Sony's Next Generation Portable unveiling - PSP2 in disguise

[Exophase]

What I was saying (and what I should've worded better, apparently) was that the metric 'M tiles * N TMUs/tile in a multi-tile setup = same number of TMUs in a non-tiled setup' is flawed. In a non-tiled setup you can have all TMUs working toward a single fragment at a time. No so in a multi-tile setup - there the number of TMUs per tile puts a hard cap to how many TMUs can work toward a common fragment. Ergo my hapless mentioning of all units working on the same tile.

If we really meant to compare an MP, locale-division setup to something more canonical, we'd need to go to further lengths, looking at ALU/ROP, ALU/TMU, TMU/ROP, etc, ratios, not just summing things up.

You can't have all TMUs in a chip working towards a single fragment in any other highly parallel but single-core GPU setup. All of them follow the same hierarchical layout that dedicates some number of TMUs to a single SIMD stream.

But I don't really see what difference it makes in any reasonable GPU workload. Fragments with a lot of TMU dependencies will take longer to execute, but four will be computable in parallel. You should always have at least this level of parallelism in anything worth doing on a GPU. There may be less simultaneous load-balancing granularity overall, but that's again compensated for by having good thread load-balancing.

The actual ratios you're looking for are 2 TMUs to 4 USSE2s, each of which has a vec4-ish FMADD ALU (can co-issue, and if that means USSE1 style operations it'd be vec2 FP16 or vec2 FP32 with a shared input). As far as I understand it each USSE2 is capable outputting a pixel per cycle, but I don't know what constitutes as an ROP in this case. I think part of the blending is handled in the fragment shading and part as a fixed function output per-cycle. I don't know if the USSE2 has dedicated resources for that or needs to take instruction issue slots and ALUs.

This topic actually brings to my mind an interesting consideration. Each of the four USSE2s in an SGX543 should be capable of operating on a completely independent thread, and switching between another 3 (I think?) threads with zero overhead. So that means that there are 16 threads in flight in parallel from an MP4 configuration with quick switching among a pool of 64. This should mean that there's little overhead in running a whole bunch of separate execution paths or even different shaders on the GPU in parallel, compared to other GPUs. I wonder what kind of impact that'll have on general purpose compute.

 

[darkblu]

You can't have all TMUs in a chip working towards a single fragment in any other highly parallel but single-core GPU setup. All of them follow the same hierarchical layout that dedicates some number of TMUs to a single SIMD stream.

The 'All-for-one' was a for-the-argument's-sake example, don't read it literally. It was meant to illustrate the principle, which is:

The moment you introduce some kind of thread affinity to your resources (or add yet another affinity), their use cases change compared to a situation where you don't do that. Tiling, or any kind of locale-based affinity in your TMU pool means they don't exactly equal the same number of TMUs without such affinity, not without any degree of discrimination. Basically, you add extra burden to your resource allocation schemes (or alleviate them in the mirror case - whichever you prefer).

 

[Lazy8s]

I think each USSE can handle 16 simultaneous threads.

Considering GPGPU, a balance of more SGX cores for the A9s would've made sense, too.

I don't see that eDRAM would make much sense for VRAM here. This isn't an IMR and 128MB would be impossibly huge in this context.

 

[Exophase]

They can handle 16 simultaneous threads, but there was something about a subset of 4 being in the current working set. I didn't really get the finer details of what this meant.. Maybe it means that there's zero-cycle switch among those 4 but then a higher penalty to swap in one of the others into that set.. maybe it means that all of the USSEs on a core share those 16 but each individual USSE has 4.. maybe I'm kinda sounding stupid and should let an IMG person explain it ;D

JohnH is especially good for times like these :>

 

[RudeCurve]

3G networks have latency issues, though. Making hard to imagine proper action gaming over 3G as a viable solution.

3G is absolutely useable for online MP in my experience, however if everyone in the game is connected via 3G then latency may become an issue. With that said as long as some of the players within a game are connected through land line+WiFi it should work fine.

 

[DieH@rd]

I wonder if PSP2 games can indeed reach "ps3" level of visuals if they were rendered in 480p [instead of native 544p]...

 

[Arwin]

No need. PSP2 games don't need any AA for one, and I understand from Epic that games using UE3 can and are putting out pretty much the same game on PSP2, PS3 and 360. And if NGP can indeed emulate PSP games as stated by Sony, it may even have some bandwidth advantages in some areas.

