Discussion in 'Mobile Devices and SoCs' started by thop, Sep 22, 2010.
I didn't. Sorry about that.
This is correct. It is difficult to tell in what region we are from the posted numbers though - it will depend on the specific process technology as well as implementation details. Greater than linear increases are effectively the norm though. Today, desktop silicon seem to be at a point where roughly power=O(f^3). A chip such as this at the rumoured frequencies should luckily be way below that level. Power will be O(f)<true scaling<O(f^3) without knowing more about this chip than I'm likely ever to. I'd guess in the lower part of that scale, but even something as vague as that is going into speculation.
Its more than any previous generation home console and the 3DS also uses a much faster storage medium, so no it seems more than enough to deliver "Xbox+" level visuals which seems to be Nintendo's general aim for the system. This isn't a general computing device with a multitasking OS and slow internal storage like a smartphone is. Its much more RAM than any iOS game (which isn't ES 2.0 exclusive, i.e 95%+ of them, then) game has access to.
You really think a setup like that is going to be able to deliver something like RE: Revelations that blows away anything produced last generation or on modern smartphones while still lasting for ~10 hours on a small ~800mah battery!?
I think a lot of posters here really can't see the woods for the trees. Dual ARM11s doesn't strike me as the cheapest choice for the CPU and the chips Nintendo are using are clearly capable of much higher clocks but if dual 266mhz ARM11 can deliver similar or better performance than a single ARM11 clocked @ 450mhz+ while consuming less power and possibly aid in BC then it seems like a pretty smart design decision to me.
This thing has to last 10 hours in high end 3D games on a tiny ~800mah battery while also powering dual screens, that's not as trivial as some are making out, far from it. Surely the fact that your iPhone can't even manage 3 hours of battery life in a 3D game while only powering a single screen should tell you that much!? The CPUs will be clocked as highly as they can be while remaining within the desired power budget, Nintendo don't just artifically gimp their hardware for no good reason, they know better than anyone what makes a decent portable gaming device and battery life ranks right at the top of the list, fancy paper specs and high clockspeeds don't even come into it.
At the end of the day theoretical capabilities matter very little, what's important is how good the games look and is there really anyone that is disapointed with what Capcom is managing to pull off in their Resident Evil demo or the fact that in just a few short months Team Ninja have got an iteration of DOA that looks better than DOA3 up and running? The Iphone may have it beat on paper but outside of Epic Citadel I've seen absolutely nothing on that platform that can remotely compete with what even mediocre 3DS developers have cooked up in just a few months with the hardware. Heck, even Epic Citadel isn't packing nearly as much technology as RE: Revelations is, there's no character models in Epic's demo, no dynamic shadow maps, no depth of field, no object based motion blur, no self shadowing and its probably not rendering in HDR either and yet the 3DS iteration of MT Framework can do all of that and more with similar scene complexity and particle/alpha effects. Oh, and what's more, 3DS titles will manage to run for more than 3x as long on a smaller battery, how is that not an impressive piece of engineering?
Nintendo have designed a dedicated portable gaming device that must be able to turn a profit at less than $200 from launch, not a general computing device running a multitasking OS like an $800 smartphone. Its a completely different market with completely different priorities and restraints, so ofcourse the hardware looks completely different, an off the shelf smartphone SOC would have been a horrible fit.
It doesn't HAVE to be using an 800mAh battery, it could be at least a little bigger. The one in DS Lite was 1000mAh and battery density has improved slightly since then. DSi did decrease it, but they probably decreased power consumption with a process bump (similar to PSP-2000)
I also don't think DS really lasted 10 hours typically, at least not with the backlight set to a normal level.
I think if power consumption is the number one concern then a better CPU could possibly improve it, not worsen it... I don't have actual numbers in front of me but my gut feeling tells me that a Cortex-A9 has better perf/Watt than an ARM11; I know that the Watt/MHz numbers are lower but that's 65nm vs 45nm; still, even if the consumption is much lower between the nodes, the perf/MHz is much better on Cortex-A9. And Nintendo would win a lot more recognition with it.
