3DS got two ARM11s

It really sounds unreasonable. I'd take dual 533MHz ARM11 easily, or even single ~500MHz Cortex A5 with NEON. Or at least having some dedicated hardware for vector processing, like Broadcom's BCM2727 in Nokia N8.
The vector processor in the BCM2727 isn't directly available to 3rd party developers, so how is that relevant? It makes more sense to compare it to the GPU and potential fixed-function accelerators in practice. The fact it's programmable is an (interesting) implementation detail.

With this kind of hardware, we just know the 3DS games are condemned to scripted physics, AI and even low-quality sound music/effects (please, no more midi sound, really.. it's enough) during their whole lifetime (until 2016?).
The first two are obvious and I said so myself, but I'm not certain about the last one. It's perfectly possible that there is dedicated hardware for audio, and even if there isn't you seem to be exaggerating the processing cost involved (it's just 20MHz per MP3 and you can do it in advance for the usual small sound effects).

The console is coming at the same time as dual-~1GHz A9 smartphones hit the market.. I just can't believe they think they can pull another "it's okay 'cause we're creative" on us, with 5 year-old hardware.
I certainly agree in a general sense, for what it's worth; it's mostly a disagreement about magnitude and important details.

When? How about 3 months ago?
And how about not removing the context of the quote? You said 'as soon as 3D displays become standard'. A single model is as far from standard as can be - while I agree OS support will probably come in less than 2 years, the number of features supported by various OSes that never became mainstream is a staggering number. This will nearly certainly become mainstream eventually, but my point is that if it doesn't within 3 years of the 3DS' release, then that's not a big deal for Nintendo - and I assume that's what Nintendo is thinking here too. But if it does happen faster as you seem to believe, then yes I agree it would be a substantial problem for them.

The DS' worldwide success is all about casual gaming.
There is a difference between casual gaming and casual games. The DS is all about what many would call casual games, but people who own one used them a lot more than cellphone gaming on average, so it's not casual gaming per-se. That's presumably the distinction MfA wanted to point out.
 
What sense does it make to instead take two at half the clock speed?
The advantage is power: the power consumption grows more than linearly with the frequency; also when you have two cores, it's easier to shut down one if you don't need it (though if Nintendo runs a kind of OS and/or libraries on one of the cores, then this won't apply).
 
I would go with a triple core ARM11 @ 450MHz (32KB instruction+32KB data L1 cache, 256 KB L2 cache/core), no GPU, MoSys 80MB 1T-SRAM, and bigger game cards.
 
I was expecting 533MHz ARM11 + 67MHz ARM9 and people were saying it was surely going to be faster... this I did not expect. Still unconfirmed, but damn.

I have a good feeling you won't be able to run user code on one of the CPUs, regardless of what they are, since that has been the model both Nintendo and Sony has used thus far.

For ARM11 @ 266MHz I expect 16/16KB L1 cache (with 32/32KB being a maybe) and 0KB of L2 cache. Was expecting CPUs that'd be a little bit more competitive against PSP than this....

How do you think dual ARM11@266Mhz + Vertex shader compares to dual MIPS32 R4000 @333Mhz though?

Also you shouldn't assume that their isn't a low clocked ARM9 or something similar in their as well, there's no reason to believe that IGN have all the details. When they got similar details about Wii they didn't find out about the little details (like Starlet ect).
 
Last edited by a moderator:
I would go with a triple core ARM11 @ 450MHz (32KB instruction+32KB data L1 cache, 256 KB L2 cache/core), no GPU, MoSys 80MB 1T-SRAM, and bigger game cards.

Why? Not only would that be more power hungry but it would be unable to come close to the graphical power of the current design.

Incidentally, I must be the only one here who didn't flinch at these specs. Its not as if the games we've seen so far will become less impressive because the specs aren't as glamorous as some would have hoped. Everything we've heard from developers suggests that the system's performance is quite surprising given the main specs, which is why I expected moderate looking specs.
 
Last edited by a moderator:
How do you think dual ARM11@266Mhz + Vertex shader compares to dual MIPS32 R4000 @333Mhz though?

That's not telling the entire story, PSP has both fixed function T&L hardware on the GPU and a powerful VFPU coprocessor on the CPU. I don't know how VFPU based shading compares to the vertex shading on DMP, which we're told is programmable, but I doubt either are going to be lacking. The vector computation capabilities on 3D consoles is rarely the bottleneck.

