NGGP: NextGen Garbage Pile (aka: No one reads the topics or stays on topic) *spawn*

Status
Not open for further replies.
Not sure if this is the same thing or not, but earlier today I did randomly come across this note in the June 2012 VGLeaks for Orbis. Much of that info still holds for today so it may be helpful in guesswork.

“>50% longer memory latency for CPU/ GPU compared to PC!!!”

http://www.vgleaks.com/world-exclusive-ps4-in-deep-first-specs/#comment-5026
What is that trying to say? It has latency 150%+ that of PCs? Which PCs? Or is it saying slightly over half the latency of PCs? Again, which makes me ask which PCs. I assume it's the later, otherwise the author wouldn't attach 3 exclamation points to such an awful number. Besides, it seems counter intuitive for a closed box to have increased latency compared to whatever nebulous PC equivalent (I imagine) it's being compared with
 
What is that trying to say? It has latency 150%+ that of PCs? Which PCs? Or is it saying slightly over half the latency of PCs? Again, which makes me ask which PCs. I assume it's the later, otherwise the author wouldn't attach 3 exclamation points to such an awful number. Besides, it seems counter intuitive for a closed box to have increased latency compared to whatever nebulous PC equivalent (I imagine) it's being compared with

latency would be a large increase for normal game operations (outside of graphics) and because of that it would impact the graphics performance, no? it's not like they'd do all the random access reads and stream the rest - so the gpu reads would be intermixed with the cpu reads making it a latency party

i'm just guesstimating here, but that's how i visualize it
 
latency would be a large increase for normal game operations (outside of graphics) and because of that it would impact the graphics performance, no? it's not like they'd do all the random access reads and stream the rest - so the gpu reads would be intermixed with the cpu reads making it a latency party

i'm just guesstimating here, but that's how i visualize it
What?

I'm asking if it's saying the latency is more or less...
 
It definately sounds like more and I could understand the more than cpu possibly even though it's potentially much nearer if GDDR5 memory latency is that much worse then DDR3 but the 50% more then GPU makes no sense
 
It definately sounds like more and I could understand the more than cpu possibly even though it's potentially much nearer if GDDR5 memory latency is that much worse then DDR3 but the 50% more then GPU makes no sense
Maybe, but what kind of DDR3? It's all so vague. Sometimes, I wonder if this isn't some giant ruse to throw us off track with false "leaks".
 
and this is before taking into consideration that the OS *may* (if plans haven't changed- referring to yukon leak) operate on it's own exclusive CPU+GPU arrangement, as well as all the audio work being piped through the SoC/dedicated DSP.

Nah, I don't think so, most of the hardware specifics have changed since that roadmap, things that haven't are their overall strategy (like 6-8x power of 360, LiveWall, Smartglass, always on etc).

So there is no separate APU for the OS - unless vgleaks and all the other recent rumours are wrong.
 
Maybe, but what kind of DDR3? It's all so vague. Sometimes, I wonder if this isn't some giant ruse to throw us off track with false "leaks".

Just posting what I found earlier today. ;)

If it were false leaks, it's strange that everything semi-reliable seems to fit together into coherent timelines. Just saying. I'd assume that latency thing was relative to something similar on PC circa summer 2012, probably something similar to the dev kit I'd imagine. Not sure why they'd want to highlight that as it sounds really bad though.
 
What is that trying to say? It has latency 150%+ that of PCs? Which PCs? Or is it saying slightly over half the latency of PCs? Again, which makes me ask which PCs. I assume it's the later, otherwise the author wouldn't attach 3 exclamation points to such an awful number. Besides, it seems counter intuitive for a closed box to have increased latency compared to whatever nebulous PC equivalent (I imagine) it's being compared with

The normal way of understanding ">50% longer memory latency" is additional latency in comparison, hence "longer".

