Next Nintendo handheld

Here we're talking about an entire CPU subsystem that costs ~$1 to make.

That doesn't mean Nintendo will go for it. Wii showed that they will go for something surprisingly weak for reasons that have nothing to do with fabrication cost.
 
I meant to say/joke that depending on the timeframe and the budget, nintendo can ship a NDS2 with performance near the wii

...and what if they use a low power version of the yet low power cpu in the wii?
i know that it's a simple G3 derived cpu from ibm that everyone can buy (well... kind of...)
Considering cost/size/power/thermal budget, can be a solution if they go from the actual 90nm to low volt lower clock 45/32nm soi hightK?
:p
 
Uhh how do you mean? Because the gba is more powerfull than the snes and the DS is more powerfull than the n64 I think.

The GBA didn't come out around SNES launch, and the DS didn't come out during N64 era either. Now the DS 2 or whatever it'll be called will likely be out within the Wii's lifetime.
 
Benchmarks seem to indicate a clock speed nearer to 100 MHz for the MBX Lite + VGP Lite in the iPhone.

The half resolution display compared to Dreamcast would make iPhone's fill rate roughly equal to the Dreamcast's on the relative scale.
 
OMAP2420 development boards clocked their graphics parts at about 50 MHz, and the iPhone scores similarly to OMAP2420 devices in GLBenchmark.
 
OMAP2420 development boards clocked their graphics parts at about 50 MHz, and the iPhone scores similarly to OMAP2420 devices in GLBenchmark.

I am sorry, but I miss how this makes the iPhone's MBX Lite clock at nearly 100 MHz unless the definition of similarly is stretched really thin ;). Typo?

Edit: unless the OMAP2420 devices have more pixel pipelines (thus higher fill-rate), for example 2x the pipelines an MBX Lite has, and still the iPhone's MBX lite scores very similarly suggesting that it makes up for the lack of pixel pipeline with a higher clock-speed.

Right ?
 
Last edited by a moderator:
Right.

OMAP2420 uses the full MBX, not the Lite pipeline rated at 2 clocks per pixel.

Actually, after relooking at the specific measurements taken by GLBenchmark, I don't think they necessarily indicate anything about a clock speed difference between the iPhone and OMAP2420 devices.
 
Right.

OMAP2420 uses the full MBX, not the Lite pipeline rated at 2 clocks per pixel.

Actually, after relooking at the specific measurements taken by GLBenchmark, I don't think they necessarily indicate anything about a clock speed difference between the iPhone and OMAP2420 devices.

The clock speeds of the iphone's mbx lite and omap2420 mbx seem to be the same (50MHz) judging by the benchmark results. Omap2420's are giving fill rates ranging from 30k-58k pixels, while iphone is only giving 24k-26k pixels. In fact, it seems that iphone's driver for the GPU are more stable since the fill rate is closer to the clock speed than omap's.
 
So wait, the iPhone vaunted to be more powerful then NDS and PSP combined has a lower fillrate then NDS alone? :oops:

I.S.T. said:
Isn't the DS a fairly... odd design?
Yea that's what I meant - lots of paralel processors clustered together, PSP hw layout is pretty simplistic in comparison.
 
So wait, the iPhone vaunted to be more powerful then NDS and PSP combined has a lower fillrate then NDS alone? :oops:


Yea that's what I meant - lots of paralel processors clustered together, PSP hw layout is pretty simplistic in comparison.

I think 25k pixels is the raw fillrate. Effective fillrate should be much higher. Anyways, even if the iphone has a lower fillrate than NDS, at least it has a much higher polygon throughput. I don't think iPhone is more powerful than the PSP even alone, but overall it should have better image quality (not draw distance) given the free bilinear filtering and FSAA.
 
there were a few mistakes in my previous posts. Those fill rate counts should be in mpixels not kpixels. Sorry for the confusion...
 
knux said:
I think 25M pixels is the raw fillrate. Effective fillrate should be much higher.
NDS is a scanline renderer - which puts them roughly on 'even' terms regarding overdraw.
Polygon throughput is a very different story though, yea.
 
Last edited by a moderator:
The abstractedness of OpenGL ES in those benchmarks also prevents the approach of the processors' full performance.
 
One question about the actual NDS, of around how many transistors the NDS is made off (or cumulated die size at a given process)?
I tried to find out on ARM site or wiki and didn't find enought datas to make an estimate.
 
Back
Top