A Nintendo Switch has been taken apart

That seems to be causing a few issues in Breath of the Wild and i wonder if that's the reason why they kept the 900p resolution instead of going with 1080p, but then again MK8 has DoF as well in replay mode and it's a 1080p game on Switch.

I wonder if they managed to mitigate the issue for the final version of the game, and if with more time in the oven (or if the game was made with the Switch in mind from the scratch) they would've managed to found some workarounds for that and boost it to 1080p.

I wouldn't necessarily say that DoF blurring on its own would prevent a game from hitting 1080p/60fps, but that on a tile-based renderer you would generally want to minimise the number of render passes which require a full-resolution buffer to be fed out to main memory and then read back in again. A DoF pass would be one of these, but it's possible BotW has more of them. I've tried my best to maintain a media blackout of the game since its first announcement, though, so it's pretty difficult for me to speculate what they may be.

Alternatively it's possible that it's simply a production decision made by the BotW team. They may have just preferred to add graphical effects, etc., in docked mode rather than push the resolution all the way up to 1080p.

Makes sense. I wonder what kind of bandwidth can we expect from the eventual SRAM.

A general rule of on-die SRAM would be that it will give you "enough" bandwidth for whatever it's designed for. My guess is that if there's any extra memory on there it's in the form of increased GPU L2, as it's a relatively simple change which would allow them to use larger tiles, or more render targets per tile, or perhaps just to free more space for texture caching and so forth.

Didn't we see a post not long ago from a Ubi developer saying that 25.6GB/s would indeed be a huge bottleneck even accounting for TBR? Specifically because of these types of post processing effects like depth of field you mention? It could be that MK8 and RMX are able to run at 1080p/60fps because there is a large on-die cache or something.

Regarding the DoF issue in Zelda mentioned by Digital Foundry, I believe that was specifically for the demo version and didn't even occur consistently. So it might be something we don't see in the retail version which would indicate the effective bandwidth is higher than 25.6GB/s right?

Do you have a link for the Ubisoft comment? It's not something I remember (although I may have simply missed it at the time).

Regarding the "large on-die cache", that's exactly how TBR already works (the big increase in L2 cache of Maxwell over Kepler was a clue for this, even before anyone figured out they moved to TBR). A larger cache would allow the GPU to use larger tiles, which should improve efficiency somewhat, but a step-change in performance would only occur once the cache is large enough to hold the entire framebuffer, which would likely take up almost all of the die area of the chip pictured.

There is still the difference in efficiency between the more modern Maxwell architecture and the older R7000 (I think) architecture in PS4/XB1, which should mean the Maxwell flops do perform better than R7000 flops. Maybe not by 40% or 30% but there should be some gains purely from more modern architecture.

The other thing to mention is, if the NVN API is as good as everyone is saying it is, couldn't that be a similar advantage to the one Nvidia hardware has in a PC environment? Again, maybe not to the same extent, but still potentially an effective flop advantage. I think a game like Snake Pass (which struggles to reach 60fps on PS4) running at 1080p/30fps locked on Switch shows that the Switch will surely be punching above its weight when it comes to pure raw numbers.

I would guess that Snake Pass has some graphical effects toned down or removed from PS4 to Switch, and I also wouldn't read too much into pre-release frame rates. It'll be interesting to see a proper comparison of the two versions once the game releases, though, hopefully Digital Foundry will do an article on it.
 
Some of you guys need to stop with the "I read before..." and "Didn't we see..." posts if you don't have links to back it up. This is how these rumors get started

It exists in a beyond3D thread. I can confirm that much, it is a ~80 page thread though? so I don't really feel like tracking it down. He wasn't directly working on the Switch and it was just his assumption, other Beyond3D users did disagree with him, so it's not a fact.

EDIT: Found it; https://forum.beyond3d.com/threads/...ulation-discussion.59709/page-58#post-1959110
 
for positive spin, how about the lack of games gives you a lower barrier to entry.
- Don't have the money for lots of launch games? No problem.
- Don't have the time to play lots of launch games? No problem.

