demigod
Member
This is clearly false. You can play RotTR with a 970 in 4k, just not at 60fps.
Prices on 4k TVs are dropping. You can get a sony 55inch for under 1k.
PS4K information (~2x GPU power w/ clock+, new CPU, price, tent. Q1 2017)
- Thread starter OsirisBlack
- Start date
This is clearly false. You can play RotTR with a 970 in 4k, just not at 60fps.
Prices on 4k TVs are dropping. You can get a sony 55inch for under 1k.
-ImaginaryInsider
Member
Roughly 10% of USA owned a 4k TV as of December of last year, it's supposed to be near 20% by end of this year, and 50% by the end of decade.
And that's with slightly higher prices and no real 4k content...
But smaller GPUs can render 4 1080P frames & the one I posted can render 4k at 60fps.
If the PS4 main GPU is doing the heavy lifting the smaller GPU could render 3 less demanding pixels to use for the up rendering.
jeff_rigby
Banned
Part of the 25X more efficient by 2020 plans by AMD is that more powerful APUs have multiple GPU power islands where GPU blocks can be turned on and off as needed. That and how memory is used are some of the current efficiency savers. dGPUs are not a Game Console Semi-custom feature, it makes no sense just as having expandable memory doesn't make sense; PCs need to have both.
Having an OS that supports multiple power island GPUs or DSPs or FPGA is the problem and issue and GPGPU. OpenVX, Vulcan and HSAIL are answers to supporting new efficiencies be they hardware or design. The Hardware manufacturer provides drivers as APIs that an OS like Windows or Linux can use. The issue is the APIs must be accepted standards (Khronos) and widely used and is what AMD is dealing with and Game Consoles an answer.
Game Developers will resist using an API only offered by Game Consoles until PC hardware is on the market that also offers those APIs. That's all the new dGPUs and APUs offered by AMD.by this October.
jeff_rigby
Banned
How about an accelerator that would be much more efficient (greater than 5X up to 100X in a DPU configuration) than a GPU at the task in question. I'm also talking up-scaling not up-rendering. What ever system is to be used in the PS4K and I think the Launch PS4 for games will also be used to up and down render media for the UHD digital bridge and must not use much power so a second GPU even on die is less practical. I use the term render because color palette and Dynamic range also need to be handled.
The VPU and ISP blocks in the following slide (credit onQ123) are essentially the same as Xtensa DPU accelerators that are in the PS4 and XB1 and the MCU like another Xtensa accelerator for True Audio. Notice what they support which is mirrored in the Game consoles and understand that the Xtensa accelerators are equally efficient at CODECS, vision and up and down scaling video.
It's a Cadence - Tensilica - Xtensa DPU The XB1 is Cadence (ARM IP) and AMD IP and the PS4 Southbridge contains the Xtensa processor/DPU on an ARM AXI bus.
The GPU you posted is more than an order of magnitude under powered for rendering a standard PS4 game at 4k 30fps. It essentially just supports output at the resolution and frame rate like playing a video.
Those other 3 pixels are each equally as demanding as the first. That is why uprendering to 4k is more taxing than natively rendering at 4k. If they were less demanding to create such as applying an algorithm to the original image then it would just be upscaling.
You need to look into how rasterization is done on a GPU to get a sense of just how much processing is actually being done. Try reading through this:
http://www.scratchapixel.com/lesson...plementation/overview-rasterization-algorithm
Upscaling is already incorporated into most 4k tvs. It would be pointless to have the hardware do what the tv already can.
The task in question, "up-rendering", is still rendering 8.3 million unique pixels, same as a native 4k frame. If there was something 5 times more efficient than GPUs at that task we wouldn't be using GPUs.
That is a 2006 paper on global illumination, which is a type of lighting model. It has nothing to do with final output resolution. Edit: I see where you were drawing parallels, but rendering per pixel presents a different challenge than a per area impact of a light on the G-buffer.
Check interleaved rendering and sampling.
https://www.google.com/url?sa=t&sou...8ijinpKYCS4JkzG-Q&sig2=oZVXs0_s8Vubec00yytGzg
You can use as small rendering targets you want to get as high resolution you want.
Completely non-interleaved method would be to render image in small tiles and combine in the end.
Aztechnology
Member
I like the change of direction in this thread.
