Just FYI, the 3DS version of download play doesn't actually run from RAM. The download play title actually gets temporarily installed to system memory (
source). I suspect Switch will be similar if it supports the feature.
That's interesting. I suppose it makes sense to store the most recently played download-play game, as it saves people from having to download it again, and users are unlikely to swap back and forth too often between download play games. I wonder if part of Switch's reserved flash memory is set aside for this.
Not sure if this has been asked already, but what can we discern about the Switch's GPU/CPU etc from a tear down?
There are a few different "degrees" of tear-down that could tell us different things. The first is the only one that's actually a tear-down, the other two are die analysis.
- Standard Tear-Down - A disassembly of all components, revealing lots of ICs and product codes. iFixit will have one of these (usually only a few days after launch), and they do a very good job, with high quality photos of every component. The most immediate info we can gain from this is the RAM quantity, bus width and speed, from reading the product code on the RAM module(s). We'll also get to see how large the main SoC die is, but not much else about it. We'll also see some other interesting stuff, such as how the cooling system works and just generally what all the insides of the Switch (and controllers) look like.
- SoC Die Photo - This would either require Chipworks, or an intrepid individual who has a good camera and macro lens and is willing to spend hours carefully sanding the die down. This would likely tell us the GPU configuration (i.e. number of SMs) and the number of CPU cores, and possibly if there are any additional large memory pools on-die, as with Wii U and 3DS. It's unlikely that we would be able to tell exactly what CPU cores are used, or if there are any changes to the GPU. This would be most useful if we also had a TX1 die photo to compare against (which, afaik, we don't), as it would allow us to quickly identify the differences, and possibly alert us to any changes to the CPU, GPU or the memory subsystems, or anything else they might have changed. This won't, however, tell us what manufacturing process is being used (i.e. 20nm/16nm), as both have almost identical transistor densities. For that we need:
- SoC Cross-Section Photo - This is beyond the realm of home tinkerers, and would require someone with a scanning electron microscope (i.e. Chipworks) to cut the die down the middle and then analyse the cross-section to determine whether it's 20nm or 16nm by checking for finfet gates.
It's a shame that they didn't go any further (or didn't show any photos of it, if they did). It just seems like a tease to leave it at that.
That said, it is a bit interesting that the fan is so close to the vent on top of the Switch, I would have assumed more space there to accommodate cooling fins.
I don't know if what they've said is true or not, but they have ONE major advantage that we don't...they actually have the system and we can see that they opened it up. We don't know how far they went from there, either. Odds are they went further, but didn't post any pictures. What they found, we don't know. Truthfully, I'm on the side of their claim being BS, but they have the machine in front of them and we don't....
It's worth keeping in mind that, prior to their discussion of specs, they wrote this (Google translated):
However, documents that have been popped up in an Internet forum already provide numerous technical basic data. Whether the documents are authentic can be difficult to determine until the precious goods are unscrewed. If the information proves to be correct, the Switch will sleep under the hood of the Switch among other things:
The words "internet forum" link to NeoGAF (although not this thread, sadly, as that would have been pretty amusing). They're getting their info from us, so I definitely wouldn't read too much info into "X2" or "GTX1050".