Its a bit twisted but as I assume I'll buy in for Mario Odessey anyway, I might as well get it now and then deal with games one at a time, rather than suddenly having a bunch that I wont have time to play at Christmas
Twisted, yes ;)
Still all they needed was 1 or 2 more decent launch games to make it worth while, hope Zelda has a lot of replay value because even after launch the games look to few and far between.
 
Do you have a link for the Ubisoft comment? It's not something I remember (although I may have simply missed it at the time).

Regarding the "large on-die cache", that's exactly how TBR already works (the big increase in L2 cache of Maxwell over Kepler was a clue for this, even before anyone figured out they moved to TBR). A larger cache would allow the GPU to use larger tiles, which should improve efficiency somewhat, but a step-change in performance would only occur once the cache is large enough to hold the entire framebuffer, which would likely take up almost all of the die area of the chip pictured.

Found it- https://forum.beyond3d.com/threads/...ulation-discussion.59709/page-58#post-1959110

Turns out it's sebbbi again.

Some of you guys need to stop with the "I read before..." and "Didn't we see..." posts if you don't have links to back it up. This is how these rumors get started

It exists in a beyond3D thread. I can confirm that much, it is a ~80 page thread though? so I don't really feel like tracking it down. He wasn't directly working on the Switch and it was just his assumption, other Beyond3D users did disagree with him, so it's not a fact.

I'm not saying what this user posted is fact as we quickly determined he had no info on the actual Switch hardware, but his perspective helped us understand the limitations of TBR when it comes to helping with bandwidth issues.

And it took me a while to find this because it was in that huge locked thread that started back in October. I should have found the source before I brought it up, my apologies.
 
I still find it mind blowing that a technology company will not list its tech specs and we have to guess or take it apart to find out what is in it.

Can you imagine someone like Apple releasing a new iPhone and not telling anyone what is inside and just expecting people to buy it? What about TVs or virtually any other electronics sold? Specs don't matter to everyone but they should be out there for the people that care.
 
I still find it mind blowing that a technology company will not list its tech specs and we have to guess or take it apart to find out what is in it.

Can you imagine someone like Apple releasing a new iPhone and not telling anyone what is inside and just expecting people to buy it? What about TVs or virtually any other electronics sold? Specs don't matter to everyone but they should be out there for the people that care.

Arent apple a bit coy when it comes to specs too?
 
Found it- https://forum.beyond3d.com/threads/...ulation-discussion.59709/page-58#post-1959110

Turns out it's sebbbi again.





I'm not saying what this user posted is fact as we quickly determined he had no info on the actual Switch hardware, but his perspective helped us understand the limitations of TBR when it comes to helping with bandwidth issues.

And it took me a while to find this because it was in that huge locked thread that started back in October. I should have found the source before I brought it up, my apologies.

Beyond3D is an enthusiast forum just like GAF. Unless I missed it, I don't see anything mentioning Ubisoft in that link. It's another perspective though
 
I still find it mind blowing that a technology company will not list its tech specs and we have to guess or take it apart to find out what is in it.

Can you imagine someone like Apple releasing a new iPhone and not telling anyone what is inside and just expecting people to buy it? What about TVs or virtually any other electronics sold? Specs don't matter to everyone but they should be out there for the people that care.

Truth to be told Apple doesn't give that much details aside from hyping: New AX chip!! More powerful than our AX-1!! With our new cores!! and at the same time not telling anything about memory or how they stack up against the competition.
 
Finally some sense. For some weird reason people still think that when presented with 2 options, Nintendo would go for the more modern, higher end version. They won't. They will try and see if they can build an acceptable system with the lowest options.
You aren't going to see high end stuff from Nintendo anymore (not since when,the 64?) But they didn't intentionally chose the poorest parts for the Switch,that is not logical this time.
 
Beyond3D is an enthusiast forum just like GAF. Unless I missed it, I don't see anything mentioning Ubisoft in that link. It's another perspective though
Sebbbi is the former Ubisoft developer we were talking about, he posts there too.

You aren't going to see high end stuff from Nintendo anymore (not since when,the 64?) But they didn't intentionally chose the poorest parts for the Switch,that is not logical this time.

Gamecube doesn't exist, Switch isn't high end portable, etc etc
 
Arent apple a bit coy when it comes to specs too?

