I find it shallow and pedantic.
Baseless Next Generation Rumors with no Technical Merits [pre E3 2019] *spawn*
- Thread starter Arkham night 2
- Start date
-
- Tags
- non sequitur
- Status
I find it shallow and pedantic.
Because the grey area between amd did everything and amd did nothing, seems to be lost in this forum.
It's semantics about the meaning of co-engineered versus co-designed. From a rumor which is complete bullshit anyway.
RTG is working on their RT solution so that's confirmed. When it arrives is a ?. Which models might be easier to narrow down. We know that RT is still compute heavy, so let's just use more of their documentation here:
So if we look at the flow chart there is a decision point on whether or not you are using triangles. Left is for non triangles, so say quads. You have to provide your own intersection shader for that. If it's triangles it's on the right - that goes straight to fixed function. You would also go left if you didn't have hardware acceleration for fixed function, then you're back to using a software intersection shader.
Looking around the flow chart, there's a lot of shading happening around here, i'm not sure how it compares to our current methods though. I would say to support RT you're going to need a good amount of TF power to support it properly whether you are accelerated or not. Nvidia went as low as the 2060 RTX for instance. So that should give you an idea of where AMD can start providing RT.
As for AMD and RT; my thoughts here is that AMD has been building out their RT solution for some time, this wasn't sprung on them. The only reason why Sony or MS will have hardware ray tracing is because AMD has been working on it. For a quick reference to the 2018 blog post:
So the collaboration with these partners have been in discussion well before the official announcement in 2018. The only reason I suspect that AMD is late to releasing their version of RT acceleration is because they are still working on what they feel is going to be providing their hardware the best performance.
Its much easier to do ray tracing if all your rays are coherent, but when they aren't and they are scattered it's troublesome, so scheduling is a big one that AMD might be trying harder to tackle (or will continually be a place for improvement in future RT generations). The ray tracing acceleration structure is another, they don't have to choose BVH if they don't want to, and perhaps there are more items on that flow chart that they can accelerate as well, which may open up different bottlenecks than the ones nvidia has.
My 2 cents on the topic we've been wading towards; Too much credit to Sony or MS for work that will largely be designed and developed by AMD; By release I won't be surprised at how similar their two solutions will be.
How they customize their hardware to get better performance out of the architecture will be the focus of some debate, but I've no doubts that everything will be largely the same tech. With MS we have some insight on how they design their hardware by looking at their methodology with X1X; we can make some assumptions that they will apply that to their next console; perhaps even apply that to RT as well as we've seen some GDC demos referencing downscaling DXR down to Xbox One (thus it's possible they have released DXR on the SDK kits for dev testing but will never see production use) But outside of those types of things, I'm not going assume much more.
Last edited:
It's not unbelievable. Many comments already talk about areas Sony could contribute. The discussion is about what actually Sony could contribute and how dependent RT hardware is on Sony's contribution.
The unbelievable part is the implication from the original rumour that AMD couldn't create an RTRT design themselves and needed Sony to coengineer the solution.
D
Deleted member 13524
Guest
No, the unbelievable part is you and others didn't actually read the original rumor nor the latest rumor, otherwise you'd know they make no such implication. The original rumor doesn't even mention raytracing at all.
There's now this general trend of taking 3rd-hand information from someone in the internet and just riding along with it, and then arguing back and forth based on these completely fake arguments.
I expected better from this forum, especially from long-time users.
Talk about changing goalposts..- You wrote "Sony haven't substantially engineered a console GPU in nearly 20 years."
- I wrote "Sony engineered the PSP's GPU". Your statement is factually false.
- You now try to argue the PSP isn't a console for you, nor is the PS Vita.
Handheld consoles are obviously consoles. Their creators call them consoles and their audience call them consoles.
It's not a matter of semantics if your personal definition is very clearly wrong.Andrew House while CEO of SCE said:
Even if someone in this world would agree with your personal taxonomy your argument still wouldn't hold up. You were trying argue that Sony doesn't have the ASIC engineering talent to co-develop a ray-tracing module with AMD because "they haven't been working on GPUs for over 20 years". Now you're arguing that engineers who design mobile GPUs aren't elligible to co-develop hardware RT solutions. Which makes even less sense.
You're misquoting. I wrote the Cell does graphics work, not all graphics work.
If we're going to be overly pedantic, no GPU does "all the graphics work".
Your claimed there are people saying "Sony can turbo-charge AMDs designs to a degree that AMD couldn't manage". No one made that claim. Not in this forum, not in the original rumor from Forbes and not in the reddit leak that was accompanied by pictures of a Sony event.
All you can find are statements saying Sony has engineering teams that can co-develop hardware blocks for an ASIC with AMD. The "turbo-charging" part was entirely made up by you, and others before you.
Sony co-developing GPU modules for RTRT when developing a Semi-Custom solution doesn't mean they're inherently better than whatever AMD would come up with by their own. It just means it should be more suitable for the customer's/Sony's needs.
Last edited by a moderator:
D
Deleted member 11852
Guest
Nobody here knows what only Sony actually knows about RT. That's the point isn't it!
I posted a little commentary on a linked to Reddit post and a vague question from ultragpu about what that might mean...
That's it! The furore over that little remark is completely over the top. I didn't accuse anyone here of making outlandish claims, nor...anything! It's remarkable how readily people's knickers get twisted.ultragpu said:
As for discussing outlandish rumours and fake news, that's exactly the purpose of this thread! Baseless rumour discussion. Feel free to stay out of it if you don't want to discuss baseless rumours. If you only want to discuss evidence-supported rumours from quality sources, stick to the technical thread.
The gpu is a custom variant of Navi. It will support ray tracing. That's all we got.Nobody here knows what only Sony actually knows about RT. That's the point isn't it!