And "closed box" is not a panacea or short for "always better." In fact in this case you have a Pitcairn class GPU sharing memory and probably a memory controller with 8 CPU cores. That, right there, is a very good reason to assume there could be some additional latency compared to a PC GPU with discrete memory. EDIT: Another possible reason to believe additional latency would be the GDDR5 tolerances/availability. They are going for 4GB total; depending on the new these new high density memory modules, their tolerances, and how forward looking Sony is being, they may be targeting higher memory latencies to accommodate the increase in memory density as well as to control costs. Lowering the latency requirements may cast a broader net over potential GDDR5 vendors/quality; going with the best memory possible will force Sony into a niche market with fewer competitors and higher costs.
 
Nah, I don't think so, most of the hardware specifics have changed since that roadmap, things that haven't are their overall strategy (like 6-8x power of 360, LiveWall, Smartglass, always on etc).

So there is no separate APU for the OS - unless vgleaks and all the other recent rumours are wrong.

They (vgleaks) haven't detailed the DSP yet, it was an option in their poll tho.

And devs wouldn't be detailed on OS hardware as they would have no access to it. Just like BC hardware.

We'd only know about it from a leak from MS itself.
 
They (vgleaks) haven't detailed the DSP yet, it was an option in their poll tho.

And devs wouldn't be detailed on OS hardware as they would have no access to it. Just like BC hardware.

We'd only know about it from a leak from MS itself.

The DSP is for audio/video though.

The devs (and hence vgleaks) aren't detailed on OS hardware as it doesn't exist, otherwise why would there be all these rumours of 2 cores and 3 GB reserved by the OS/apps etc.
 
I notice no one has posted this new DF article yet, quite interesting:
http://www.eurogamer.net/articles/df-hardware-the-cloud-is-coming-home

Low latency PS4 Remote Play for Vita (like Nvidia's Project Shield) sounds great, might pick up a Vita just for that, if I can lie in bed and get the full fat Battlefield 4, GT6, Uncharted 4 experience.
He also confirms that MS is working on an ARM based, Xbox TV type device:

So, if Sony is well set for home cloud streaming, where does that leave Microsoft and its next-gen system, codenamed Durango? Well there's no doubt that it's capable of acting as a central gaming server in exactly the same way as Orbis, but sources suggest that a "second screen" GamePad-style strategy could well be more along the lines of an extension to its existing, under-used Smartglass tech, which would greatly benefit from the ability to receive streaming video - something that's unworkable for games on the current Xbox 360 set-up. However, from what we've managed to piece together, Microsoft's approach to streaming in general is somewhat more complex. An Apple TV-style unit based on ARM architecture, running Windows RT and perhaps featuring Xbox branding is under discussion. This would be a very basic games machine running Windows 8 apps, but geared more towards streaming video services like Netflix.

This piece of kit could operate as a standalone unit of course (an Apple TV competitor if you like), but it could also work just as well as a media extender on a home network, linked to Durango. In essence, the strategy here would be a reversal of Valve's ideas for its entry-level Steambox bringing remote PC gameplay into the lounge: Durango would take centre-stage in the living room, but the ARM box would allow for media - and gameplay - to be transmitted around the home. It sounds like an intriguing idea in theory, the only stumbling block we can think of being Microsoft's strong push for developers to integrate Kinect functionality wherever possible into next-gen gameplay. Replicating camera functions on an extender would be problematic, not least in terms of streaming all the data from a secondary camera back to the host console in the lounge.

EDIT: Anyone know who liquidboy is? He's apparently from B3D but he just confirmed bgasassin's 2x FLOPS statement on GAF:
http://www.neogaf.com/forum/showpost.php?p=47810977&postcount=1164
 
The DSP is for audio/video though.

The devs (and hence vgleaks) aren't detailed on OS hardware as it doesn't exist, otherwise why would there be all these rumours of 2 cores and 3 GB reserved by the OS/apps etc.

Yes. Just like in the leak!

Described as:

Video/Audio Accel
2 HD Dec, 1HD Enc, DSP, XMA

Because if you saw the leak:

one: App level hardware was 6-8 ARM or x86 cores
two: SoC CPU was 2 ARM or 2 x86 cores

There is a high possibility they ditched the extra CPU (as that would have made 3 separate CPUs) or the rumor of the 2 cores is wrong? I'd put my money on the 2nd possibility.