You can click on the tech specs on the product page and it will tell you in the documentation with the phone itself. They may not be in your face with the tech specs but they don't hide them either. Meanwhile Nintendo has went on record saying that tech specs don't matter. If they don't matter, why do you make new consoles? Why aren't we still playing the NES if all that matters is "fun"?

I love Nintendo but they are a frustrating company to support and defend.
 
I never really saw the estimations for the battery, being 6.4watts for 2.5 hours.

If the rest of the device is say 2.4watts, that SoC should be able to draw up to 4watts, I don't see how the eurogamer clocks could hit that even on 20nm, my calculation puts it under 2.4watts just for the SoC alone, and that isn't an Oled, so there is no way the rest of the system is drawing 4watts.

That 2.5hours is going to be a pretty heavy use case scenario... If this did move to 16nm there is absolutely no way that SoC has those clocks IMO.

Edit: Seems the A53 could be there, if the devkits hadn't added the interconnect yet, devkits would have to use a core for the OS, if Final hardware changed this, then they could move the OS to a couple A53 cores, and even possibly offer up the other 2 for games.
 
but has he actually worked with the switch..... if not he needs to be quiet. If the developers that are working and creating games on it are happy.... then his words mean nothing to me

I'm not bringing this up to say anything specific about the Switch, it was just in a conversation I was having with Thraktor about the limitations caused by a RAM bandwidth being so low (25.6GB/s) and how that apparently can't be remedied just by using Nvidia's TBR.

We know sebbbi hasn't developed for the Switch but we can still use his expert knowledge of developing in general to make observations about what the Switch can and can't do with various rumored hardware. I personally think Nintendo would customize their RAM situation more in order to avoid the bottleneck mentioned by sebbbi, which is why I brought up his post.

I never really saw the estimations for the battery, being 6.4watts for 2.5 hours.

If the rest of the device is say 2.4watts, that SoC should be able to draw up to 4watts, I don't see how the eurogamer clocks could hit that even on 20nm, my calculation puts it under 2.4watts just for the SoC alone, and that isn't an Oled, so there is no way the rest of the system is drawing 4watts.

That 2.5hours is going to be a pretty heavy use case scenario... If this did move to 16nm there is absolutely no way that SoC has those clocks IMO.

Edit: Seems the A53 could be there, if the devkits hadn't added the interconnect yet, devkits would have to use a core for the OS, if Final hardware changed this, then they could move the OS to a couple A53 cores, and even possibly offer up the other 2 for games.

The July devkit docs showed that the A53s were removed. Those docs show us what is on the hardware in the left column and what is usable by the devs in the right, so the A53s not showing up in the left would seem to confirm they aren't there at all.

And anyway, doesn't the Jetson Tegra X1 devkit have the A53s removed too?
 
I wouldn't necessarily say that DoF blurring on its own would prevent a game from hitting 1080p/60fps, but that on a tile-based renderer you would generally want to minimise the number of render passes which require a full-resolution buffer to be fed out to main memory and then read back in again. A DoF pass would be one of these, but it's possible BotW has more of them. I've tried my best to maintain a media blackout of the game since its first announcement, though, so it's pretty difficult for me to speculate what they may be.

Alternatively it's possible that it's simply a production decision made by the BotW team. They may have just preferred to add graphical effects, etc., in docked mode rather than push the resolution all the way up to 1080p.



A general rule of on-die SRAM would be that it will give you "enough" bandwidth for whatever it's designed for. My guess is that if there's any extra memory on there it's in the form of increased GPU L2, as it's a relatively simple change which would allow them to use larger tiles, or more render targets per tile, or perhaps just to free more space for texture caching and so forth.
I agree, just wondering what the numbers (both in terms of size and bandwidth) would be.

I never really saw the estimations for the battery, being 6.4watts for 2.5 hours.

If the rest of the device is say 2.4watts, that SoC should be able to draw up to 4watts, I don't see how the eurogamer clocks could hit that even on 20nm, my calculation puts it under 2.4watts just for the SoC alone, and that isn't an Oled, so there is no way the rest of the system is drawing 4watts.

