My point was that even if the innacuracy of the person measuring and removing bits Nintendo doesn't want were accounted for the extra RAM might take extra space which we are not really sure is accounted for, especially if we can't speculate on its size, not without knowing the exact customisations Nintendo's going with.
Oh, I agree, and I don't think there's any evidence for extra memory pools/increased cache/etc., but I also don't think there's really any evidence against it. It's unlikely to be something we could really determine without a die photo (and preferably another die photo of the TX1 to compare against).
Wii U's 35MB of embedded ram is also coming into play here, or am I missing something? I mean if Wii U is using all available embedded ram, doesn't that mean it can push more alpha textures than it's gpu has any right to? especially the 3MB of very fast t1sram from legacy gamecube on die?
It's also reckless to use this metric when we've seen Wii U with the same effects causing dips (alpha textures, fire, ect) and that is the comparison we are making.
I wasn't really comparing to Wii U (which would be difficult, as the different rasterisation paradigms and memory system setups would make differences very situational). I was responding (indirectly) to a post asking, to paraphrase, "If a game runs at 720p in handheld then why can't it run at 1080p docked?".
Depth of field effects are particularly bandwidth-intensive, but most importantly for a tile-based rendering GPU like Switch's, they can't be tiled. (Not even with the best possible implementation of Vulkan's render passes) Hence, frame rate drops which occur during DoF effects are a pretty good indication that a game is bandwidth limited.
I expect Nintendo to iron out the frame rate by the time the game releases, as their internal teams have a very good track record with frame rates (and the demo is likely based on a version of the game which is several months old by this point), but it's still a useful indicator at this point of where their technical challenges lie. I'd also expect them to improve some other graphical features in docked mode (e.g. draw distance), because if you're bandwidth-bound and have hardware idling you may as well use that up in some way.
I thought the same during the NX rumors, but the problem I have now is that the "Switch" name and logo start to fall apart when you don't revolve around the single core device like an SCD would.
Yeah, I had thought the same thing a while ago, and was actually considering making a thread on it, but to be honest I've come to the conclusion it's not that big of an issue. For now, Switch is a word which we associate with certain features of the hardware, but in a year or so it will just be a brand name we associate with Nintendo's device (or devices). When you hear Playstation you don't really think about the meaning of the words play or station, you just think of one of Sony's consoles. Ditto with pretty much any other brand name which is based on real words, it's only very early on in the brand's existence that you associate it with those words, and after that it's just a brand.
I also don't think this SCD would have to be that expensive sold standalone or optionally bundled at a discount, but I think it's important to note that the current dock despite doing essentially nothing has an inflated $90 price, possibly preparing consumers for whatever comes next.
Back to the name and logo, a dedicated home console would immediately lose "switching" to handheld, and is only left with "switching" the joy-cons between the grip and individual motion controllers, and that's if it didn't just come bundled with a Pro controller instead. Add in smaller handhelds and the joy-cons can't be attached, so either you lug them around separately or build some of the functions into the device but lose some game compatibility too which I doubt they want. The 3DS XL already sells better than the smaller unit anyways, right?
Yes, the 3DS XL sells better than the 3DS, but that's why they're starting off with a large handheld that doubles as a home console. That doesn't mean there's zero interest in anything smaller, though (and the hypothetical 5" Switch Pocket would be closer to the XL than the regular 3DS anyway). More importantly, though, a portable-only version of Switch with fixed controls could be a lot cheaper:
- No active cooling
- No dock (yes, it's overpriced at $90, but there's still some cost to it)
- No joy-con batteries
- No joy-con wireless
- No joy-con rails and physical interface
- Likely fewer linear actuators required for HD rumble
- Smaller screen
- Smaller battery (potentially)
- Less powerful charger
- No IR camera (maybe)
Having a cheaper option would be very valuable to Nintendo, and there are plenty of people who would prefer something a bit more portable than the current Switch anyway.
As for game compatibility, as far as we've seen so far the only game which wouldn't be compatible with a handheld with fixed controls is 1, 2, Switch, and it should be fairly obvious to customers that it requires joy-cons (and with Bluetooth it would in theory be possible to connect joy-cons to a "Switch Pocket", anyway). Even ARMS, a game that's very obviously built around motion controls, supports traditional controls as well, so it seems like Nintendo is at the very least keeping their options open for future form-factors.
A 1080p upgrade down the line with full backwards compatibility is what I would most expect, just like the DSi and NN3DS, with or without the SCD being true. Until this happens I would not expect any of the games being exclusive to better hardware ("some games will only play on the second device"), just like the PS4 Pro will not have exclusives.
I would have expected them to avoid the "some games only work on one of the Switch devices" too, by making a home console that plays the same games in higher resolutions. They could do so with a relatively affordable ~1.6TFlop machine, going by our current understanding of Switch's performance, and in theory get games running at up to 4K.
The problem is two-fold. Firstly, if they were actually to release a dock/SCD/standalone console using GP106-class hardware, then it would be far too powerful for this approach, even clocked down and with SMs disabled. A game running at native 4K on this kind of device would probably need to run at about 480p on Switch in portable mode, and games that run at sub-4K on it would be pushing 3DS-level resolutions in portable mode. Developers could certainly do more than just alter resolution, with improved effects, lighting, etc. to use up the extra horsepower, but with such a large performance gap this would be quite a burden on some developers.
The second issue is western third parties. If Nintendo wants to use a device like this to bring western AAA games onto their hardware, then forcing them to also bring their games to the portable would only hamper this effort, as with a couple of exceptions western publishers/developers don't seem to have any interest in supporting the current hardware. And if it is to be based on GP106, then you'd be going from saying "hey developers, make games for this console, it'll easily be able to handle any of your games with little effort" to saying "hey developers, make games for this console which can handle easy ports, but you also have to put a load of effort into making them work on much less powerful hardware too!".
I meant more "likely". Anyway i'll repost
this quote from Manfred Linzner about Wii U memory bandwidth, i still remember people saying that it was bandwidth starved "cuz 12.8GB/s", and none of them pointed at the CPU's rather large cache or those 32MB of eDRAM being part of a more complex but balanced memory subsystem, not even despite the fact that literally 0 devs ever complained about bandwidth despite shitting on the CPU at every turn. Now you're saying there are other possibilities like having more external bandwidth on Switch, which certainly makes more sense to me than "25GB/s period". Still I don't see any major challenge in using a 128bit bus in a device this size, and it's certainly not a cheap one either.
Hopefully marcan will test this d1 if we don't get other leaks.
We don't really need marcan or any kind of hacking to figure out the memory bandwidth, it'll just be a matter of checking the codes on the RAM modules once there's a teardown.
You're right in that Wii U wasn't bandwidth starved by the 12.8GB/s of DDR3 memory bandwidth by virtue of having an eDRAM pool with (I believe) about 70GB/s of bandwidth where most buffer accesses would go, but in a certain sense the same is true of Switch. There may be 25/50GB/s of main memory bandwidth, but like Wii U, most buffer accesses won't hit that, but will rather stay on the GPU's L2 (so long as they're tiled). That L2 bandwidth is likely to be pretty high, potentially higher than Wii U's eDRAM.
The GM107's L2 cache reportedly has a bandwidth of 512 bytes per clock, and even if Switch's is only one quarter that it would put it at almost 100GB/s.
On a related note, Switch should also have a quite substantial pixel fill-rate advantage over Wii U. If they keep the 16 ROP configuration from TX1 then we're looking at 12.3 GP/s, compared to a reported 4.4 GP/s for Wii U.