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

Status
Not open for further replies.
Well, the article says CPU can access Garlic memory.

All AMD apus can do this but its slow because when the cpu wants to write to local memory it has to go through the gpu core to do so. And cpu reads are uncached and only a single oustanding read is supported.
 
Last edited by a moderator:
Then we are again in a Llano like HSA level?.

No. Its not like AMD is doing total redesigns everytime it drops a new APU. AMD is expanding upon the older apus and slowly integrating its overall HSA design over time. Some things that are HSA related have probably been present in AMD's apus since the beginning.

HSA doesn't require processors to have the capability to directly address all memory.
 
Last edited by a moderator:
No. Its not like AMD is doing total redesigns everytime it drops a new APU. AMD is expanding upon the older apus and slowly integrating its overall HSA design over time. Some things that are HSA related have probably been present in AMD's apus since the beginning.

HSA doesn't require processors to have the capability to directly address all memory.

And in which way would you say PS4 HSA design is an evolution of the HSA Llano design?.
 
I still don't get how they came to the conclusion that this is huma. It doesn't seem like anything new. There is nothing new on how the onion and garlic are different than the ones inside trinity or llano.
 
There are some new bits compared to Southern Islands that give more context to the how a few of the additional bit fields work in the Sea Islands document.

For example, the mention of the override bits that allow the application to use the same shader and just flip a switch to go between the various caching modes shows up in Sea Islands under some new status register fields.

There are performance optimizations in this, but I too am not certain that this alone really disqualifies similar APUs.
I did mischaracterize one thing about the HSA memory model that was made more clear after I read some of the HSA slides from Hot Chips.
At least for HSA, in terms of coherence visibility is something that is explicitly controlled by the software with the necessary preliminary flush, explicit distinction between synchronizing memory accesses, and a final write-back at certain synchronization points.
It's not clear what HSA-like code on the CPU side would need to do differently, aside from some kind of low-level protocol for requesting the necessary invalidates. Kudos if the memory management hardware could somehow map CPU accesses to little invalidate requests for convenience without extra code effort, but that's probably not necessary since HSA's big thing for such details is "a finalizer did it".

Onion+ is interesting in that it bypasses the rather brutal memory purge latency, and Orbis seems to heavily leverage selective invalidates at the start and end of a dispatch. That leads to some interesting comparisons in terms of the degree of performance impact, but it doesn't appear materially different in results from a more coarse total cache invalidate.

If there is a hUMA distinction to be made, I'm not seeing evidence that it's a problem with coherence. If there is a distinction, perhaps it is with one of the other bullet points AMD has for hUMA.

P.S.
One thing to note is the length of the on-die GPU memory window when doing an invalidate. Bookkeeping aside, that's several hundred nanoseconds to drain the memory pipeline that's still on-die. GDDR5 or not, traffic through the cache hierarchy past the L2 has a very high minimum trip duration.
 
"Ravi Sinha: What can you tell us about AMD’s plans for PC CPUs and GPUs in the near future? Will there be technology that syncs up closer to the games we’ll be seeing on the PS4 and Xbox One, or will they go above and beyond what consoles will be capable of?

Robert Hallock: People often compare the hardware of the next-gen consoles and the PC, compare specs on paper, and conclude that these consoles must be “PCs in a box.” That is patently untrue. While there are many commonalities, there were platform architecture decisions made for the consoles that set them apart from the PC in a significant way: how developers access the hardware, the Xbox One’s ESRAM, and the PlayStation 4’s UMA are all powerful examples of such decisions."

link:
http://gamingbolt.com/the-big-interview-amd-on-ps4xbox-one-graphics-technologies-pc-gaming-and-more
 
Any more info on the face recognition and voice control abilities of the new PS Eye that got spotted at PAX?
 
Isn't huma kind of a qualified derogatory specification of capabilities where the platform must meet ALL of the specifications?

If so them PS4/XB1 could be mostly qualified (IE gaining some/most of the benefits) but still unable to use the term?
 
And in which way would you say PS4 HSA design is an evolution of the HSA Llano design?.

http://techreport.com/news/24737/amd-sheds-light-on-kaveri-uniform-memory-architecture

Will hUMA mean CPUs and discrete GPUs can share a unified pool of memory, too? Not quite. When the question came up during the briefing, AMD said hUMA "doesn't directly unify those pools, but it does put them in a unified address space." The company then stressed that bandwidth won't be consistent between the CPU and discrete GPU memory pools—that is, GDDR5 graphics memory will be quicker, while DDR3 system memory will lag behind, so some hoop-jumping will still be required. (As an interesting side note, AMD then added that "people will be able to build APUs with either type of memory [DDR or GDDR] and then share any type of memory between the different processing cores on the APU.")

http://arstechnica.com/gaming/2013/...4s-hardware-power-controller-features-at-gdc/

The 176 gigabytes of total bandwidth provided by that GDDR5 RAM are much more efficient than the 40 gigabytes a second provided by the standard DDR3 RAM used in most current computer systems. The unified address space should also cause fewer headaches for developers than the mixed architecture of the PS3, Norden said.

Llano lacks an unified address space.

I'm a lay person but the utility of hUMA revolves around removing the need for an application to explicitly move data around to expose it to different processors by providing a virtual unified address space where a lot of the data movement is invisibly handled by the hardware.
 
Last edited by a moderator:
So we now know the final clock speeds of the X1.

Is there any indication of what the clock speeds are in the PS4 and if they also received a similar bump?

I can't find any official clock speed specifications of the PS4, just rumors and estimates. Sony's web site is also mum on the issue.
 
So we now know the final clock speeds of the X1.

Is there any indication of what the clock speeds are in the PS4 and if they also received a similar bump?

I can't find any official clock speed specifications of the PS4, just rumors and estimates. Sony's web site is also mum on the issue.

It's not confirmed. And if they don't match MS clock, i don't think it will ever be.
 
So we now know the final clock speeds of the X1.

Is there any indication of what the clock speeds are in the PS4 and if they also received a similar bump?

I can't find any official clock speed specifications of the PS4, just rumors and estimates. Sony's web site is also mum on the issue.

Sony internal developpers Guerilla/Evolution have been making intresting framerate claims for final products that look pretty far from the current code. Dont know if it means anything
 
Sony internal developpers Guerilla/Evolution have been making intresting framerate claims for final products that look pretty far from the current code. Dont know if it means anything

uhm? what kind of claims?
 
So we now know the final clock speeds of the X1.

Is there any indication of what the clock speeds are in the PS4 and if they also received a similar bump?

I can't find any official clock speed specifications of the PS4, just rumors and estimates. Sony's web site is also mum on the issue.

Well we know that something in the PS4 can reach 2.75GHz.

Some people said that it was the wifi but I never seen a wifi radio clocked at 2.75GHz.
 
Well we know that something in the PS4 can reach 2.75GHz.

Some people said that it was the wifi but I never seen a wifi radio clocked at 2.75GHz.

The GDDR5.
 
Status
Not open for further replies.
Back
Top