Obliterator
Member
Every time jeff rigby posts I feel like my technical knowledge is woefully inadequate
Aztechnology
Member
Well considering he's a computer engineer? with probable experience either fabricating or otherwise I wouldn't feel too bad. I am curious to know where your extensive knowledge stems from though Jeff.
Panajev2001a
GAF's Pleasant Genius
Up rendering, in terms of emulating low level graphics pipeline depending on specific memory layouts to perform rendering and post processing (exact resolution/frame buffer size as part of the variables used/dependent upon in a way) it is the safest and most compatible option. You are letting the pipeline re render the same exact frame in the way it would have normally done it just with subtle sub-pixel shifts (affecting how each pixel is calculated depending on where the sample position is, but not affecting the rest of the pipeline) and then accumulating the final frames in the host computer's memory. It is a little bit like having 4 or more PS2 running inside the PS4 and letting the PS4 synchronise them and merge their outputs... essentially something like what they were doing with the GSCube workstation (no emulation involved there though).
Obliterator
Member
Oh I hear ya, I absolutely love reading his posts as his knowledge is amazing I just only can understand about half of it most of the time lol
GhostTrick
Banned
But uprendering still require a lot of power as you're rendering the native game at a higher resolution.
so what? its absolutely irrelevant that the garbage GPU you posted CAN render at 4k/60...the question is what can it render at that res/fr??...
it wont even touch a PS3 game at those specs, let alone a PS4 game...i promise you that
omg...you really need to stop.
less demanding pixels? you really have no clue as to what youre talking about...im done with this discussion, youre so clueless i can't keep up with your constant spin cycle...its tiring...
admittedly im too tired to read that right now, but based on your comment, sure you could render an image in small tiles and stitch it together to form a monster image of super high resolution...
was it the Gran Turismo games that did this with multiple PS3's?? each PS3 would render a 1920x1080 "corner" of the screen and then stitch them together?...
thats not what has been discussed here though...what we are being told is happening is that the GPU is rendering a 1080p frame...then its taking the 3 previous 1080p frames and by shifting some pixels around, injecting some unicorn semen, and smushing them together you create some magical 4k image...
what i think youre talking about (again too tired to read right now) is rendering separate tiles (say 4 1080p tiles in the case of 4k) and then placing them next to each other to form a single 4k image...that would actually create a native 4k image, though you would need the horsepower to render all 4 at the same time...the PS4k will not be able to do this...
Aztechnology
Member
Same here it sparks my interest so much. The really technical analysis makes me want to study it. I have a background in some computer science and modeling/animation. But there's something about computer engineering discussion that just... engages me. Sometimes I think I'm in the wrong field (education, haven't graduated).
jeff_rigby
Banned
Kaveri uses the same generation Xtensa processors for UVD that the XB1 and PS4 use for their vision processing and codecs. From the section on UVD in Kaveri:
Regardless of a semantic or any understanding or mis-understanding of rendering or up-scaling, it is accomplished with a Xtensa accelerator.https://www.guru3d.com/articles-pages/amd-a10-7800-kaveri-apu-review said:
Enough already all of you.
Gesh, I try to educate you guys on what's in the game consoles and NO one listens. This will be implemented by both consoles and is a reason that AMD, XB1 and PS4 are all using the same family of Accelerator.
Cthulhu_Steev
Member
I hope I can just stack two PS4s for more power so I don't have to buy PS4K, Cerny designed them to stack, like Voltron.
Aztechnology
Member
Many of us appreciate it Jeff!
jeff_rigby
Banned
No, I have an EE background so I can understand the terms and do lots of reading. I'm 64 so most of my experience is analog .
captjohnboyd
Member
Get the fuck out of here?! 64? I never would have guessed it. I'm in my mid (to late) 30's and I like to stay up on all the newest tech often simply because the process interests me not necessarily the actual using of the item so I imagine whatever new fangled things the kids are playing with 30 years from now will still interest me in some form or another but I can only hope that I have half the passion you do by the time I'm your age
Agreed.
Basic idea is to divert rendering at Vertex or geometry shader to get duplicated version for outputs. (Same as instanced left/right eye rendering for VR)
Anyway agree what you say it's guite silly way to do it.
