Nintendo 3DS hardware thread

Unless it uses lenses to focus the light, a parallax barrier does cause dimming. After all, you're blocking half the pixels for each eye.

 

The parallax barrier is actually powered as well. I didn't think this was the case, but tear-downs have shown that it uses monochromatic liquid crystal strips that activate when you turn on 3D mode. But the second screen and the brightness issues are much larger power drains.

 

Yes, my bad. For some reason I thought 3DS was using a lenticular lens despite using the correct name for its 3D tech! Thus the same screen brightness as a normal screen would need a brighter backlight, which will increase power draw.

When 3DS's 3D is set to off, have we ahd it confirmed if it just doubles pixels or if it actually renders the second half of the resolution for higher IQ?

 

Last time I heard it just doubles pixels. And that's also what it looks like when you turn it on or off (have tried it several times in store). I think 3DS game footage also confirms it.

 

I agree. But on the other hand, I'm getting the impression it may sell bucketloads just the same.

 

Doubles pixels to 800*240 irc.
Also, the battery life with stereo 3D off is substantially longer (25% longer), which should prove that the extra power required for backlighting is the culprit.

That doesn't excuse the fact that they should've used a much larger battery from the beginning, though.

 

Anyway.. any info on the VFP co processors? Are these units that are in any ARM11 chip or are they optional and what kind of performance advantage might they have?

Anyway, nice to see the GPU is 268Mhz, considering some on here assumed the clock speed would be 133Mhz. Also interesting to see that its got 6MB FCRAM VRAM, I wonder what the bandwidth is and also what the bandwidth is to the two 64MB FCRAM modules.

 

Well, to be fair, the common assumption for the ARM11 clock speed was 266MHZ.

 

I was talking about the GPU, not the CPU.

A lot of people (not just on here of course) assumed Nintendo would take the standard 200Mhz PICA200 chip and reduce its speed to half that of the systems CPU.

 

You can read all about ARM11 VFP on infocenter.arm.com. It's pretty much just a standard single-issue scalar floating point. It does have a cumbersome "vector" mode (hence its name) but it's not like SIMD, it just sequences operations. And it has been removed from newer ARM processors.

Without them there'd simply be no floating point.

The CPU is basically comparable to the original iPhone from way back in 2007. Only 2/3rds the clock speed, but at least there are two of them. Only who knows what Nintendo is running on one of the ARM cores, which originally was prohibited from running user code at all.

 

Is there some reason to avoid something like a Cortex A8 w/ Neon? Or two.

 

Only that Cortex-A9 w/NEON is probably a better option in most regards. And you can't multicore Cortex-A8s, at least not with cache coherency. That was probably the real reason they went with ARM11. Cortex-A9 may have been too unavailable or too new when they started designing.

Although, for a Nintendo handheld they probably could get away with not having the separate cores cache coherent. Worked fine on DS, even if that never allowed user code on the other core..

 

The most shocking part is probably the clock speed. I guess a dual ARM11MPcore could be considered acceptable at ~650MHz, or even 536MHz (2*268, since Nintendo is always so obsessed with synchronous operation).

Nonetheless, I doubt a 536MHz Cortex A8 LP for applications + a 268MHz ARM9 for I/O tasks (Starlet style) would take significantly more space, or that it would consume significantly more battery.
I think greedyness is was the main factor for choosing such low-end hardware, to be honest.

 

There seems to be a few VFP's (some made for later processors like A8/A9), I assume the 3DS is using VFPv2?

Also seems they are optional, what kind of throughput do they have?

 

Playing the Devils advocate here, just why should nintendo have gone with higher power hardware? What would it accomplish?

It is clear that they envisioned the 3DS as having its own unique games library, apart from maintaining backwards compatibility with the DS (which at the time of the design was the largest mobile games platform in existance). Of course, the 3DS was and is the only autostereoscopic 3D device around, reinforcing the message of a games library that is unique to the device. So ease of porting titles to the 3DS wasn't something that Nintendo needed to concern themselves with. They designed to low cost, the only concession being the 3D-display, which appears to take a fair chunk of the budget for the hardware. And it has already paid off, since it allowed them to lower their prices when it became clear that they had misread the market, and they now seem to enjoy quite reasonable sales of the device, at least as it stands. Without resorting to $100 memory cards to try to recoup costs.

It makes some sense.

Technology enthusiasts generally feel that more is better. Some here even felt that it was Nintendos duty as market leaders to push the technology envelope! But technology is a means to an end, and without pointing out how Nintendo, and the totality of their market, would have benefitted overall from more powerful hardware, I can't help feeling that people are promoting more powerful hardware for it's own sake. Which is a valid hardware enthusiast standpoint, but then you also have to recognize that hardware entusiasts aren't Nintendos target demographic with the 3DS, apart from the autostereoscopic display.

 

ARM11s that have VFP have an implementation called VFP11, that is VFPv2. I don't think VFP is optional for ARM11 in exactly the sense you may be thinking, it's more like there are different ARM11 models and some have VFP while others don't. You'd think 3DS would be using ARM11 MPCore, which has VFP.

VFP11 is single-issue over three pipelines: FMAC, FDIV/FSQRT, and load/store. They can run in parallel and complete out of order. It's single issue, but since you can sequence vector operations it's possible to overlap them. Useful if you're calculating reciprocals in parallel with doing matrix multiplies, and useful for getting stuff into registers without slowing you down.

 

Thanks for the info

 

I found this: http://www.dmprof.com/e_news/e_event/2540/

DMP exhibited a 3ds at TSMC booth !! so i'll bet on a 65nm process node.

 

DMP is offering soft IP via TSMC. DMP is showing off 3DS to demonstrate their IP capabilities. This doesn't mean 3DS was fabbed at TSMC. 3DS's SoC allegedly has 6MB of embedded FCRAM from Fujitsu. It would therefore seem to make sense that the SoC is manufactured by Fujitsu, but iSuppli apparently thinks it's manufactured by Sharp: http://www.eetimes.com/electronics-news/4214569/Teardowns-find-Fujitsu-FCRAM-in-Nintendo-3DS

 

Is it confirmed that the edram in the 3DS's soc is FCRAM ? It could be 1TSRAM and Mosys is offering soft IP via TSMC. Why will you show a product that is not yours !!