• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

PlayStation 4 hUMA implementation and memory enhancements details - Vgleaks

We in

tributetwo4eooovvs6w.gif
 

satam55

Banned
http://www.vgleaks.com/more-exclusi...plementation-and-memory-enhancements-details/






Edit Added Direct X 11 support:



http://www.geek.com/games/sony-iimprove-directx-11-for-the-ps4-blu-ray-1544364/




----------------------------------------------------------------------------------------------------------

Basically an update on the implementation of hUMA from the prior thread...
http://www.neogaf.com/forum/showthread.php?t=662537&highlight=vgleaks

-----------------------------------------------------------------------------------------------------------
In multiple threads some of us where debating if it was thread worthy or not... I took the precautionary measures in making one still



Ha! showed you!... wait a minute

As the OP, you probably shoud've mentioned in the title "exclusive PlayStation 4 hUMA implementation and memory enhancements details".
 
Dang... by the way in any sort of way or form... will this help out the CPU overall power as in give it a extra boost

since both the ps4 and xbox one have similar CPUs but not GPUs


It doesn't help the CPU power in the traditional sense (like making it faster, for instance) but it helps with redundancies and this combined with GPU compute will allow the CPU to have more cycles to do what it does best - work on the more intense processes (like complex AI, physics, etc).


An incredibly simple example that will make Durante yell at me (which he probably will anyway) is using their water example above. Let's say we're making a jet ski game. Typically the water needs to be rendered in a straight line copying the data back and forth between the CPU and GPU. The CPU is saying "the jetski is making a wake" and the GPU draws the waves coming off of it. This system allows both to be done simultaneously without swapping back and forth - so it's happening in parallel instead of in a straight line.

So what does that mean? Well, if our jetski on a non-hUMA product takes 20% of the available resources to render the water and the real time waves created by the jetskis - it may now (and i'm pulling a number of my ass here, but the % isn't the point) may take 10%. So that's 10% "extra" that they have. Once you add everything... the AI, the lighting, all of the animations, the graphical effects, particles, etc... you have more overhead to add more because of the cycles you saved on the water rendering using hUMA. So the physics may be even more in depth... or the water particle effects may be EXTRA crazy.

So, yeah, it doesn't make the processor more powerful but it makes the entire system more efficient. Which in turns makes games better. So to the end user it will feel like more power - but it's not literally more power. Upclocking would *literally* be more power, but hUMA has nothing to do with that.
 

RoboPlato

I'd be in the dick
It doesn't help the CPU power in the traditional sense (like making it faster, for instance) but it helps with redundancies and this combined with GPU compute will allow the CPU to have more cycles to do what it does best - work on the more intense processes (like complex AI, physics, etc).


An incredibly simple example that will make Durante yell at me (which he probably will anyway) is using their water example above. Let's say we're making a jet ski game. Typically the water needs to be rendered in a straight line copying the data back and forth between the CPU and GPU. The CPU is saying "the jetski is making a wake" and the GPU draws the waves coming off of it. This system allows both to be done simultaneously without swapping back and forth - so it's happening in parallel instead of in a straight line.

So what does that mean? Well, if our jetski on a non-hUMA product takes 20% of the available resources to render the water and the real time waves created by the jetskis - it may now (and i'm pulling a number of my ass here, but the % isn't the point) may take 10%. So that's 10% "extra" that they have. Once you add everything... the AI, the lighting, all of the animations, the graphical effects, particles, etc... you have more overhead to add more because of the cycles you saved on the water rendering using hUMA. So the physics may be even more in depth... or the water particle effects may be EXTRA crazy.

So, yeah, it doesn't make the processor more powerful but it makes the entire system more efficient. Which in turns makes games better. So to the end user it will feel like more power - but it's not literally more power. Upclocking would *literally* be more power, but hUMA has nothing to do with that.

Thanks for the example. This is how I understood it but wanted to be sure.
 
It doesn't help the CPU power in the traditional sense (like making it faster, for instance) but it helps with redundancies and this combined with GPU compute will allow the CPU to have more cycles to do what it does best - work on the more intense processes (like complex AI, physics, etc).


An incredibly simple example that will make Durante yell at me (which he probably will anyway) is using their water example above. Let's say we're making a jet ski game. Typically the water needs to be rendered in a straight line copying the data back and forth between the CPU and GPU. The CPU is saying "the jetski is making a wake" and the GPU draws the waves coming off of it. This system allows both to be done simultaneously without swapping back and forth - so it's happening in parallel instead of in a straight line.