Especially now that rendering is not about filling a single buffer. (Temporal, deferred etc methods make this even harder and might even cancel some of the usability by using same sample locations.)
The cost is also tremendous as you have to setup polygons several times, as well as g-buffers and all processing involved.. (Possibly even CPU stuff.)
It really might become messy quite fast.
Simply rendering into bigger buffer is both easier and faster.
Aztechnology
Member
That's awesome, you definitely know your stuff.
Seconded!
Ignoring the BS being thrown around, Panajev gave a pretty clean explanation of what it actually is:
That said, modern game engines are designed so that the pipeline shouldn't have an issue with a change in rendering resolution, thus it seems it would be an unnecessary cost to create the frame in this manner.
Xtensa is an accelerator for a lot of general multimedia applications, of course one is going to be included in the next iterations of the consoles as they have to handle all different quality levels of media that will need it anyway. That said if the system is already outputting native 1080p or 4k there is no point in wasting the the extra .5 watt or so to upscale that when there is a better processor for that task in the TV.
For all it matters that portion of Xtensa might as well shut off under the circumstances we were discussing. Not saying it's irrelevant, just that it wasn't pertinent to the discussion.
Edit: BTW, I appreciate the extensive write up as a fellow EE with a lot of years to catch up on.
Caayn
Member
Except with one I get the feeling that he knows what he's talking about while the other just rambles along having no clue what he just said.
Not really... While Jeff could be an old school Jedi, onQ is more like Jar Jar Binks...
jeff_rigby
Banned
Possibly but TVs have power restrictions that the Game Console or Computer don't have so they (game consoles and Computers) can do more processing. This is why Kaveri does it rather than leaving it to the TV. Edit: Kaveri has a HDMI port in case you tried to argue that Kaveri is usually outputting to a monitor with no video processing. 4K TVs are the best monitor.
AMD has stated that the same hardware used for HEVC codecs in the XB1 is used in AMD UVD and that is the Xtensa accelerator. It can do much much more and AMD has had them in APUs since 2010 along with the Trustzone processor as Xtensa processors/DPUs need a AXI ARM bus. ALL AMD APUs can support gesture recognition using the Xtensa accelerator. Starting with Kaveri (UVD 4.2) it can support HEVC and Carrizo (UVD 6) can support HEVC with a duty cycle allowing it to turn off part of the time while processing a HEVC video stream.
In the beginning I also thought that Jeff knows some things but after so many posts with mainly buzzwords, the connection of things that don't really have a connection and putting these into posts in a way that it obfuscates the message and his inability to express in layman terms what he tries to say made me realize that he doesn't really understand what he posts.
Not that the stuff he posts (the actual links) but his conclusions and connections for all this are just guesswork.
Aztechnology
Member
Panajev2001a
GAF's Pleasant Genius
I agree with you, modern engines are quite likely not to make the same usage of their video memory as the PS2 ones made of GS's eDRAM so this technique is likely over engineered for that. In terms of the scaler chip being built in or not/enabled or not for 1080p to 4K scaling, it is true that Sony has great TV's to sell, but if I cared a lot about my device video quality (maybe part of my unique selling points) I would prefer to do the scaling on chip and send a reliably great signal to the TV instead of trusting an unknown chip on the Unknown TV.
As a ECE graduate, I sympathise with lovers of both EE and CS fields
. Researched posts are appreciated and the intention is good.
EventHorizon
Member
Wow I can't believe I made it to the end of this thread. I was literally Googling words and concepts every few posts or so in order to try to keep up. Here is my layman's take away from the topics discussed.
- Uprendering is not upscaling. It is impossible for upscaling to add information to an image whereas uprendering does
- Uprendering to a 4K image takes more processing power than rendering to 4K directly
- A full GPU is needed to create the extra pixels needed for uprendering. DPUs cannot take over this task
- If the PS4K uses uprendering, it will uprender to an intermediate resolution (1440p?) and upscale to 4K from there
- Uprendering and Dual GPUs biggest benefit, if implemented, would be seamless backwards compatibility which comes at the expense of some performance
I think Sony will probably make an API or framework available to improve the image quality of games targeted towards the OG PS4 but it would require the developer's intervention. Hence the need for a patch for the improvements to be enabled. My guess is that this would instead be intended for new games going forward so they would look a bit better on the PS4K for developers who don't want to expend much effort to support it. Uprendering could be a part of that framework.