That 2.5hours is going to be a pretty heavy use case scenario... If this did move to 16nm there is absolutely no way that SoC has those clocks IMO.

Edit: Seems the A53 could be there, if the devkits hadn't added the interconnect yet, devkits would have to use a core for the OS, if Final hardware changed this, then they could move the OS to a couple A53 cores, and even possibly offer up the other 2 for games.

Like suggested by other users, i think they might have removed the A53 cores to accomodate a pool of SRAM without making the SoC too big, which makes sense. I still wish they used 4 A72 cores for games and a couple of A35 to run the OS. The 128bit bus thing is just stupid to me, as even the 3DS had it but who knows, maybe they thought it wasn't necessary.
 
Edit: Seems the A53 could be there, if the devkits hadn't added the interconnect yet, devkits would have to use a core for the OS, if Final hardware changed this, then they could move the OS to a couple A53 cores, and even possibly offer up the other 2 for games.

Not impossible, but they'd have to redesign the CCI, as the TX1 itself only supports the primitive either/or cluster switching. Can only have the high or low performance cluster on at once, so the OS couldn't run on on the A53s concurrently with the A57s running a game.

Not insubstantial work to update that, but not impossible.
 
You can click on the tech specs on the product page and it will tell you in the documentation with the phone itself. They may not be in your face with the tech specs but they don't hide them either. Meanwhile Nintendo has went on record saying that tech specs don't matter. If they don't matter, why do you make new consoles? Why aren't we still playing the NES if all that matters is "fun"?

I love Nintendo but they are a frustrating company to support and defend.
Well,you are taking "specs don't matter" too literal. Nintendo won't talk specs since Wii because they know people will use them against them and they rather focus on something else,bad PR.
An example would be 3DS versus Vita were despite 3DS being much weaker it offered a strong environment from first and third party developers without expensive dev budget

From what we know,Switch is a good handheld by Nvidia,however it Will be compared to PS4(pro) that is a battle they can't win.
 
Like suggested by other users, i think they might have removed the A53 cores to accomodate a pool of SRAM without making the SoC too big, which makes sense. I still wish they used 4 A72 cores for games and a couple of A35 to run the OS. The 128bit bus thing is just stupid to me, as even the 3DS had it but who knows, maybe they thought it wasn't necessary.

We still have no clue what's in the SoC, so I wouldn't count out A72s yet, though I would imagine # of cores for developers would have remained consistent throughout the devkits so we can reasonably assume 3 cores for devs.

Yeah but we don't know what Nintendo has done to overcome that "potential issue." that what I meant is I get he is a developer and knows a lot. but he should a least say along with his remarks that Nintendo and NVidia could have made changes to get around the potential bottleneck.

In almost all of his posts there he says something like "if the Eurogamer article is 100% right about the specs, this is what problems there may be:"

So I think he does a good job of contextualizing his comments.
 
I'm not bringing this up to say anything specific about the Switch, it was just in a conversation I was having with Thraktor about the limitations caused by a RAM bandwidth being so low (25.6GB/s) and how that apparently can't be remedied just by using Nvidia's TBR.

We know sebbbi hasn't developed for the Switch but we can still use his expert knowledge of developing in general to make observations about what the Switch can and can't do with various rumored hardware. I personally think Nintendo would customize their RAM situation more in order to avoid the bottleneck mentioned by sebbbi, which is why I brought up his post.



The July devkit docs showed that the A53s were removed. Those docs show us what is on the hardware in the left column and what is usable by the devs in the right, so the A53s not showing up in the left would seem to confirm they aren't there at all.

And anyway, doesn't the Jetson Tegra X1 devkit have the A53s removed too?

^that could be possible, I'd assume they would just be disabled otherwise.
 
The July devkit docs showed that the A53s were removed. Those docs show us what is on the hardware in the left column and what is usable by the devs in the right, so the A53s not showing up in the left would seem to confirm they aren't there at all.

Surely that would only show that none of them are usable by developers. After all if none are usable then there is no point listing them at all.
 
Surely that would only show that none of them are usable by developers. After all if none are usable then there is no point listing them at all.