So what does that mean? Well, if our jetski on a non-hUMA product takes 20% of the available resources to render the water and the real time waves created by the jetskis - it may now (and i'm pulling a number of my ass here, but the % isn't the point) may take 10%. So that's 10% "extra" that they have. Once you add everything... the AI, the lighting, all of the animations, the graphical effects, particles, etc... you have more overhead to add more because of the cycles you saved on the water rendering using hUMA. So the physics may be even more in depth... or the water particle effects may be EXTRA crazy.

So, yeah, it doesn't make the processor more powerful but it makes the entire system more efficient. Which in turns makes games better. So to the end user it will feel like more power - but it's not literally more power. Upclocking would *literally* be more power, but hUMA has nothing to do with that.

thank you, this makes a lot of sense.
 
It doesn't help the CPU power in the traditional sense (like making it faster, for instance) but it helps with redundancies and this combined with GPU compute will allow the CPU to have more cycles to do what it does best - work on the more intense processes (like complex AI, physics, etc).


An incredibly simple example that will make Durante yell at me (which he probably will anyway) is using their water example above. Let's say we're making a jet ski game. Typically the water needs to be rendered in a straight line copying the data back and forth between the CPU and GPU. The CPU is saying "the jetski is making a wake" and the GPU draws the waves coming off of it. This system allows both to be done simultaneously without swapping back and forth - so it's happening in parallel instead of in a straight line.

So what does that mean? Well, if our jetski on a non-hUMA product takes 20% of the available resources to render the water and the real time waves created by the jetskis - it may now (and i'm pulling a number of my ass here, but the % isn't the point) may take 10%. So that's 10% "extra" that they have. Once you add everything... the AI, the lighting, all of the animations, the graphical effects, particles, etc... you have more overhead to add more because of the cycles you saved on the water rendering using hUMA. So the physics may be even more in depth... or the water particle effects may be EXTRA crazy.

So, yeah, it doesn't make the processor more powerful but it makes the entire system more efficient. Which in turns makes games better. So to the end user it will feel like more power - but it's not literally more power. Upclocking would *literally* be more power, but hUMA has nothing to do with that.
Okay thanks for clarifying it up

As the OP, you probably shoud've mentioned in the title "exclusive PlayStation 4 hUMA implementation and memory enhancements details".
I couldn't fit the word exclusive in the OP title... it was either that or Vgleaks in the OP title and I choose to have the website reference name for better understanding of where this is coming from
 

bigmf

Member
I need someone to explain this to me in a Futurama-Landlady-with-a-lazy-eye manner, filled with the word "kajigger".
 

Pimpbaa

Member
That’s achieved not only through an modified DirectX 11.1 API, but also a secondary low-level API specifically for the PS4 hardware.

A modified DirectX 11.1 API? This kinda made me question whether whoever wrote this knows what the fuck they are talking about.
 
Is multiply really the right word? I'm asking because I honestly don't know if that was just a slip of the tongue, or if you're being literal. In the latter case...weak hardware redeemed.

Multiply in width, not speed. And maybe multiply isn't the right word - add is better.

Think of it as counting to ten.


One person has to go

1-2-3-4-5-6-7-8-9-10



But if you add width (more people) and do it simultaneously you can get to 10 quicker
1-2
3-4
5-6
7-8
9-10

All of those happening at the same time. You're done before the single person hits 3.

Edit: This is the philosophy behind multi-core processors. They are having a harder time speeding processors up without chip temperatures becoming absurdly high so multi-core is the solution. A 2.2ghz processor on its own would be garbage. Having 8 cores that can all be on separate tasks makes it amazing. hUMA is the same conceptually, but with memory. In the past memory had to be swapped back and forth. Like on the 360 - it has a shared pool but that just means that it can designate between what the CPU and GPU can read as needed. But if you needed the CPU and GPU to see the same data you would have to transfer it from one side to the other. With hUMA both the CPU and GPU can read it at the same time, saving memory/cpu/gpu overhead by not ever having to swap it.

hUMA is the start. It only works for certain functions. But i think parallel computing is likely more of our future than just straight increases in speed.
 

Kleegamefan

K. LEE GAIDEN
Multiply in width, not speed. And maybe multiply isn't the right word - add is better.

Think of it as counting to ten.