 

[onQ]

No need. PSP2 games don't need any AA for one, and I understand from Epic that games using UE3 can and are putting out pretty much the same game on PSP2, PS3 and 360. And if NGP can indeed emulate PSP games as stated by Sony, it may even have some bandwidth advantages in some areas.

1st when I read people saying that it would be PS3 level graphics I thought it was just hype then I seen the videos & the specs now I'm thinking that maybe it can even put out a better looking game than most of the games on the PS3 & 360 when looked at on the 5 inch screen.

& it's kinda scary to think what the next gen consoles will be like now that I have seen how far they have come with a handheld , I'm thinking we will have 16 core GPUs & CPUs or maybe higher

 

[Aeoniss]

No need. PSP2 games don't need any AA for one, and I understand from Epic that games using UE3 can and are putting out pretty much the same game on PSP2, PS3 and 360. And if NGP can indeed emulate PSP games as stated by Sony, it may even have some bandwidth advantages in some areas.

Why does it not require AA? I'm kind of confused.

 

[BRiT]

Why does it not require AA? I'm kind of confused.

Perhaps the screen being small enough that it would be hard(er) to notice aliasing, is what was meant?

 

[Mobius1aic]

1st when I read people saying that it would be PS3 level graphics I thought it was just hype then I seen the videos & the specs now I'm thinking that maybe it can even put out a better looking game than most of the games on the PS3 & 360 when looked at on the 5 inch screen.

& it's kinda scary to think what the next gen consoles will be like now that I have seen how far they have come with a handheld , I'm thinking we will have 16 core GPUs & CPUs or maybe higher

I for one think people are hyping up the "PS3-ness" of the NGP way too hard. Feature set wise, the NGP will be able to do whatever the PS3 can just like the PSP pretty much can do whatever the PS2 can, but not at the same level of quality or amount.

I'm most interested in the quad core Cortex A9. Admittedly I don't know too much about ARM CPUs, but I would like to hear a well to do person go over their opinions on ARM vs x86 (specifically AMD's Fusion) on not just the merits of capability per watt and clock, but in terms of architecture related capability.

Also does anyone have any links to a chart or whatever plotting out clock speed to power consumption for any specific chips? Yes, I know it rises exponentially, but a chart is always a nice visual reference. I was at least able to find this chart below, but nothing for ARM or Fusion.

Interesting as to how that version of C2Q "only" uses 13W at 1.2 GHz with 4 full x86 cores, while Bobcat at 1 GHz with 2x 86 + 16 SIMDs is at 9 Watt TDP (but at 18W TDP when at 1.6 GHz). Makes me wonder how much the x86 cores and SIMDs are tolling the power consumption.

 

[Ailuros]

darkblu,

I tried with a simple example to explain how I've understood the whole thing and fell flat on my nose apparently (see John's explanation).

Frankly I still don't understand what you're trying to say. A second more simplified attempt to explain my chain of thought is that I don't see where the actual difference is as long as workload gets distributed over cores in a dynamic fashion. What's even more confusing for me as a layman is what the difference actually would be between a block of MP cores within a SoC compared to a GPU block within another SoC with multiple processing clusters (where each cluster has its own TMU block), especially if in the first case workload distribution is handled with hw assistance (not a pure sw level as with other multi-core GPU configs) and the latter multi-cluster block also uses TBR.

After all that there are of course other finer detailed differences, like differences in tiling methods, deferred texturing, thread handling or what else. But those are above the TMU reasoning of yours I'm trying to understand here.

Exophase,

At this time and stage there's not much IMG really needs to "hide" about Series5/XT anymore (apart from the "secret sauce"). There are a LOT of aspects I'm curious about considering the architecture itself. Some might not even matter for a handheld console like the NXP, but it's more out of technical interest than anything else.

They also claim procedural geometry amongst other things for SGX and if you look at their VGX150 (vector graphics IP core) whitepaper there are some details in there that suggest some sort of programmable geometry. Now I wouldn't think that whatever form of tessellation is supported it to be as flexible as in DX11, but I'm still curious what actually sits in there.