Huh? The normal Tegra X1 has 4x A53s and 4x A57s, so having the July spec sheet show (on the hardware side, not just the usable side) only 4x A57s should indicate that there are no A53s present in the hardware, right?

So it's possible Switch is using A72s? And what are the chances that Nintendo included more than 4 GB of RAM?

A72 is possible and would indicate a 16nm process, but there's no way to know until someone does a cross section I believe. RAM wise I think we're pretty much set at 4GB, as the modules in the OP have been 99% identified as 2GB each. Though we aren't 100% sure the pictures in the OP represent the retail hardware, although it's a reasonable assumption. I'd say 90% chance it's 4GB.
 
Huh? The normal Tegra X1 has 4x A53s and 4x A57s, so having the July spec sheet show (on the hardware side, not just the usable side) only 4x A57s should indicate that there are no A53s present in the hardware, right?



A72 is possible and would indicate a 16nm process, but there's no way to know until someone does a cross section I believe. RAM wise I think we're pretty much set at 4GB, as the modules in the OP have been 99% identified as 2GB each. Though we aren't 100% sure the pictures in the OP represent the retail hardware, although it's a reasonable assumption. I'd say 90% chance it's 4GB.

Thanks! Any word on the RAM type/config?
 
Found it- https://forum.beyond3d.com/threads/...ulation-discussion.59709/page-58#post-1959110

Turns out it's sebbbi again.



I'm not saying what this user posted is fact as we quickly determined he had no info on the actual Switch hardware, but his perspective helped us understand the limitations of TBR when it comes to helping with bandwidth issues.

And it took me a while to find this because it was in that huge locked thread that started back in October. I should have found the source before I brought it up, my apologies.

Thanks for the link. As you say, he doesn't actually have any experience with Switch itself, and although I don't doubt his experience on these matters, I don't think the following applies quite as much to Switch as he assumes:

Around 50% of modern game engine frame time goes to running compute shaders (lighting, post processing, AA, AO, reflections, etc). Maxwell's tiled rasterizer has zero impact on compute shaders.

In particular, he refers to lighting being handled via compute shaders, which would be the case for a deferred shading pipeline on virtually any modern API & hardware (in creating the G-buffer you would use render-to-texture (RTT), and then the actual lighting calculations would be handled by compute shaders reading the texture, which can't normally be tiled). The issue is that, from what I have read, a tightly optimised game using a Vulkan-derived API on a tile-based GPU is literally the only scenario where this isn't the case. That is, it seems to apply to everything but Switch.

From my reading of the Vulkan documentation (and I'm completely open to being corrected by anyone more knowledgeable on this, by the way), when you're implementing deferred lighting on a tile-based GPU you're not supposed to use RTT, but rather create the g-buffer(s) as what Vulkan calls an "attachment" within a render pass. While you're operating within a render pass in Vulkan you are somewhat restricted on what you can do with these attachments, most specifically a pixel shader can only operate on data from the same pixel across the various buffers and attachments the render pass has access to. This limits some effects from being implemented in this way, but it guarantees that everything within a given render pass can be tiled, giving both the programmer and the GPU explicit information on where and how tiling will occur.

In particular, this should allow the standard deferred lighting techniques to be implemented in a fully tile-able manner on Switch. In fact, Vulkan's entire paradigm of render passes, sub passes and attachments seems to have been explicitly motivated by the desire to allow deferred lighting to work efficiently on tile-based GPUs.

This isn't a free lunch by any means, though. Multi-platform developers would have to put potentially quite a bit of effort into re-working their entire rendering pipeline to make most efficient use of Switch's TBR, and they'd have to maintain that in parallel to their main renderer, as the optimisations wouldn't apply to any other platform. The other items in sebbbi's list would still be difficult to tile (post processing, AA, AO, reflections), although to varying degrees (techniques like post-process AA and AO have a relatively low sample area compared to something like DoF blurring, and they're sufficiently standardised that Nvidia and Nintendo could potentially provide means to implement them in a reasonably efficient manner, such as with overlapping tiles).