One person has to go

1-2-3-4-5-6-7-8-9-10



But if you add width (more people) and do it simultaneously you can get to 10 quicker
1-2
3-4
5-6
7-8
9-10

All of those happening at the same time. You're done before the single person hits 3.

A great way of explaining parallelism
 

todd360

Member
One piece power rating. Courtesy of Cboat

ps4 is timeskip luffy
xbox one is timeskip zoro
pc is nami's boobs....err wait no pc is shanks
 

Zephyx

Member
Multiply in width, not speed. And maybe multiply isn't the right word - add is better.

Think of it as counting to ten.


One person has to go

1-2-3-4-5-6-7-8-9-10



But if you add width (more people) and do it simultaneously you can get to 10 quicker
1-2
3-4
5-6
7-8
9-10

All of those happening at the same time. You're done before the single person hits 3.

In layman's terms, this is the simplest explanation for it. Basically, the goal of hUMA is to utilize every cycle available to both GPU and CPU so there is none to very little idle time between the two in order to maximize performance.

However, it is still dependent on the skills of the developers to make use of the available architecture since frameworks and engines have to be refactored to fully utilize this functionality.
 
It doesn't help the CPU power in the traditional sense (like making it faster, for instance) but it helps with redundancies and this combined with GPU compute will allow the CPU to have more cycles to do what it does best - work on the more intense processes (like complex AI, physics, etc).


An incredibly simple example that will make Durante yell at me (which he probably will anyway) is using their water example above. Let's say we're making a jet ski game. Typically the water needs to be rendered in a straight line copying the data back and forth between the CPU and GPU. The CPU is saying "the jetski is making a wake" and the GPU draws the waves coming off of it. This system allows both to be done simultaneously without swapping back and forth - so it's happening in parallel instead of in a straight line.

So what does that mean? Well, if our jetski on a non-hUMA product takes 20% of the available resources to render the water and the real time waves created by the jetskis - it may now (and i'm pulling a number of my ass here, but the % isn't the point) may take 10%. So that's 10% "extra" that they have. Once you add everything... the AI, the lighting, all of the animations, the graphical effects, particles, etc... you have more overhead to add more because of the cycles you saved on the water rendering using hUMA. So the physics may be even more in depth... or the water particle effects may be EXTRA crazy.

So, yeah, it doesn't make the processor more powerful but it makes the entire system more efficient. Which in turns makes games better. So to the end user it will feel like more power - but it's not literally more power. Upclocking would *literally* be more power, but hUMA has nothing to do with that.

Hmm my takeaway is that every PS4 game will require a jet ski sequence to show off that new huma goodness

I kid, thanks for the explanation

I think I'm starting to wrap my head around this
 

itxaka

Defeatist
there is “cross talk” in that one compute dispatch may forcé an invalidate or a premature flush of another dispatch’s SC memory

They couldn't do cross chat on the ps3, not sure if they would be able to do cross talk in the ps4.



is a bad joke I know, I got nothing :(
 

mekes

Member
I think we need someone to translate the stuff, because I don't know what the stuff was that I just read.
 

todd360

Member
I think we're somewhere around the part where the shounen protag says "Time to stop playing around, let me show you my real power!"

Maybe.

Ps4: You won't be able...to keep up with me any more.
Xbox One: What?
Ps4: All of my techniques...
Ps4: ...are going to evolve by one step!
Ps4: Gear...hUMA!!!!!!!!!!
 
I think we need someone to translate the stuff, because I don't know what the stuff was that I just read.

Basically the RSX was like a post-paid contract plan with a hard minute cap and individual phone line data caps. The PS4 GPU is a no contract plan with rollover minutes and a shared data plan. For real.
 
alan666 said:
i didn't think that the PS4 was using DirectX though ?

PS4 doesn't actually use DX11.2 API. It's just on the slide to simplify what their API's (PSSL) feature set will be.

Instead of saying we support <insert list of new features> it's easier to say our API will have features of DX11.2 but we'll have even more beyond that (hence the +).

Think of it like...DBZ comparisons. DirectX unfortunately has become somewhat synomous with GPU power.
 

tokkun

Member
Quick summary:

Bypass Bits
-The purpose of the bypass bits is to avoid conflict misses in the caches (as expected)
-They have the ability to selectively bypass only the L1 or both the L1 and L2.

