PlayStation 4 (codename Orbis) technical hardware investigation (news and rumours)

Status
Not open for further replies.
What a teaser!
I suppose you are not talking about the extended ACEs with more queues as this was documented already as a feature for upcoming graphics cards using the "GCN1.1" iteration (as well as this volatile flag stuff Mark Cerny talked about as customization and the FLAT memory model) and one could argue it is probably not so invasive to the core (okay, the flat memory accesses would count, I guess).
You can send me a PN. I can be as silent as a grave.

Edit:
I completely forgot about the additional high priority command processor for the PS4. Okay, I agree, there was definitely some customization.
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.
 
Well 6.5GB is over 5GB, so is 7GB..............he doesn't give the exact game size, nor the exact memory available. Game could also be 5GB in size but require 6.5GB of memory.

Yes, of course. That's definitely possible. Possibly that its 6.5 but he just said over 5 for some random reason. That is, technically, possible. I won't argue that's not possible.

EDIT

Just to be clear, my smart ass sarcasm aside, I fully agree that is definitely possible. He doesn't give an exact number.
 
Last edited by a moderator:
I almost forgot about this thread so I was asking this from the xbox thread.
I'm sure this would have been covered by now but is there any conclusion on the feasibility of Tiled Resources or something similar on PS4 via OpenGL yet? Is PS4's UMA not suitable for AMD's PRT?
 
PRT is a GCN feature. We have every reason to believe it's supported and the feature should already be exposed in libgcm. PRT is always useful because it cuts down unneeded memory access by the GPU irregardless of the system topology.
 
Is OpenGL the language for programming on the PS4, or is it some proprietary version (or extensions)? Isn't OpenGL somene extendable for this sort of thing?

I'm an artist, not a graphics programmer. So, this stuff is interesting to me, but not too directly effecting my work unless the tech lead yells at me to change something. :p
 
PRT is a GCN feature. We have every reason to believe it's supported and the feature should already be exposed in libgcm. PRT is always useful because it cuts down unneeded memory access by the GPU irregardless of the system topology.
Yeah logically you would think so, I hope they do a demo on PS4 at some point preferably in a real game environment.
 
I wouldn't be surprised if PS4 supports OpenGL too in Sony's attempts to make the platform as accessible as possible. We've heard that devs can go low-level, but Sony aren't forcing that. Here's a rumour from 2012 on the idea of PS4 supporting OpenGL. I guess AMD will provide the drivers and devtools, so it's just a matter of connecting the lib to the OS.
 
If the PS4 runs at 800MHz it can process 1.6 billion triangles/second but this number isn't useful for determining how many triangles you'll see on screen. You'll only be able to render this many if there's no more than one unique vertex per triangle and the number of vertex attributes is low enough that memory bandwidth doesn't become a problem. Plus the triangles must all be small enough that the rasterizer isn't a bottleneck.

Games engines don't just render one pass at full rate displaying all triangles on the scene. Some or all geometry is rendered multiple times to contribute to shadows or other effects. Plus you never achieve 100% efficiency out of all parts of the rendering pipeline. Even a contrived demo will only achieve 90+% efficiency.
 
like the dark sorcerer demo pushes 1 million polygons , so my question is can ps4 push 1.6 billion polygons ?
Yes, in graphics speak, triangle = polygon.

If the PS4 runs at 800MHz it can process 1.6 billion triangles/second but this number isn't useful for determining how many triangles you'll see on screen.
At 1080p, 2 million pixels, one triangle per pixel, the most you could ever see would be 120 million per second. ;) I guess we could supersample it x10 for 1.2 billion triangles a second...
 
If the PS4 runs at 800MHz it can process 1.6 billion triangles/second but this number isn't useful for determining how many triangles you'll see on screen. You'll only be able to render this many if there's no more than one unique vertex per triangle and the number of vertex attributes is low enough that memory bandwidth doesn't become a problem. Plus the triangles must all be small enough that the rasterizer isn't a bottleneck.

Games engines don't just render one pass at full rate displaying all triangles on the scene. Some or all geometry is rendered multiple times to contribute to shadows or other effects. Plus you never achieve 100% efficiency out of all parts of the rendering pipeline. Even a contrived demo will only achieve 90+% efficiency.

thanx for the reply @3dcgi and @shifty
even if a demo can achieve 90% , then its very impressive . The quantic dream demo only used 1 million - by the end of the next gen if they can achieve 90% efficiency then it will avatar quality !:oops:
 
GDDR5 parallel read/write capacity

Reading the Elpida docs on GDDR5 at wikipedia, there is mention of parallel read/writes being possible. With DDR3 these operations are serialized, the question is even if the 8N burst of GDDR5 increases latency, how much the read/write parallelism will impact in practice?

GDDR5 interface transfers two 32-bit wide data words per write clock (WCK) cycle to/from the I/O pins. Corresponding to the 8n-prefetch, a single write or read access consists of a 256-bit wide two CK clock cycle data transfer at the internal memory core and eight corresponding 32-bit wide one-half WCK clock cycle data transfers at the I/O pins. At 5 Gbit/s data rate per pin, CK clock runs with 1.25 GHz and WCK with 2.5 GHz.
 
Reading the Elpida docs on GDDR5 at wikipedia, there is mention of parallel read/writes being possible. With DDR3 these operations are serialized, the question is even if the 8N burst of GDDR5 increases latency, how much the read/write parallelism will impact in practice?
But the sum doesn't add up to more than the physical peak performance of 5Gbit per pin. The PHY has a peak limit. Wouldn't the write cache be enough to hide latency?
 
thanx for the reply @3dcgi and @shifty
even if a demo can achieve 90% , then its very impressive . The quantic dream demo only used 1 million - by the end of the next gen if they can achieve 90% efficiency then it will avatar quality !:oops:
You won't see Avatar quality this gen. By contrived demo I mean something that doesn't draw anything interesting just triangles splatted to a screen. When you add complicated shaders and other effects you'll never achieve the peak theoretical triangle rate over the course of an entire frame. You can achieve the peak rate in bursts when the content is suited for it.

Shadow mapping is an example where you could be limited by the triangle rate though these triangles are in addition to the scene triangle count because it means objects are being rendered multiple times.
 
Status
Not open for further replies.
Back
Top