Sony's Next Generation Portable unveiling - PSP2 in disguise

Well, there's Angry Birds for PS3...

lol funny when I said that I kinda knew that someone would jump through that loop hole,


but really the games look just as good as the PS3 & 360 games that came out when the consoles was 1st released & it's a handheld so to me that's PS3 level graphics on a handheld




____________

people are saying that the 3DS has GameCube level graphics (witch is true) but I don't see anyone saying that it's not GameCube level graphics because the 3DS games are 400x240 & not 640x480 like the GameCube games ,


no one is saying that NGP has the same power as the PS3 they are saying that the games look like PS3 games on a handheld,


& I think with the tech that's in this handheld with more ram & not needing to push as much pixels as the PS3 does on a big screen that they can make some games that look better than some PS3 games when looked at on a 940x544 5inch screen

when you look at Uncharted NGP you're looking at early work of Sony Bend on new hardware that's almost a year away from release , not the work of Naughty Dog on hardware that they been working with for a few years.
 
Yap, I'd go as far as stating the NGP has more power per screen pixel regarding 3D graphics than the PS3 has for 1080p rendering. I don't get why so many people seem so shocked with the press statements.
But AFAIK most games are rendered at 720p (or lower) and then upscaled, so it's a tricky issue.

As for general processing for A.I., sound, physics and other stuff, that'd be a whole other story.. unless they eventually do GPGPU in the system (but then, that could hamper the 3D performance).
 
on another note is there anything else out now that use this type of CPU / GPU combination so we can look to to see what's the results of having games programed for a Quad-Core CPU / Quad-Core GPU configuration?
 
on another note is there anything else out now that use this type of CPU / GPU combination so we can look to to see what's the results of having games programed for a Quad-Core CPU / Quad-Core GPU configuration?


No, not now. But then again, neither is the NGP out now.

High-end smartphones withh quad A9s may never come out at all, since the A15 devices may come out before there's a market demand for quad A9s.
 
Yap, I'd go as far as stating the NGP has more power per screen pixel regarding 3D graphics than the PS3 has for 1080p rendering. I don't get why so many people seem so shocked with the press statements.
But AFAIK most games are rendered at 720p (or lower) and then upscaled, so it's a tricky issue.

When I first read about PS3 level of performance the first thought was "too good to be true". Then I was slowly toying with a MP8 scenario at high frequencies (which would truly be a lot closer to RSX performance) and when the NGP got announced I couldn't get rid of the "is that all?" feeling. And then of course I bitch-slapped myself and thought again of power consumption....

As I said I don't know who's responsible for those kind of exaggerations, but I'm personally anything but for them.

As for general processing for A.I., sound, physics and other stuff, that'd be a whole other story.. unless they eventually do GPGPU in the system (but then, that could hamper the 3D performance).

Wouldn't AI be better off on the CPU & sound on a dedicated low power consumption DSP/audio processor?

SGX was designed around GPGPU; wherever there are ALU resources free for anything outside 3D most GP I could imagine outside the above on a hand-held would be obviously better off on a GPU than on a CPU. There should be way better chances to maximize ALU utilization vs. CPUs and most important the GPU while employing higher parallelism with low frequencies wins against CPUs/low parallelism/high frequencies. That should be valid though for pretty much any embedded GPU out there capable of GPGPU.
 
High-end smartphones withh quad A9s may never come out at all, since the A15 devices may come out before there's a market demand for quad A9s.

Tegra3 AP30 for smart-phones is according to NVIDIA either dual or quad core A9.
 
lol funny when I said that I kinda knew that someone would jump through that loop hole,


but really the games look just as good as the PS3 & 360 games that came out when the consoles was 1st released & it's a handheld so to me that's PS3 level graphics on a handheld




____________

people are saying that the 3DS has GameCube level graphics (witch is true) but I don't see anyone saying that it's not GameCube level graphics because the 3DS games are 400x240 & not 640x480 like the GameCube games ,


no one is saying that NGP has the same power as the PS3 they are saying that the games look like PS3 games on a handheld,


& I think with the tech that's in this handheld with more ram & not needing to push as much pixels as the PS3 does on a big screen that they can make some games that look better than some PS3 games when looked at on a 940x544 5inch screen

when you look at Uncharted NGP you're looking at early work of Sony Bend on new hardware that's almost a year away from release , not the work of Naughty Dog on hardware that they been working with for a few years.

In the case of the 3DS, there actually have been arguments about how powerful it is in raw power and geometry to the Gamecube. The difference is that the 3DS has the major advantage of having a more modern shader system that gives it some graphical features closer to current-gen (and NGP) games.