I also think having L2 cache could improve consumption; the cache itself uses a good amount of power, but it cuts away cycles spent where the CPU is doing nothing waiting on main memory, and it likely decreases consumption for RAM as well. Not that I'm saying Nintendo won't use L2, but I doubt it.
The impressive footage we've seen thus far is really all a credit to the GPU design and has little to do with the CPU choice. The CPU limitations will probably be more apparent in more subtle ways that are hard to see in videos, especially when most of them are with little going on outside of scripted content.
Seems to me that the best solution would have been a sort of "turbo" arrangement where developers have access to higher clocks if they put one of the cores to sleep. That way devs who aren't making good use of threading won't get hit as hard. But I'm still betting on Nintendo running their own code on one of the cores.
Also, $800 smart phone, seriously? Instead compare to iPod Touch which is $229 and still makes a profit, probably a bigger one than any Nintendo handheld would because Apple goes for huge margins and doesn't make as much in software sales (although they get a big cut from the App Store, Nintendo probably still makes a lot more from its tens of millions of first party sales and multitudes of licensing fees from third parties)
The extra 256 MB of RAM not included in the new iPod touch (and any extra battery capacity it lacks compared to iPhone 4) adds marginally to the bill of materials, so the price of the processing-related hardware that it takes to render iPhone 4's Epic Citadel was not outside of the budget of a dedicated gaming portable like the 3DS.
Epic lists shader tech used in Epic Citadel and Unreal Engine 3 for iOS at a page on their website, so some notion that the demo is not computationally intensive and its looks are mostly attributable to pre-baked texture work (as if many high quality texture layers were not extremely expensive on a hardware level anyway) is misguided.
Indeed it has! ...And it is one described by developers as "bizarre" in the way it is programmed - perhaps because it may have been cooked up in-house by Nintendo R&D, and not being experienced in 3D graphics hardware design... *shrug* I dunno.
Anyway, it's there, and decently fast for what it does too by accounts. It only does point sample texture filtering though (but at least seems to offer perspective correction), which may explain why you thought the DS doesn't have a GPU. All other consumer-level 3D accelerators have offered at least bilinear filtering, so it may be surprising to some when the DS did not.
Designing even a relatively basic GPU core like the DS's would be outside the expertise of Nintendo's hardware engineers, so they definitely brought in somebody's work.
Based on DS's rendering algorithm and the recently sparked possibilty that Stellar Semiconductor was not completely devoured by Broadcom for all these years, the DS might've gotten it's GPU from them -- yes, it's quite a stretch.
Is a CortexA9 design really realistic for a device that aims to ship ~5 million units that costs signfiicantly <$200 to build by year end? We still haven't seen a smartphone ship with a CortexA9 core yet and they ship in much lower numbers than the 3DS will and at much higher prices to boot. If the 3DS was a mid/late 2011 product, then it would be a none issue but its going to be hitting store shelves in one or more regions in just a couple of months.
My battery life figures were obviously ballpark estimates but they're inline with previous Nintendo hardware and Nintendo's rhetoric on the subject. I'd be incredibly surprised if it has more than a 1000mah battery and can't last a good 7/8 hours at reasonable brightness, and anyway you shake it, that's well beyond the battery performance of any smartphone running a high end 3D game and not something which can be so easily discounted when looking at 3DS specifications.
I've read all that and I've also read Capcom's article on the 3DS iteration of MT Framwork and despite the fact that Epic Citadel is a very impressive demo (I've never claimed otherwise) it definitely doesn't seem to be doing nearly as much in terms of shaders/lighting/shadows as Capcom's demo is. The quality of the art, lightmap and texture work are absolutely a huge part of what makes Citadel look so impressive imo, that's not to say it isn't using some nice shader effects as well, just not as many as high end 3DS titles are.
