Next Nintendo handheld

[Colonel]

Well I havent said much but Nintendo somehow is able to get about 15-20 hours out of there handhelds and if they did it with the Gameboy, Gameboy color, GBA, and DS, Im sure they can prob. get it out of there next one. Never underestimate Nintendo.

 

[tangey]

THe pandora project is probably the best benchmark out there regards what can be acheived with a Cortex A8 + SGX configuration.

Ok its a open platform, with full keyboard and probably a lot of ancillary stuff that a dedicated handheld wouldn't have, but still acts as a decent indicator.

THe FAQ suggest its up to emulating the original Playstation, and many existing handhelds

http://en.wikipedia.org/wiki/Pandora_(console)

500Mhz processor, 10+hrs battery life

 

[Panajev2001a]

you can take a look at current handhelds, batteries wont improve much. In the case of the PSP you are looking at a 2000mAh Accu, which lasts 4-6 hours, meaning the whole PSP (LCD,2xMIPS CPU, Gfx Chip, RAM) draws less then 0.5 Watt. Should give you an idea how much you can spend on 1 CPU.. it aint much and I cant see anything but MIPS or ARM fitting the bill (both beeing way leaner then PPC btw). 1-2 cutdown SPEs might be possible, but I doubt it.

And if you think there will be whole cores (out of 2-3 available) sucking power while beeing locked away for the OS, then you are nuts

Ahem... how much interaction do you think devs have with the Media Engine and its friends on the current PSP (we are talking about a several GOPS capable configurable sound processor [the VME], a video decoding engine, and a full MIPS R4000i+FPU core... all locked away for the OS) ?

Here I am talking about a single SPE doing ALL of that (video decoding, sound processing, DRM, OS assistance, etc...).

Also >=0.5 Watts might well be achievable... a 2200 mAh battery for the current PSP is already out now .

Sure, low voltage is key... right now the current 45 nm CELL BE can reach 3.2 GHz with a Vdd of 0.9V and sucking out about 20 Watts... and we are talking about full good ol' PPE (which is one of the hottest power hungry portions of the chip), 25.6 GB/s XDR MC, 35 GB/s FlexIO, full-speed and full-scale EIB, etc...

Replace the PPE with a very low power MIPS/ARM core, take out 4 SPE's, scale down or replace the EIB, re-do XDR MC and FlexIO interfaces to optimize them for low power operation (and different bandwidth requirements a portable platform would have given the performance it is targeting), and clock it all at 600 MHz (how low can they get the voltage to go? Not clear...)...

http://www.realworldtech.com/page.cfm?ArticleID=RWT022508002434&p=3 (and other pages in that same article)

I am not sure a 1/100th of the power consumption of the current CELL BE 45 nm at 3.2 GHz can be achieved (it would place it at 0.2 Watts) though...

We could also remove the FlexIO interface completely, if we were to keep a rectangular die for the processor and embed the GPU core in it.

An SPE is 6.47 mm^2 at 45 nm (at least in the version they showed at ISSCC 2008).

The largest (in this PR

http://findarticles.com/p/articles/mi_m0ECZ/is_2005_August_1/ai_n14844669 )

PowerVR SGX core is about 8 mm^2@90 nm so even considering a 45% scaling from 90 nm to 45 nm would net us 3.6 mm^2 (and everyone criticized SPE scaling from 90 nm to 45 nm because it "only" got a 43.9% scaling).

We could almost double that core and still kinda fit in the same space of a missing SPE...

CPU core + 2-3 SPE + ~6 mm^2 PowerVR SGX core all in the same die, clocked at 600 MHz and with the SGX core clocked at 300 MHz or less (depending on the internal bus speed).

It should pack quite a bit of punch .

 

[RussSchultz]

Given the timeframe and the negligible die size, it'll nearly certainly be a Cortex-A9 at 600MHz+ with a plain FPU (no NEON) and a moderately sized L2 cache, probably 256KiB. FWIW, that would make it faster than any other handheld in the world today that I'm aware of, but that's more of a testament to ARM and Moore's Law than Nintendo. Here we're talking about an entire CPU subsystem that costs ~$1 to make...

You are way underestimating costs. The packaging ALONE would make the cost to nintindo ~$1 (IOW, the most cost effective packaging options for a SoC of this size/class bottom out around $.50, which would margin up to ~$1.)