Four Memory Buffer Usage Examples
-These are just different examples of common graphics patterns and how you would place your data structures in the different types of memory.
-There are three different coherence classes: UC, GC, SC
--UC is non-cached. Useful for data that will never be shared between cores.
--GC is cache-coherent, but only with regards to the GPU. Useful for data that will be shared between GPU cores, but not accessed by the CPU.
--SC is cache-coherent system-wide.

Strategies for Scalar Loads
-When a read or write is issued from a single thread it uses a separate cache that is shared one per every three graphics cores.

Performance
-Information about how long it takes to invalidate or write-back the cache
-It seems that when a writeback occurs they writeback all the dirty lines in the entire cache rather than on a per-line basis?

Additional Optimizations
-They talk about modifications to the coherency protocol. The modifications suggest that reading valid/invalid and dirty bits is an expensive operation (not surprising if they do full writeback, but I assume they don't do full invalidation, so I feel like I'm missing some key detail here).
 

slapnuts

Junior Member
So with PS4 having HuMa, GDDR5, more powerful GPU, and some other stuff I forgot.

How wide is the gap between Xbox One and PS4 in none technical terms.

We cant really say because we have not had any xbox one in-depth details yet but from the looks of it ..it is pretty safe to say PS4 will have the edge over Xbone this generation but not by leaps and bounds but i feel it will definitely be noticeable to the eye in 1st party games, 3rd party games will be more of a wash as usual.
 

slapnuts

Junior Member
It doesn't help the CPU power in the traditional sense (like making it faster, for instance) but it helps with redundancies and this combined with GPU compute will allow the CPU to have more cycles to do what it does best - work on the more intense processes (like complex AI, physics, etc).


An incredibly simple example that will make Durante yell at me (which he probably will anyway) is using their water example above. Let's say we're making a jet ski game. Typically the water needs to be rendered in a straight line copying the data back and forth between the CPU and GPU. The CPU is saying "the jetski is making a wake" and the GPU draws the waves coming off of it. This system allows both to be done simultaneously without swapping back and forth - so it's happening in parallel instead of in a straight line.

So what does that mean? Well, if our jetski on a non-hUMA product takes 20% of the available resources to render the water and the real time waves created by the jetskis - it may now (and i'm pulling a number of my ass here, but the % isn't the point) may take 10%. So that's 10% "extra" that they have. Once you add everything... the AI, the lighting, all of the animations, the graphical effects, particles, etc... you have more overhead to add more because of the cycles you saved on the water rendering using hUMA. So the physics may be even more in depth... or the water particle effects may be EXTRA crazy.

So, yeah, it doesn't make the processor more powerful but it makes the entire system more efficient. Which in turns makes games better. So to the end user it will feel like more power - but it's not literally more power. Upclocking would *literally* be more power, but hUMA has nothing to do with that.

Very good explanation and it is pretty spot on from my own technical experience
 

Spongebob

Banned
It doesn't help the CPU power in the traditional sense (like making it faster, for instance) but it helps with redundancies and this combined with GPU compute will allow the CPU to have more cycles to do what it does best - work on the more intense processes (like complex AI, physics, etc).


An incredibly simple example that will make Durante yell at me (which he probably will anyway) is using their water example above. Let's say we're making a jet ski game. Typically the water needs to be rendered in a straight line copying the data back and forth between the CPU and GPU. The CPU is saying "the jetski is making a wake" and the GPU draws the waves coming off of it. This system allows both to be done simultaneously without swapping back and forth - so it's happening in parallel instead of in a straight line.

So what does that mean? Well, if our jetski on a non-hUMA product takes 20% of the available resources to render the water and the real time waves created by the jetskis - it may now (and i'm pulling a number of my ass here, but the % isn't the point) may take 10%. So that's 10% "extra" that they have. Once you add everything... the AI, the lighting, all of the animations, the graphical effects, particles, etc... you have more overhead to add more because of the cycles you saved on the water rendering using hUMA. So the physics may be even more in depth... or the water particle effects may be EXTRA crazy.

So, yeah, it doesn't make the processor more powerful but it makes the entire system more efficient. Which in turns makes games better. So to the end user it will feel like more power - but it's not literally more power. Upclocking would *literally* be more power, but hUMA has nothing to do with that.
Excellent post, thanks for the example.
 
Top Bottom