Processing cores like the A9 or an SGX take up square millimeters of space and cost literally just a few bucks in silicon, whether they're brand new or not. Their real cost to a mobile system is, of course, their power consumption currently.
I realise that but the A9 is still a relatively new CPU design and there's obviously good reasons why we still don't have any shipping smartphones using it yet. If phone manufacturers that refresh every year and often compete on specifications are unable to ship a few 100k units of A9 based hardware then surely its not a stretch to expect that Nintendo would have trouble getting ~5 million units using the same silicon into the marketby year end?
Does the A9 not cost more to license compared to an ARM11 at least? I'm genuinely ignorant about that and perhaps wrongly assumed that licensing a more modern ARM core was more expensive even if the cost of the extra silicon was completely negligible.
The licensing fee may be higher and the royalty rate per unit definitely would be, yet the point about A9 availability within the time frame of a 3DS launch definitely would've been the big obstacle.
I think the GPU was done by ARM, same as GBA's 2D renderer. Not sure where I saw the reference to this but I know I saw it somewhere.
A9 costs more but it's a pretty reasonable tradeoff. Chances are most of what I said about A8 would apply too, for what it's worth, A9 would just apply moreso.
As far as availability goes, I don't think volume really makes that much of a difference. The catching point is that most of the leading edge cell phone manufacturers have been relying on third party SoCs which adds an extra layer of time for production. Even Samsung, who is selling phones with their first party SoC, still are making said SoC available to the public and only offered phones after that point. What I'm getting at is that it's a different scenario when Nintendo has a chip manufactured specifically for them, and have much more specific and less complex needs than what a typical cell phone SoC these days provides.
It's pretty widely believed that Nintendo switched to DMP quite late in the game (as little as less than a year ago), so it's not as if they can't accommodate technology that's effectively new to them, without a long amount of transition time.
I spent more than a few moments closely observing the demo on an iPad - there it has *zero* dynamic per-pixel lighting. It's so apparent at places it's not even funny (diffuse highlights and shadows in disagreement with light sources, non-existent or blatantly pre-baked specular components, etc). It does seem to have a (rather shallow) parallax mapping effect on a portion of the surfaces though. Perhaps the demo does do everything the brochure claims on the iPhone4 (higher relative fillrate, etc). On the iPad though it's essentially a static world with two textures (albedo and GI map), with a subtle bloom filter as a post-effect.
I never heard that.. are you sure? That's truly a big waste of potential.
I'd say the most impressive performance features of the N8 (like editing 12MPix photos in realtime very fast) are due to those vector processors.
Of course that simply playing some sound files won't tax even a low-end CPU. But if a developer would like to do 3D positional audio with ~32 voices of ~128kbps MP3 quality and with some reverb effects to the boot... well, he won't.
Not unless there's some dedicated audio dsp hardware, like you said.
Sorry, didn't mean to. My point was just to show that there's already a real product in the market using a parallax barrier screen and that's a huge step from showing early prototypes in a tech conference.
The commercial success of some DS titles doesn't really belong to the discussion. I was talking about platform's success.
Check out the list of best-selling games for the DS:
Now take away all the imaginative low-budget, depth-less and simple games (all the brain-trainings, pet, puzzles and more). What would you get, in terms of console sales and market relevance? A portable Gamecube..
What I'm saying is that I don't think Nintendo will be able pull a stunt like that in 2011->2015.
In 2005 there was no unified gaming platform in cellphones (except for ngage, which was terribly executed). Now there iOS, Android and even Symbian^3+Qt.
Developing low cost and simple games to sell them in a cellphone platform is now very accessible for both big and indie developers.
Why would I carry a DS to play simple and fast games if I can do the exact same thing with a device that I have to carry with me all the time?
The only selling points that a 3DS could offer above smartphones with a 3D screen is more battery life, better controls and AAA games.