On the other hand, I'm not sure if the NGP has any significant graphical features that the PS3 didn't have. It appears that it may have more RAM and is alot easier to develop for than the PS3, though, so it will be fun to see how far developers will push the system.
 
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.
Dynamic distribution is a given, but it does not help the fact the resources allocated to a given tile of work with remain bound until the tile is ready - i.e. there's an implied coarse granularity of work distribution, and thus resource allocation onto tasks.

In the most basic of examples - where you have the exact same, homogeneous workload spread across (1) a tiled MP setup, and (2) a non-tiled setup, you're right to not see a difference - as long as all resources remain 100% utilized, the sheer quantitative ratio between the total computational resources of (1) and (2) will give us the performance picture. The difference creeps in when you introduce a heterogeneous workload to said setups. Here's a simple 'bad use case' for the tiled MP setup:

Let's have a heterogeneous workload. Task A is absolutely ALU-bound, and Task B - ultra-light on ALU, TMU-bound instead. Those two run concurrently on the MP setup. Now, the TMUs of the tiles processing task A are twiddling their thumbs, while the tiles processing task B could use some extra TMUs, alas, they cannot, as those vacant TMUs have a certain thread affinity effective on them. On the other side of the spectrum, the non-tiled, common-pool-of-resources setup (aka #2) will exhibit a better utilization of its TMUs (clearly here we assume no hard-wiring of TMUs to ALUs). In the long run, such utilization bubbles in the MP setup will hardly be much more than statistical noise, but we can assume it'd not be impossible to create a deliberate workload that makes the long-term picture quite grim on the MP setup.

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.
Well, so far I've been referring to the non-MP setup as a hypothetical 'pool of resources' - ALUs, TMUs, ROPs, what have you. The moment we introduce some sort of topology to that 'GPU soup' the situation stops being so ideal and becomes bound by its own set of resource-allocation limitations, thread affinities, etc.
 
Backside trackpad isn't a Sony innovation. I used one months ago on Motorola's Charm smartphone.

As a control for a gaming devce, though, it could be very nice.
 
with a 28nm handheld then the graphics stick that Sony is dangling wont really exist
Being a bit pedantic, but surely they are dangling a carrot.

Let's have a heterogeneous workload. Task A is absolutely ALU-bound, and Task B - ultra-light on ALU, TMU-bound instead. Those two run concurrently on the MP setup. ...
I tried to understand your example - do you mind if I attempt to rephrase it?

I assume we have N Processing Units (PUs), each of which have an "A" unit and a "B" unit. We also have a mixture of processes, M, which have some mixture of "a" work and "b" work which use A and B units respectively.

Basically, you are saying that there is a worst case scenario where we have a set of processes {Ma} that is almost exclusively type "a" work that all gets sent to one subset, {Na}, of the N PUs, and another set of processes, {Mb} that is almost all "b" work, that gets assigned to set {Nb}. In the case of {Na}, the B units are sitting idle and, vice versa, for {Nb}.

Is that an accurate summary?

In the first instance, SGX has multiple threads per USSE so it seems to me that it would be very bad luck that one unit would exclusively get items from either {Ma} or {Mb}.

Secondly, if one only had a single core, surely it will suffer from exactly the same situation.

At least if DX11-specific performance optimizations aren't being used.
Which would be (honest question)?
AFAIK, tesselation, new texture compression algorythms and SM5.0..
It should be able to use PVRTC-I and PVRTC-II at 4bpp and 2bpp formats. Also, the artificial restriction on texture sizes that exists on certain devices (for backwards compatibility) probably won't exist.
 
Dynamic distribution is a given, but it does not help the fact the resources allocated to a given tile of work with remain bound until the tile is ready - i.e. there's an implied coarse granularity of work distribution, and thus resource allocation onto tasks.

In the most basic of examples - where you have the exact same, homogeneous workload spread across (1) a tiled MP setup, and (2) a non-tiled setup, you're right to not see a difference - as long as all resources remain 100% utilized, the sheer quantitative ratio between the total computational resources of (1) and (2) will give us the performance picture. The difference creeps in when you introduce a heterogeneous workload to said setups. Here's a simple 'bad use case' for the tiled MP setup:

Let's have a heterogeneous workload. Task A is absolutely ALU-bound, and Task B - ultra-light on ALU, TMU-bound instead. Those two run concurrently on the MP setup. Now, the TMUs of the tiles processing task A are twiddling their thumbs, while the tiles processing task B could use some extra TMUs, alas, they cannot, as those vacant TMUs have a certain thread affinity effective on them. On the other side of the spectrum, the non-tiled, common-pool-of-resources setup (aka #2) will exhibit a better utilization of its TMUs (clearly here we assume no hard-wiring of TMUs to ALUs). In the long run, such utilization bubbles in the MP setup will hardly be much more than statistical noise, but we can assume it'd not be impossible to create a deliberate workload that makes the long-term picture quite grim on the MP setup.