 

[INKster]

I believe that, not only will Nintendo go with ARM again, as i said earlier, as it will probably use an ordinary ARM11 core or two, not Cortex A8/A9.

If we think about it, going from 67MHz + 33MHz ARM7/9 cores to a ~500/600MHz ARM11 solution is still quite a jump in overall performance.
Also, the number of suppliers and customization options for ARM11 cores is much more extensive, and, as we know from the Wii, Nintendo likes to rake in as much as it can through hardware profit margins too, not just software sales.

 

[Arun]

You are way underestimating costs. The packaging ALONE would make the cost to nintindo ~$1 (IOW, the most cost effective packaging options for a SoC of this size/class bottom out around $.50, which would margin up to ~$1.)

Good point, although I actually wasn't - I meant that the incremental cost of that CPU in what would otherwise be an identical CPU-less SoC would be less than $1. It's not like they can get away without a GPU or a memory controller (sorry Nintendo ), so the incremental cost is what matters. Of course, there's an incremental package cost too which I am not able to properly estimate, although I would tend to believe it is on the order of cents...

Panajev2001a: Well as I said I don't disagree that SPEs are a viable option. I don't think there's a clear advantage over, say, ARM's NEON (which would achieve similar gflop ratings) but it's definitely viable. However, I believe reserving a full SPE on a handheld is absolutely ridiculous, no way in hell anything similar is ever going to happen, and so I still believe 2 SPEs (or maybe 3) is the best approach. Obviously if they had 2-3 at decent clocks they could use a non-unified PS-only GPU and let the developers do interesting things too. Once again though, the advantage versus Nintendo would be nowhere as large as this generation's.

 

[Panajev2001a]

Once again though, the advantage versus Nintendo would be nowhere as large as this generation's.

I know and I do not think it'd matter .

Also, about the reserved SPE: you conceded that 3 SPE's (my idea is to have an SGX core, thus unified shaders, in the place of the 4th missing SPE thus avoiding FlexIO, having a UMA set-up) could be doable.

Still, games must run sound (multi-channel too), TCP/IP Networking, video decoding, secure/robust DRM (isolated mode is a very neat feature SPE's have), in-game VoIP and messaging, etc...

The idea is to have a light main CPU core that would offload most heavy duty processing to this SPE as well as stuff that on the current PSP is run by a separate MIPS CPU+2 programmable HW engines.

I am not sure how much time left for general processing this SPE would have, but more importantly I do not know how tightly it would be able to work with the other two SPE's when in isolation mode without compromising the security of this kind of sandbox mode.

So you would be making a choice between doing DRM/OS tasks/etc... the secure way like you do on PS3 Game OS (being able to almost directly use that code in PSP2 Game OS) and implementing it all through software on the main CPU + some HW engines for sound and video...

HW sound and video processing could very well be considerably more power efficient on a portable, so I could give you that, but I think that Sony will want to unleash a nice and protected digitally delivering DRM beast (video, music, and game stores) with PSP2 and they have one battle-tested solution on one hand and an untested/to-be-developed solution on the other.

It is something to wonder about IMHO .

 

[Fafalada]

With just one catch: That would have absurdly low single-threaded performance

Are there any Pocket-form handhelds that actually have CPU performance that isn't low?

Once again though, the advantage versus Nintendo would be nowhere as large as this generation's.

Isn't this a given anyway? All Nintendo needs to add is filtering and up the polygon throughput a bit and they'll have achieved "parity" in the eyes of average handheld buyer.

Ok they have Hardware for rendering for the Triangles and stuff but no real GPU?

I'm assuming your definition of "real" is "rasterizer + geometry processor" in which case, yes, DS has a very 'real' GPU. So no, there's nothing to free up on CPU on that respect. They could work on the rasterizer limitations though, current ones are pretty harsh.

 

[Arun]

Are there any Pocket-form handhelds that actually have CPU performance that isn't low?

ARM Cortex-A9 and MIPS 74K on 40nm should shatter that notion to pieces, yes.

Isn't this a given anyway? All Nintendo needs to add is filtering and up the polygon throughput a bit and they'll have achieved "parity" in the eyes of average handheld buyer.