With such low-end computing capabilities, AAA games will be limited to weak AIs and scripted events. That could become very troublesome for the 3DS, especially if faced against a 3D PSP2 with much higher performance.
I wouldn't even go to A9s, even a single 600-720MHz Cortex A8 would make a big difference from 2*266MHz ARM11.
And there would definitely be no problem in availability for getting A8s ready for the 3DS launch.
In that case even if the difference is only a few $ then it can hardly go down as an insignificant cost. Even conservative estimates would put the 3DSes eventual LTD sales in the 100 million region, so a $3/$4 increase could easily end up costing Nintendo a cool half a billion $ when all is said and done. That's just not a cost you can easily write off unless you can genuinely justify a tangible advantage/benefit and I'm unsure Nintendo can.
That's cool and all but what 3DS title is really going to need those audio characteristics? Most gamers will be hearing the audio through the awfully tinny internal speakers or some cheap $5 stereo earphones, this isn't a device that's going to be hooked into a $5000+ home theatre. Perfect 3D positional audio seems a complete waste given that.
I never had or heard any complaints about the quality of ingame audio on the NDS and at the very least its going to be a significant step above that. The more audio a developer packs in, the more expensive the game cards they have to use will be as well (and 2GB is the current maximum even if they're willing to swallow significant fixed costs), so it has much bigger tradeoffs besides processing costs. It just seems like a case of seriously misplaced priorities to me. This isn't something that is all that crucial for the target market, so it makes perfect sense to compromise in this area, especially if it affords better Graphics hardware and increased battery life.
Again, designing low cost dedicated portable gaming hardware is all about making tradeoffs and compromises, its simply impossible to offer something that is all things to all men without compromising on the things that are utterly crucial to success (like a small BOM, small formfactor and excellent battery life).
Does a 720mhz A8 really have the same (or lower) power draw than dual 266mhz ARM11s? I'm somewhat doubtful and if it doesn't then it was never an option open to Nintendo. What about BC as well? Its quite possible that this dual ARM11 setup has been chosen as its allowed Nintendo to offer full BC without legacy hardware and if that is the case would a single A8 be able to offer this feature?
All of these potential alternatives have to be discussed under those qualifiers, truth is we'll probably never know whether Nintendo got it right but I think its naive to doubt that their engineers don't have some pretty solid justifications for going with the hardware they did. They've designed some utterly fantastic systems over the years, the GCN and NDS were both perfect examples of supremely efficient designs for their target market.
Not exactly, a S3D display is worthless without any worhtwile content that takes advantage of it, and that's one area where Nintendo are going to be several years ahead of anyone else. The device you linked doesn't even have a S3D camera, I honestly don't understand the practical use for the S3D on that particular device, its nothing more than a silly marketing gimmick without the content to back it up.
At launch the 3DS will have more S3D content than any other device on the market. It packs a stereo3D camera (a first for an affordable consumer device), will ahve dozens of S3D games available and even several S3D movies.
An Android or iOS device may be able to offer a S3D screen within the next 12 months but how long is it going to be before it has anything like the amount of S3D content that the 3DS will have at launch? I'd say 3 years would be a rather ambitious estimate.
This can't be serious, surely? The NDS has one of the most fantastic lineups in history, and has an immense amount of excellent content for just about every demographic, only the PS2 could ever really hope to compare. The third party "core" output from Japan is simply fantastic, and lightyears above any modern home console in terms of worthwile Japanese support. Heck, the JRPG genre is pretty much exclusive to the NDS these days (although the PSP is getting some decent games in this area recently, admittedly), what worthwile JRPGs did the GCN have!?
We're talking about the most successful gaming device in history here, I didn't see any mainline Dragon Quest games launching exclusively on the GCN.
Its as if you're suggesting these are meaningless side benefits, when in reality, time and again these have been proven to be the 3 of the most important, if not the most important, facets of what goes into making a successful portable gaming device. We've been here before with both the GB and NDS, uncompetitive graphics technology didn't seem to do the success of those two systems any harm at all.