More a case of what fields Sony could bring in meaningful talent. eg. Do Sony have chip engineers that'd be of use? Like I said early, AFAIK their only chip design is imaging. Is that the case or am I wrong?Nobody here knows what only Sony actually knows about RT. That's the point isn't it!
Sensors is what they have world leading fabs capabilities.
They also make gigantic video processors in their broadcast equipment and $100k+ cameras, frame interpolation asics for TVs, down to companion processors for smartphones. They just don't fab them themselves.
D
Deleted member 13524
Guest
Oh my bad. I really thought you had written this:
Which has been repeated over and over again in this thread and the other.
¯\_(ツ)_/¯
Someone from Sony had to pick up the ARM IP, PowerVR IP, DRAM controller IP (probably Synopsys?) etc. and put it all together into a single SoC for the PS Vita back in 2011.
That wasn't that long ago. Assuming they may have started working on a Vita 2 after the Vita launched, there might be a team that worked on stuff that ultimately didn't see the light of day after a couple of years into development. They didn't know the platform was going to fail at the time, and the failure certainly wasn't because of the people who developed the SoC.
It's only been repeated in response to ongoing arguments about what's being argued about.
For clarity, hopefully enough to end this cycle of nonsense, no-one here is saying AMD need Sony to make RT hardware, and neither are they saying Sony have nothing to contribute to a collaborative project.
Conversations are difficult when things like this happen:
I didn't misquote or misrepresent. That's demonstrably not true. The context of my response is crystal clear. You said Cell was originally intended to do all the graphics work. I responded about the pre-release ideas for the PS3. I responded directly under the paragraph in question, talking only is past tense. If anyone can be arsed reading back they'll see.
"Handheld" opposed to "console" is very common shorthand, particularly for people who've been around games for a long time. I didn't invent it, or move goalposts to include it. Unless you think I can time travel and seed most of the internets gaming forums: https://www.google.com/search?q=forum+handheld+vs+console
I never claimed mobile GPU designers can't develop hardware RT. I'm the person who brought PowerVR and their mobile ray tracing hardware up in this thread about two days ago [edit: Nope, Shifty got there first]. And you know I did, because you responded to the post. Then you link to it existing like I don't know, like some kind of triumph.
Edit: NM. I don't think anything good can come from continuing to discuss these matters.
I didn't misquote or misrepresent. That's demonstrably not true. The context of my response is crystal clear. You said Cell was originally intended to do all the graphics work. I responded about the pre-release ideas for the PS3. I responded directly under the paragraph in question, talking only is past tense. If anyone can be arsed reading back they'll see.
"Handheld" opposed to "console" is very common shorthand, particularly for people who've been around games for a long time. I didn't invent it, or move goalposts to include it. Unless you think I can time travel and seed most of the internets gaming forums: https://www.google.com/search?q=forum+handheld+vs+console
I never claimed mobile GPU designers can't develop hardware RT. I'm the person who brought PowerVR and their mobile ray tracing hardware up in this thread about two days ago [edit: Nope, Shifty got there first]. And you know I did, because you responded to the post. Then you link to it existing like I don't know, like some kind of triumph.
Edit: NM. I don't think anything good can come from continuing to discuss these matters.
Last edited:
I must have missed some information from a rumor or an actual source, but why are some assuming (not necessarily here) Xbox-Next or Anaconda will have a dedicated GPU and CPU, but unwilling to believe PS5 has such a design, other than an APU?
Personally, I believe both will have an APU design, nothing eclipsing 200w (under-load) power-draw, especially within todays console space designs.
Personally, I believe both will have an APU design, nothing eclipsing 200w (under-load) power-draw, especially within todays console space designs.
Not sure, I think it was the idea of the hand written leak of chiplets being the setup for the XboxNexts. But I still think that's just way too costly to be price efficient at the 399-499 marks.
RTG is working on their RT solution so that's confirmed. When it arrives is a ?. Which models might be easier to narrow down. We know that RT is still compute heavy, so let's just use more of their documentation here:
So if we look at the flow chart there is a decision point on whether or not you are using triangles. Left is for non triangles, so say quads. You have to provide your own intersection shader for that. If it's triangles it's on the right - that goes straight to fixed function. You would also go left if you didn't have hardware acceleration for fixed function, then you're back to using a software intersection shader.
So ... based on that flow chart, do you think it would be possible to mix and match different types of intersection test? E.g. use voxel (terrain) and tri (characters)? Err ... put both in the same acceleration structure and enumerate either type?
Hmm RTX 2060 is 6.5 TF. Lockhart is rumoured to be ~ 4TF, which on the surface of it doesn't look great. Then again, would Nvidia even try selling RT to people who didn't mind gaming at 1080p/30 fps? There's probably a point where they want the minimum die area and the maximum incentive to go up to a higher margin product like the 2060.
That's my thought too.
As an aside, I wonder if Nvidia tried to use hardware RT acceleration to flog Turing to either Sony or MS? That might have put some pressure on AMD as might favourable moves to MS for an evolution of DX. *shrug*
Nvidia had to have been working on this since 2016, at least.
I've been wondering if the strain of building console processors has pushed Navi back. Even without RT, having to change the Navi design to allow for the possibly for compatibility with PS4 and X1 might have been more than AMD were comfortable with. With the the Pro, X1X, and now PS5 and Xbox next(s) they've had a lot of custom work on their hands that requires ongoing compatibility.
Yeah. Sony and MS are no doubt great at knowing what they want, but it's ultimately AMD that have to deliver it. Even with X1X and all the profiling MS did it still came down to AMD making it happen. E.g. "we're going to need 12 memory channels" or "these changes to the cache behaviour, please".
Not seen anything like that about Scarlett.
But for PS5 the belief is know the APU, Gonzalo.
- Status