 

[onQ]

I for one think people are hyping up the "PS3-ness" of the NGP way too hard. Feature set wise, the NGP will be able to do whatever the PS3 can just like the PSP pretty much can do whatever the PS2 can, but not at the same level of quality or amount.

I don't think it's hype. look at the games that they showed running on the system in real time then look back at what was shown on the PS3 about a year before it was released, this really is PS3 level graphics because Uncharted NGP look as good as the ps3 launch titles

I remember people saying that the PS3 couldn't match this (Gears of War) because it didn't have enough ram because it was split into 2 sets of 256 now you're saying that what we seen on the NGP being called PS3 leveled graphics is hyping it up way too hard?

so where does PS3 level graphics start?

 

[Fafalada]

No one said that writeable save space on the carts was mutually exclusive with other forms of save management.

Of course. But having it there means game-functionality will be there with no online, and no writeable storage on the unit itself. Ie. that will most likely be your entry level unit (no 3G - at least no subscription required, and no storage).
So the 'no-internal' storage(at least in base unit) is very likely correct - to be fair it does sound like a move to differentiate itself away from the media-player/phone market, unlike the original.

But I wonder about ramifications for DD(as well as online play) if significant portion of userbase doesn't opt for the flash-card upgrade.

Feature set wise, the NGP will be able to do whatever the PS3 can just like the PSP pretty much can do whatever the PS2 can

That wasn't really the case for PSP though - the fixed function nature of the chipset meant there was a lot of things that just couldn't be done or at least couldn't be done in hw-accelerated way.

And if NGP can indeed emulate PSP games as stated by Sony, it may even have some bandwidth advantages in some areas.

Not sure why that would be the case - original PSPs biggest weakness was bandwith in the first place.

 

[onQ]

so what's the real Ram configuration on this thing?

1st I heard 1GB

then I heard 128MB of vram & 1GB of ram but the 1GB was only for the dev kits & the retail units would be 512MB

then someone said it would be more ram than what's in the PS3 (which still goes with what was said before) but does that mean it will have maybe 768mb or 1GB of ram + the 128MB of vram or was they referring to the 512MB ram + 128MB vram when they said it would have more ram than the PS3?

 

[kraken]

That's what she said.
Sorry, couldn't resist..


I for one think that a well implemented haptic feedback replaces the "urge" of a tactile feel.


Which reminds me of another feature that's absent from the specs: rumble.
Does the NGP have it?
Nintendo decided to keep it out of the 3DS (and DSi, DSXL), even though there was a rumble add-on for the DS/lite.

No one mentioned it but I highly doubt, I don't think that it would be healthy for the battery life.

 

[wco81]

3G is absolutely useable for online MP in my experience, however if everyone in the game is connected via 3G then latency may become an issue. With that said as long as some of the players within a game are connected through land line+WiFi it should work fine.

Maybe 3G is more for non-gaming use.

Or a future hardware revision will have LTE.

Depending on how integral mobile data is, it could have shaped the design. That is, they could be distributing through mobile carriers, in which case they don't have to rely on retail chains for moving hardware, which might have meant they could do DD instead of retail distribution of games, which would have meant instead of these proprietary flash media, they could have a regular SD slot and some internal flash memory for storing DD games.

But requiring 3G or any kind of mobile data plan would have made the business model difficult.

 

[kraken]

Well, I believe that 3g will have same purpose as on the tablets- for internet surf and download and therefore there certainly will be operators who will offer data plan, most probably the ones who are already have relation with sony-ericsson phones.

As far as gaming goes- I have tried 3g once for fps on my laptop and it was horrible

 

[Nano]

"NGP 8x more powerful than the Samsung Galaxy S"

- Peter McGuninnes ImgTec

McGuinness continued. "Each base has multiple shader pipelines. The Intel 535 that originally used in its Menlow [Atom processor] has two Shader pipes, while the Galaxy Samsung SGX540 S has a GPU that has four lines shaders. The 543 has the same number of pipelines, but each line is twice as powerful. PSP2/NGP The GPU that would have approximately eight times more potent than the Samsung Galaxy S. "

/disuss

 

[ltcommander.data]

"NGP 8x more powerful than the Samsung Galaxy S"

- Peter McGuninnes ImgTec



/disuss

http://forum.beyond3d.com/showpost.php?p=1517256&postcount=118

This thread has been discussing it since page 5.