This is forgetting the fact that the graphics in 3DS games compare much more favorably to their contemporaries than the graphics in GB and NDS games ever did. Heck, the single most impressive demonstration of mobile graphics technology atm is on the 3DS (RE:R) and even the ugly 3DS games are still lightyears ahead of your average iOS game which struggles to manage a stable 30fps with barely better than PS1 ear 3D. Epic Citadel is not the norm, its the exception and even then there's nothing in that demo outside of texture and rendering resolution that isn't being demonstrated in any number of 3DS titles.
The difference is measured in dimes, not dollars, between the royalty on a new and/or lower volume IP core and an old and/or high volume one. Indeed, the entire royalty per unit for a single core is under a buck.
Probably not - a better argument would be a dual Cortex-A8 at 266MHz not using more power than dual 266MHz ARM11s but offering superior performance. ARM doesn't actually offer dual Cortex-A8s and does offer MPCore ARM11s, but that didn't stop Nintendo from rolling out an ARM9 + ARM7 for DS. When you aren't sharing L2 there isn't that much to it, and I don't even expect the 3DS CPUs to be MPCore, just two ARM11s and without L2 cache. We'll see if I'm proven wrong; this is pretty bold speculation based more on gut feeling than anything, so don't hold me to it too much.
I don't see a 266MHz ARM11 being better at backwards compatibility for a 67MHz ARM9 than a Cortex-A8. All three have different timing characteristics and are for the most part ISA backwards compatible. I expected Nintendo to include a 67MHz ARM9 verbatim, which they still might be doing; one that's perhaps inaccessible during 3DS operation. That wouldn't take up that much die space.
I'm not going with the ends justifying the means on this one. Just because Nintendo has reigned supreme on gaming handhelds since their inception it doesn't justify every decision they've made. Okay, for Nintendo's short term bottom line it does, but as far as the userbase is concerned it's in our better interests for companies to be pushing technology forward and not just getting away with what they can get away with.
Cost and power consumption are good justifications for shooting lower on mobile devices, but IMO Nintendo shoots too low, because they can. And while other devices are pushing the technology forward but don't have the software to match Nintendo continues to hold things back. Gameboy Color is a perfect example of Nintendo doing the absolute bare minimum to generate fresh revenue. Things have gotten better since then. Have they gotten "good enough"? That much is debatable.
Yep, Nintendo has something here and it's pretty exciting.
Let's hope their CPU decisions don't end up crippling them.
There won't be much in the way of games, but I think S3D is going to hit in a much bigger way in terms of non-interactive content (movies) and that's going to be relevant on any device that can run it. And the Android phones will probably be able to run more thanks to having better hardware. Let's hope Nintendo is at least equipping this thing with some video decoders. Oh, and video encoders would be nice, for recording 3D movies.
Maybe Nintendo could sell the device for $199.99 instead of $199
NIntendo has a long history of squeezing every penny, especially in their handhelds. Their feature bullet lists always look aggressive, but everything they do include has been and is going to be offensively cheap. Microphone in the DS anyone? When I think about what they passed as audio DACs in the GBA I still want to punch the whole lot in their faces. Multiple times. Or how they saved literally half a penny per unit by removing S-Video capability from EU Wiis.
It's not surprising at all that they go cheap on the CPUs. They always have done so.
What's surprising to me (still) is that two CPU cores come out cheaper than one CPU core, especially at such low clock speeds. Maybe it really is about security, running one of them locked down for system libraries. Piracy has to be a big concern for them right now.
Two ARM11's at 266Mhz won't be cheaper then one ARM11 at 530Mhz would have been (these two CPU's are probably meant to run at 400Mhz + anyway), they'll be more expensive, however they'll use less power. I also don't think they'll use only one for gaming and the other for the system. I think we'll probably find out that there's a lower clocked ARM9 in their for those tasks,