. They can handle 64 compute jobs can't they ? I believe Maxwell Gen2 can handle 32.

At least async compute is supported on Maxwell, Kepler owners will probably have to upgrade.

 

[tuxfool]

I thought GCN 1.1 and above had significantly superior compute performance compared to Maxwell. They can handle 64 compute jobs can't they ? I believe Maxwell Gen2 can handle 32.

At least async compute is supported on Maxwell, Kepler owners will probably have to upgrade.

They all support it. It is just that Kepler sucks at compute in general. The ACEs are there to increase utilization of hardware resources for compute and they do this in hardware. From what I understand in a Nvidia GPU this functionality isn't as automated.

Because AMD wavefronts are 64 elements wide, it means that they need 64 work elements in order to fully utilize hw. Because of this they don't have the scheduling granularity of a 32 element warp in Nvidia hw. The ACEs then help when scheduling work to be performed by the CUs. It should be noted that this is all speculation.

It is also known that AMD generally throws more shading power as a percentage of GPU die area. Just look at the compute monster (on non scientific and hpc workloads) that is Fiji. It hasn't helped them much so far, but given the right workloads they could capitalize on this.

 

[Durante]

Warp/wavefront widths and the number of submission queues in the frontend are basically entirely unrelated. They are literally at opposite ends of the compute pipeline.

 

[Kezen]

They all support it. It is just that Kepler sucks at compute in general. The ACEs are there to increase utilization of hardware resources for compute and they do this in hardware. From what I understand in a Nvidia GPU this functionality isn't as automated.

Because AMD wavefronts are 64 elements wide, it means that they need 64 work elements in order to fully utilize hw. Because of this they don't have the scheduling granularity of a 32 element warp in Nvidia hw. The ACEs then help when scheduling work to be performed by the CUs. It should be noted that this is all speculation.

It is also known that AMD generally throws more shading power as a percentage of GPU die area. Just look at the compute monster (on non scientific and hpc workloads) that is Fiji. It hasn't helped them much so far, but given the right workloads they could capitalize on this.

How do you expect Maxwell cards to fare compared to their GCN equivalents ? Everywhere I've looked GCN was heralded as the strongest arch for this kind of task which is only going to be more and more relevant.
Nvidia knew compute was going to be really big, hence why they created the CUDA platform, strange that they have not put more emphasis on compute with Kepler.
The 7970 may end up having as long legs as the 8800gt.

Nvidia will adapt, I have no doubt about that.

 

[nib95]

Found it. Guy works at AMD

https://forum.beyond3d.com/threads/playstation-4-codename-orbis-technical-hardware-investigation-news-and-rumours.53602/page-117#post-1698936

Read this before but it's nice to have the direct quote again. Sony certainly did well to push the compute side of things with the PS4. It's actually one area where the PS4 leads in a big way compared to the Xbox One.

 

[Durante]

I really wish people wouldn't just say "compute" when they could possibly mean any one of 50 different things. It reminds me of late last decade when every GPU was suddenly a "GPGPU".

It's the nxgamerization of hardware discussion


Edit:
I think the fact that "compute" aptly describes basically everything every component in a computer (which isn't memory) does is somewhat indicative of it being not very well suited to discussing specifics.

 

[nib95]

I really wish people wouldn't just say "compute" when they could possibly mean any one of 50 different things. It reminds me of late last decade when every GPU was suddenly a "GPGPU".

It's the nxgamerization of hardware discussion

When people talk about the additionally considered compute specs with the PS4, I'm guessing they're referring to the number of ACE's/queues.

 

[Kezen]

I really wish people wouldn't just say "compute" when they could possibly mean any one of 50 different things. It reminds me of late last decade when every GPU was suddenly a "GPGPU".

It's the nxgamerization of hardware discussion

Sorry, my knowledge on this particular section is rather limited.
Let's be more specific then, PC gaming does not exist in isolation. It stands to sense AAA (or more modest devs it does not really matter) multiplatform devs will build their engines around the GCN's strengthes (the GPU compute workloads at which it excels) so the question is : how will Kepler/Maxwell fare then ? I assume, perhaps wrongly, that those multiplatform games will make heavy use of GCN-tailored compute workloads and leverage async compute and async shaders.

We might have gotten a glimpse of an answer with Ryse which is a showcase for GCN, it did not run bad at all on Kepler/Maxwell but not quite as well as on various GCN cards.

 

[Boombox On]

Basically if you have Dx11 and DX12. Think of Metal as dx11.5. It keeps things like the binding model from older apis, but also has the separate command buffer and submission process from the newer apis.

This means it is easier to use than than Vulkan or Dx12, but it doesn't do things like bindless resources, which enable things like command buffer reuse.

Hopefully I have characterised it correctly, but that is my understanding from what I've read. Somebody stomp on me if this is wrong.

tl;dr: Metal is easier to use but not as flexible as dx12/Vulkan.

It also doesn't seem to be feature-complete, either.

It's unlikely as the game uses features in OpenGL that don't exist yet in Metal. The game requires Hard Tessellation, Geometry Shaders and a few other bits that aren't available in Metal but are required for the game to run even on lowest settings meaning this game can't use Metal.
We always try and use all technology available to make the game the best experience on the Mac depending on what the game requires and what technology is available.

Also, it looks like Vulkan BETA drivers from NV will be ready as soon as the spec is released later this year.

Clarifying.

 

[tuxfool]

Warp/wavefront widths and the number of submission queues in the frontend are basically entirely unrelated. They are literally at opposite ends of the compute pipeline.

The numbers ACEs are unrelated, but the ACEs and their respective queues there to schedule work to be performed on the shaders. The more queues you have the more easily you can schedule a task to be performed by the shaders because you can pick between more options to optimize hw resources.

I'm speculating that because a wavefront consumes more hardware upfront than a warp, scheduling optimal shader usage on AMD gpus is more difficult, which is a reason why they would have more queues.