I wanted to mention the GSCube but people already think I'm crazy lol
jeff_rigby
Banned
Developers were not given access to many of the APIs possible with the Xtensa accelerators so launch games on Launch PS4s might benefit from this also. It's very possible that the PS4K could include a more powerful Xtensa DPU/accelerator in addition to more GPU power. They are likely going to release a new VR goggle with higher resolution which needs more processing for the video distortion and maybe better head and hand/body tracking. 1080P stereo cameras or mono with IR depth like the XB1's? 4G radio for wireless VR Goggles.
They are talking like the VR separate view on the living room TV can support multi-player and that needs more performance 2X right?
Wishmaster92
Member
Oh really?
From the 2nd GPU
Funny it didn't seem to be too complex for GT5 & GT6
Jeff is telling you about the same thing I told people about before but no one listens you can even see that he credit me in some of the post.
PS4 is at 40 million in 2 years plus , Sony don't give a damn about a NX or Xbox right now this is part of a bigger deal that includes 4K TV streaming , 4K Movie streaming , 4K Blu-rays , 4K TVs & so on.
EventHorizon
Member
Sony wants to go with simplified development. They are not going to create another complex heterogeneous processor architecture that developers would need to navigate. Openings up the Xtensa seems like it would go counter to that. For example...
Besides all that, if the Xtensa could generate pixels for display, why would they need a GPU at all? Perhaps they could put a better upscaler and antialiasing on it, but that seems about it.
However I agree that Sony will release a higher resolution VR headset which would not work on the PS4. In fact I expect Sony to ride the VR train for as long as it goes. If room level tracking takes off, they'll do that too.
What's the bandwidth and latency of 4G radio? Any solution would have to encode and decode at the very least a 1080P stream, but for an upgraded headset it'd be a lot more. That is a huge stretch to use it for VR. I could see it being used for something like Hololens since the processing is on the goggles. Only a small amount of data would need to be sent to the headset and it would be relatively to tolerant to signal interruptions. VR is far more sensitive and data hungry.
potatohead
Member
Yep.
Aztechnology
Member
You're leaving out a key interest Sony has. VR. If they don't create a box that will actually be able to handle modern games in VR at 60+ fps they're in trouble. They can drop settings alot to compensate but that's not the VR experience people want. If Microsoft releases a new VR ready console that pairs with Oculus and has the theoretical performance increase that AMD will bring with Polaris and their later APU'S revisions then I think they do care. Keeping users on their infrastructure is far far more valuable than selling a console. I'd imagine a single year of PS+ for them is more valuable than all of their console sales.
I know VR isn't actually that big of a deal now. But it will be this decade and its all about positioning of infrastructure and services. They want people to stay and grow their numbers because the money is in software and services. Perception could change if the Playstation hardware wise starts falling seriously behind. That's why I think they do care about what Microsoft and Nintendo are doing hardware wise.
potatohead
Member
We don't yet know what is palatable for VR on consoles, yet, in terms of graphics prowess.
Aztechnology
Member
We know what people's expectations are. They don't want to just play gimmicky on rail shooters or simple polygon games with little depth. People want immersive and graphically breathtaking games. The PC market around VR has proven this, I don't think it's much different on console.
As far as performance we do know what the PS4 is relatively capable of regardless of what people want to believe. The engines and graphics tech as of this moment demand certain things and the PS4 does not have the computational power to handle it like they want.
potatohead
Member
Aztechnology
Member
Here is a game called Everspace that is planned for VR.
This game would need to be graphically paired down to run at 30fps on Xbox one or PS4. Without serious image quality reductions where the game would not longer resemble the parent game graphically it would not be possible to play that on PSVR at a targeted 60fps reprojection to 120.
Aztechnology
Member
VR also make sense of a 2nd GPU for higher frame rates.
I'm not really sure how that would work in context to an APU. Forgive me I'm tech savy but I'm not engineer.
MightyPeter
Neo Member
Q1 2017...
But, but I want one naaaooooww
Looks to be quite an interesting concept, I'll gladly throw money at it asap...
But, but I want one naaaooooww
Looks to be quite an interesting concept, I'll gladly throw money at it asap...