Well, so far I've been referring to the non-MP setup as a hypothetical 'pool of resources' - ALUs, TMUs, ROPs, what have you. The moment we introduce some sort of topology to that 'GPU soup' the situation stops being so ideal and becomes bound by its own set of resource-allocation limitations, thread affinities, etc.

I think you correctly summed it up when you say "In the long run, such utilization bubbles in the MP setup will hardly be much more than statistical noise", this is exactly what we see in real scenarios, you get at most a few percentage points of fluctuation relative to the ideal due to local resource affinity. You can create a workload that breaks things but this is exceptionally unlikely in reality as long as you don't choose too course a granularity to divide your workload.

John.
 
<attempt to catch shifty's attention>
MODS: Can we split the technical discussion from the business/gameplay discussion?
</attempt to catch shifty's attention>
 
Adding a heatsink wouldnt do any good, if there is no ample air and much less circulation.
Would be much better to attach a thermal pad or heatpipe to the backside, but I dunno where it should be attached to? Metallic backplate ?:D

I doubt anything above 2W total power is infeasible for a PSP2 form-factor.
Probably launching with 28nm is planned because of that reason (diearea isnt much of a concern at 45nm, I red 35 mm² for the GPU and CPU should be < 15 mm²). ARM designs are pretty much the testbed for 28/32nm at many foundries (Clicky here) so might be there is enough confidence to get everything ready for 28nm

The original PSP-1000 has max power around 1.5~2W@ min 4 hour gameplay; I think a larger NGP may be feasible for 2.5~3W?


http://av.watch.impress.co.jp/docs/series/rt/20110128_423489.html
This Japanese article says that the weight of NGP is similar to original PSP(280g). Considering that NGP uses lighter OLED screen and excludes UMD drive, we can assume that NGP has very large batteries. Let's say two 1500 mAh @ 3.7V batteries in parallel? Therefore NGP may have power of 3W with 3.5~4 hour battery life.

What kind performance will NGP have with 3W @ 28 nm CMOS process?
Since some asian manufaturers may have 28 nm CMOS ready for in late 2011, this performance may be available for the retail model of NGP.
 
The original PSP-1000 has max power around 1.5~2W@ min 4 hour gameplay; I think a larger NGP may be feasible for 2.5~3W?


Here's a hint that it should be higher than that.

Although I believe those power figures could be to related to a prototype stage of 40nm for the SoC, and if they can get a consumer version of 28nm, the power demands could drop drastically.
 
New NGP details emerge at private event

http://www.eurogamer.net/articles/2011-02-04-new-ngp-details-emerge-at-private-event

Sony staff demoed a handful of upcoming first-party NGP titles, including Uncharted, Little Deviants and WipEout. The source said the latter was "the WipEout HD PS3 engine running on PS3 with no changes to the art platform. That means full resolution, full 60 frames per second. It looks exactly the same as it does on PS3 – all the shader effects are in there".

With Sony urging developers to create releases that work across PS3 and NGP, the implications of this are significant. "They want us to do cross-platform," said the source, explaining that the submission process has been streamlined, with only a single submission required for a title on PSN and NGP.

wow
 
But does not the PS3 version of Wipeout HD render at full 1080p? Cutting the render resolution to a quarter certainly eases things up.
 
The original PSP-1000 has max power around 1.5~2W@ min 4 hour gameplay; I think a larger NGP may be feasible for 2.5~3W?
The PSP1000 chipset had (and still has) max power way below 1W.
wlan & the display at highest brightness would push the total over 1W but then you have less than 2h battery time.

28nm is scheduled to hit production this quarter in a couple foundries, so its possible Sony is eying at this node for launch.
 
Last edited by a moderator:
"the WipEout HD PS3 engine running on PS3 with no changes to the art platform. That means full resolution, full 60 frames per second. It looks exactly the same as it does on PS3 – all the shader effects are in there".

I don't get it. So there's a PS3 game running on a PS3 (hooray?).. how is this NGP related?
 
Back
Top