It never had dedicated ram. It was always feeding from the main pool. (probably at a restricted rate for QoS)

But again. Developers would not be detailed on CPU level hardware. Even according to DaE they aren't allowed to even access the BDRom Drive. Their only concern is the CPU, GPU and RAM and the hooks to the OS.
 
But the rumours have either 1 or 2 cores reserved for Durango OS, this has been directly or implicitly mentioned by DF, Edge, Kotaku, aegies, thruway, AndyH, Proelite, DaE etc. The only thing they debate about is how many cores and how much RAM.

Now if there was a dedicated OS CPU, there wouldn't be any such debate.
 
You are jumping too many questions to draw up your straw man.

How many compute cycles are lost in AMD's GCN due to latency or moving memory due to GDDR5 latency? Would this be impacted if 8 CPUs had access to VRAM?

Only then could you move to Durango, get the ESRAM latency figures, and begin to weigh such.

Question: Does latency compound per step? In other words, do we disregard the latency between DDR3 and ESRAM/eDRAM? Does the latency total (from DDR3 to ERAM/eDRAM to L2) to be the same or near the same as GDDR5 to L2 (or stacked DDR3 to L2)?
 
I remember there was a discussion about the amount of bandwidth available per ROP, I think it started with the idea the the PS4 wouldn't have enough bandwidth available to feed all the ROPs at max. But I don't know how to calculate that, it depends on the pixel format I guess.
Well, if 16 rops can drive 102 GB/s, then 32 would easily exceed the rumored bandwidth of the PS4.
 
The normal way of understanding ">50% longer memory latency" is additional latency in comparison, hence "longer".
Agreed, but - well, I'm probably over analyzing it, but there have been so many instances where I've seen people do incomplete comparisons and incorrectly use >"statement" or <"statement" that, with no context, I automatically start wondering. Combine that with the odd vagueness of "compared to PC", and possibly translation issues (if they're getting this information from a foreign source, as is common), as some languages are structured such that comparisons in which something is compared to be "better" can be mechanically translated as being "more", and... you know, it's just easier to say I am over analyzing it. What can I say, internet pedantry.

And "closed box" is not a panacea or short for "always better." In fact in this case you have a Pitcairn class GPU sharing memory and probably a memory controller with 8 CPU cores. That, right there, is a very good reason to assume there could be some additional latency compared to a PC GPU with discrete memory.
I didn't think about that. Does that then mean this kind of extra latency is inherit to consoles with UMAs? Did Xbox/360, or any other console with a single memory pool I may have missed, also have longer latencies compared to PC part equivalents?
 
for all this latency talk maybe people should look at the data published about using the CPU in Llano as a predictor for the GPU. if i remember correctly it was around 25% perf increase for GPGPU code which if i was to take a punt is more likely to suffer stalls then regular "graphics" workloads. If these misses are such a large problem for GPU target workloads then we would be seeing bigger caches on GPU's.

also if Durango has double the throughput per core per clock for Jaguar then its not Jaguar, especally if its sustainted not peak. 256bit ALU's means massive core and soc changes, if its just from FMA, its still needs core/soc changes above the ALU's and its not going to get sustained 2x higher throughput from it alone.
 
for all this latency talk maybe people should look at the data published about using the CPU in Llano as a predictor for the GPU. if i remember correctly it was around 25% perf increase for GPGPU code which if i was to take a punt is more likely to suffer stalls then regular "graphics" workloads. If these misses are such a large problem for GPU target workloads then we would be seeing bigger caches on GPU's.

also if Durango has double the throughput per core per clock for Jaguar then its not Jaguar, especally if its sustainted not peak. 256bit ALU's means massive core and soc changes, if its just from FMA, its still needs core/soc changes above the ALU's and its not going to get sustained 2x higher throughput from it alone.

Well in all the leaks, it has never been referred to as jaguar cores, just 8 core amd x64 cpu.
 
Status
Not open for further replies.
Back
Top