So the question should be limited to how I think ARM11@266MHz compares to MIPS32@333MHz, and that depends very heavily on the memory hierarchy. In particular, the size and associativity of L1 cache, the presence of L2 cache, and the relative speed of the bus/main RAM. In terms of internal processing throughput the two are probably pretty similar, with AFAIK ARM11 having better branch performance (when predicted) and Allegrex having lower load-use penalty.

Also you shouldn't assume that their isn't a low clocked ARM9 or something similar in their as well, there's no reason to believe that IGN have all the details. When they got similar details about Wii they didn't find out about the little details (like Starlet ect).

It's possible. It might help on power consumption when in DS mode.

Secessionist said:
I would go with a triple core ARM11 @ 450MHz (32KB instruction+32KB data L1 cache, 256 KB L2 cache/core), no GPU, MoSys 80MB 1T-SRAM, and bigger game cards.

I think you greatly overestimate the ability of software rendering. That configuration would struggle to even emulate DS graphics, which is a requirement for 3DS. You wouldn't be able to deliver anything close to what we've seen so far.
 
I would go with a triple core ARM11 @ 450MHz (32KB instruction+32KB data L1 cache, 256 KB L2 cache/core), no GPU, MoSys 80MB 1T-SRAM, and bigger game cards.
Sarcasm? I'm really not able to tell here.

I mean, something that would underperform (most generously put) in the graphics department, would very likely draw more power, and would cost an arm and a leg (80MB MoSys, ahoy) - how would that be a viable alternative to the rumored specs?
 
Exophase

I didn't know PSP's GPU had hardware T&L, I've never seen that mentioned anywhere (admittedly I've never been very interested in PSP). I thought all T&L was done using the VFPU and main CPU.
 
Last edited by a moderator:
I may be off on my architecture versions, but can't you run single ARM11 cores >500MHz easily? What sense does it make to instead take two at half the clock speed? It's twice the silicon at maybe, possibly, almost the same performance, if you run MP-optimized software.

edit: yep, ARM11 is supposed to do "from 350 MHz in small area designs up to 1 GHz in speed optimized designs in 45 and 65 nm". Or with a little love and tweaking, 750MHz at 90nm.

Well to me the implication is rather obvious - the choice of clocks was made based on desired power draw. It would seem likely that they would have no problem clocking their silicon higher, and higher clocks in this range would come for free. For free other than power.

Of course, Nintendo have the power draw data for the device in its entirety. It wouldn't make sense to be overly conservative on silicon clocks if it meant, for instance, that the display consumed 85% of the total power. If it were so, increasing clocks by 50% would still have had an acceptable impact on the overall power draw. I think it is safe to assume that Nintendo has chosen clocks to achieve a reasonably balanced device in terms of power draw. Which, in turn, in this case implies that the overall power draw of the device is quite low, and presumably that battery life will be quite good. Which, for a mobile device, is incredibly important.

To me, this line of reasoning makes a whole lot more sense than assuming that Nintendo engineers don't know what they're doing.
 
Sarcasm? I'm really not able to tell here.

I mean, something that would underperform (most generously put) in the graphics department, would very likely draw more power, and would cost an arm and a leg (80MB MoSys, ahoy) - how would that be a viable alternative to the rumored specs?

Go through his posting history. You'll see the man is incapable of sarcasm.
 
I would go with a triple core ARM11 @ 450MHz (32KB instruction+32KB data L1 cache, 256 KB L2 cache/core), no GPU, MoSys 80MB 1T-SRAM, and bigger game cards.

You DO realize the DS has a GPU, right? Why would you go from a design with a GPU to one without?
 
The advantage is power: the power consumption grows more than linearly with the frequency;
I thought any > linear power increase is not strictly due to the frequency increase, but due to using a higher voltage to attain the higher frequencies.
 
I think you greatly overestimate the ability of software rendering. That configuration would struggle to even emulate DS graphics, which is a requirement for 3DS. You wouldn't be able to deliver anything close to what we've seen so far.
I thought the DS used an ARM7 and 9 for graphics? It had a GPU?

I was also thinking mostly about 2d performance, so that may explain some of the stupidity in my post.
 
Back
Top