Sony's Next Generation Portable unveiling - PSP2 in disguise

While those two are hard to compare since they're crafted for completely different power envelopes, I'd still add that Fusion is DX11 compliant. If in theory the NGP GPU block would be fully DX11 too it would also increase its die area.

 

The 2 years comment may have been exaggerated, but those 10 years sound a bit of a longshot to me.
The biggest problem is the disparity of longevity between pixels of different colors, which results in an increasingly inaccurate color display.
The blue OLEDs are rated at losing half their brightness at 14.000 hours. Reaching ~75% brightness (which is pretty noticeable) could be had after some 7.000 hours (5 years at 4 hours/day) while the Reds and Greens are at ~95%.
That could become an issue still within the console's lifespan.
It's nothing to worry about right now, though.

Sorry, but that really sounds like a pure marketing comment. Power draw wasn't even a variable in my question.
I don't know what the clocks are in that SGX543MP4, but I have a hard time believing it can beat the 500MHz Robson with 80 ALUs, 8 TMUs, full DX11 shaders and programmable tesselator.

I do believe that if you restrict a game to OpenGL ES 2.0 featureset, the SGX543MP4 could show better results in a scene with lots of overdraw where TBDR shows its advantages, but I also believe a 500MHz Robson could pull a lot better graphics if all the eye-candy features are implemented.

 

My former post shows that I obviously don't disagree on that point. However once you start counting VLiW-N as ALUs, it would be only fair to do it on both GPU blocks. It doesn't make any difference to me whether you count each SGX543 as Vec4 or VLiW4, but since we're actually talking about 80"SPs" in one case the other one is then 64"SPs". TMU amount is the same between the two (2 per 543 core), frequency has IMHO a sizable difference and yes DX11 adds a whole damn lot more, which isn't exactly necessary for a hand-held console either. For example 8096*8096 texture sizes would be a wee bit too much for an embedded design.

It looks rather like SONY will need a DX11 hand-held console quite some time after they will have released their own (most likely) DX11 console. I couldn't imagine why anyone would want any insane amounts of overdraw, but in that regard while TBDRs still should have an advantage it's certainly not as sizable as it used to during the non early Z IMR era (if one can even call it that). They've other advantages but those are besides the point. I can't imagine either that the MP4 could measure itself against a Fusion APU.

 

I do, I do! PowerVR2DC aka CLX2

 

And how big is theoretical power peak for NGP cpu/gpu or entire system? Ballpark it. Same questions for psp/ds/3ds.

 

Based on your hands-on experience with the screen, can you confirm whether it's a pentile configuration or the more traditional one. One of the few drawbacks of Samsung Amoled is their use of the pentile, the text are a bit blurry looking on my Galaxy S.

I wonder how the OLED screen compares to Samsung Super Amoled.

 

Those mentioning the 128 MB VRAM rumor...

 

(Unless Nathan's Sony internal) It's impossible because we don't have enough info...

We don't even know clock speeds yet...

 

Maybe I'm being ignorant here, but why not? If the intermediate frame buffer is set up to be in linear colour space and has sufficient precision, why would there be a (significant) problem?

Dr Seuss? Is that you?

 

He was suggesting that Marco was one of the pioneers to show that HDR and MSAA can in fact be combined on RSX.

 

what are the specs of the cameras on the NGP? I haven't seen any talk of this anywhere.

 

Is there any info out there on who has designed the Soc. Is it a Sony internal design (do they even have an ARM license ?)

If not Sony, then my guess would be Renesas. given that Renesas has already demoed an Soc with SGX543MP graphics in it. I assumed Renesas is going with MP2 for its own Soc for smartphones.

Sony is an IMG licencee (one of the "generic" CE companies in the licensee list), as is Renesas. However I would suggest that Renesas has not taken a license for MP4, only MP2, thus the reason Sony had to take a separate license from IMG.

 

Maybe it's time for someone to do exactly that?

Android is x86 compatible, I don't know if all software would have to be recompiled to x86 (which would require the source-code), but maybe we can get the Android benchmarks to work on x86 machines?


While there are no current AMD graphics drivers for Android, I know the linux driver for ati\amd cards has been open source for a while now, and maybe those could be ported to Android?

Comparing the lowest-power x86 variants to the highest-power ARM variants in an apples-to-apples environment is starting to make sense.
The power envelopes are going to cross each other somewhere in the next 18 months, that's for sure.

 

It seems rather pointless until we know the PSP2's clocks, isn't it? It's not as if either was going to be an order of magnitude faster than the other so clocks do matter here...

 

As a rasterizer, assuming 200Mhz, I think it hands RSX its ass relative to target resolution - then again even the PS2 did that. It'd be quite funny to see something like GT port end up looking better without the castrated particles - and given the ram equivalency PD probably won't need 5 years this time.

But I disgress - my main question is whose bright idea of a "game changer" was it to come up with a completely new proprietary memory-card format. Especially if the rumours about internal storage (or rather, lack of thereof) turn out true.

Though of course I'll still want to give that screen my money...

 

The debug text I'm looking at right now looks pin-sharp..

Dean

 

Based on a brief quote here from IMG's Pete McGuiness.....

http://news.cnet.com/8301-13924_3-20029825-64.html?tag=mncol;title

"The 535--which Intel used originally in its Menlow [Atom processor]--that has two shader pipelines. The Samsung Galaxy S...that's an SGX540. That has four shader pipelines. The 543 has the same number of pipelines but each pipeline is twice as powerful. The [quad-core graphics chip] in the Sony machine would be roughly eight times as powerful as [the Samsung Galaxy S]," he said. "

.....one would assume that the SGX543MP4 is running at the same clock speed as the SGX540 in the samsung, which would mean 200Mhz.

The inference also is that the PSP2 is 16 times more powerful graphically, on a clock for clock basis, than the current gen Iphone.

 

Maybe I didn't make my point correctly.
The purpose wouldn't be to exclusively evaluate the NGP's performance against a x86 machine (the NGP isn't running android anyway), but to effectively start comparing the two architectures using the very same consumer-oriented benchmark (be it GLbenchmark, neocore, 3dmark mobile, quadrant, whatever).

There's been all this talk about ARM going up the hill and x86 not being good for low power systems. But in the last 3 years we've seen tremendous efforts from both Intel and AMD (and poor VIA who doesn't seem to get a single consumer-oriented design win) to fit their architectures into smaller power envelopes.
The same with x86-centric GPUs. Everyone just says they're not comparable. Well, I say it's about time someone did compare them.

Well, but this is completely off-topic.
Maybe I will try to make that a home project of mine and who knows I'll come back with a few results myself. My 1.2 GHz Athlon Neo L310 + RV610 seems like an excelent contender to mimic a C-50 APU.

 

That would be a highly theoretical number and based on shader power exclusively, i.e. TMUs are quadrupled, and we (well, at least some of us) don't know exactly how much the architecture scales with the multi-cores.

Besides, I doubt the PSP2 would be a bandwidth monster with something like a 512bit memory controller.