Heh, they'll have a very potent OpenGL ES 2.0 GPU, thank you very much

 

[DeadlyNinja]

What do you guys think Nintendo can do with their next handheld if they focus on having a long battery life? You know, the same 10-15 hour like the DS. I'm not very good at the tech details, but I like long battery life. That's the whole reason I picked the original Gameboy over the GameGear. What kind of performance can we get assuming the DS will be replaced in 2 years. I seriously doubt they'll replace the DS next year, maybe a new model, but not a new handheld.

 

[Colonel]

What do you guys think Nintendo can do with their next handheld if they focus on having a long battery life? You know, the same 10-15 hour like the DS. I'm not very good at the tech details, but I like long battery life. That's the whole reason I picked the original Gameboy over the GameGear. What kind of performance can we get assuming the DS will be replaced in 2 years. I seriously doubt they'll replace the DS next year, maybe a new model, but not a new handheld.

That is what I was talking about in my last post. Someone atleast hears me or is on the same page about the battery power, Nintendo dont have handheld systems that dont have a long life in battery power so I think they will pick a CPU and GPU that will give them them the most.

 

[DeadlyNinja]

Yeah I know, but my question is what do you think the specs/performance would be like. Do you guys think it's possible to be a bit more powerful than say the Dreamcast for example.

 

[Panajev2001a]

Yeah I know, but my question is what do you think the specs/performance would be like. Do you guys think it's possible to be a bit more powerful than say the Dreamcast for example.

I think it can do that... the current iPod Touch/iPhone already matches Dreamcast level visuals (and thanks to the larger RAM pool can probably pull ahead too).

 

[Arun]

I think it can do that... the current iPod Touch/iPhone already matches Dreamcast level visuals (and thanks to the larger RAM pool can probably pull ahead too).

In practice maybe, but 25MPixels/s (MBX Lite @ 50MHz) is quite far away from 100MPixels/s... Anyway DS2 will be well in excess of 500MPixels/s (2 or (unlikely) 4 pixel pipelines @ 250-400MHz)

 

[Fafalada]

ARM Cortex-A9 and MIPS 74K on 40nm should shatter that notion to pieces, yes.

A device has to be made using them first. Which brings a question, what timeline are we really talking about here - both current handhelds have barely seen any price changes yet (actually they both even raised price in some territories), and NDS sellthrough is still record breaking everywhere.

It's also interesting to see how BC will be approached, PSP is simpler to emulate then NDS, but on the other hand there's the UMD factor.

 

[tongue_of_colicab]

Why would the PSP be easier to emulate? Wont that also depend on the hardware? what if nintendo does the same with the ds2 as they did with the ds? use old chips and some new for the gba bc. But if they used a ARM again would it be such a big deal to emulate the DS?

 

[I.S.T.]

Isn't the DS a fairly... odd design? I think that is what Fafalada was referring to(I could be wrong, of course. I don't want to words in the man's mouth!).

 

[Panajev2001a]

A device has to be made using them first. Which brings a question, what timeline are we really talking about here - both current handhelds have barely seen any price changes yet (actually they both even raised price in some territories), and NDS sellthrough is still record breaking everywhere.

It's also interesting to see how BC will be approached, PSP is simpler to emulate then NDS, but on the other hand there's the UMD factor.

PSN downloadable games... they are testing the waters...

IMHO, they could take an initial fit of rage from some angry consumers, but they should and quite likely could very well go with an UMD-less PSP2 which emulates PSP games downloaded from the PS3/PSN store while enjoying longer lasting battery life as well as a saner way to fight against piracy and publishers should like that .

 

[Ailuros]

In practice maybe, but 25MPixels/s (MBX Lite @ 50MHz) is quite far away from 100MPixels/s... Anyway DS2 will be well in excess of 500MPixels/s (2 or (unlikely) 4 pixel pipelines @ 250-400MHz)

MBX at (at least) 100MHz would be damn close or slightly better than Dreamcast; MBX Lite and especially @50MHz is too weak to be comparable.

To take it though from a different perspective, MBX Lite was/is aimed for mobile phones and the likes and not a handheld. Nintendo's handheld will have as a straight competitor something like the second generation PSP or similar stuff.

2 pipelines at 250-300MHz (max) sounds reasonable; 400MHz is a way too optimistic IMHLO considering the minor obstacles that have priority. What will stand against that is something that's written in the stars and my telescope is a wee bit foggy for the time

 

[roninja]

My wish would be an SGX540 but that needs to be licensed and I don't know if it has....