This is the reason I'm so impressed by MK8D and Fast:RMX seemingly hitting 1080p/60fps with deferred lighting, though. Because of the typical RTT->compute approach to deferred lighting, it would usually be a big issue on tile-based GPUs, however on Switch there doesn't seem to be a problem. It would seem to confirm that my reading of Vulkan is correct, and they are successfully tiling the entire deferred lighting process.
 
Based on the TX1 floorplans shared a couple pages back* the removal of the A53s would allow very few extra CUDA cores to be added. Maybe 4-8 at most?

SMMs are built in particular group sizes (128CC) for architectural reasons, so you wouldn't just add a few of the cuda cores like that.
 
Thanks for the link. As you say, he doesn't actually have any experience with Switch itself, and although I don't doubt his experience on these matters, I don't think the following applies quite as much to Switch as he assumes:



In particular, he refers to lighting being handled via compute shaders, which would be the case for a deferred shading pipeline on virtually any modern API & hardware (in creating the G-buffer you would use render-to-texture (RTT), and then the actual lighting calculations would be handled by compute shaders reading the texture, which can't normally be tiled). The issue is that, from what I have read, a tightly optimised game using a Vulkan-derived API on a tile-based GPU is literally the only scenario where this isn't the case. That is, it seems to apply to everything but Switch.

From my reading of the Vulkan documentation (and I'm completely open to being corrected by anyone more knowledgeable on this, by the way), when you're implementing deferred lighting on a tile-based GPU you're not supposed to use RTT, but rather create the g-buffer(s) as what Vulkan calls an "attachment" within a render pass. While you're operating within a render pass in Vulkan you are somewhat restricted on what you can do with these attachments, most specifically a pixel shader can only operate on data from the same pixel across the various buffers and attachments the render pass has access to. This limits some effects from being implemented in this way, but it guarantees that everything within a given render pass can be tiled, giving both the programmer and the GPU explicit information on where and how tiling will occur.

In particular, this should allow the standard deferred lighting techniques to be implemented in a fully tile-able manner on Switch. In fact, Vulkan's entire paradigm of render passes, sub passes and attachments seems to have been explicitly motivated by the desire to allow deferred lighting to work efficiently on tile-based GPUs.

This isn't a free lunch by any means, though. Multi-platform developers would have to put potentially quite a bit of effort into re-working their entire rendering pipeline to make most efficient use of Switch's TBR, and they'd have to maintain that in parallel to their main renderer, as the optimisations wouldn't apply to any other platform. The other items in sebbbi's list would still be difficult to tile (post processing, AA, AO, reflections), although to varying degrees (techniques like post-process AA and AO have a relatively low sample area compared to something like DoF blurring, and they're sufficiently standardised that Nvidia and Nintendo could potentially provide means to implement them in a reasonably efficient manner, such as with overlapping tiles).

This is the reason I'm so impressed by MK8D and Fast:RMX seemingly hitting 1080p/60fps with deferred lighting, though. Because of the typical RTT->compute approach to deferred lighting, it would usually be a big issue on tile-based GPUs, however on Switch there doesn't seem to be a problem. It would seem to confirm that my reading of Vulkan is correct, and they are successfully tiling the entire deferred lighting process.

Given the above, and taking into account the potential constraints you mentioned earlier in the thread, what are some current-gen AAA titles that would likely represent a "soft" ceiling for what's economically feasible in terms of Switch ports?
 

Very interesting writeup, thanks. This should bode quite well for first parties and exclusives but like you said, this type of coding requires a fair bit of optimization so I'm a bit concerned that it might be offputting to most big publishers.

SMMs are built in particular group sizes (128CC) for architectural reasons, so you wouldn't just add a few of the cuda cores like that.

Right, I was speaking 100% theoretically and based just on that image, which apparently isn't even an accurate schematic of their sizes.
 
Absolutely. Pre-production bits of info can turn invalid later down the production cycle. Changes happen. I personally believe neither LCGeek nor Nate had personal interest in deceiving anybody on these forums, and were acting in full sincerity. At the same time drive-by low-brow shit-posting is mostly left undisturbed. Go figure.

How are you so sure they're not fake insiders?
 
Huh? The normal Tegra X1 has 4x A53s and 4x A57s, so having the July spec sheet show (on the hardware side, not just the usable side) only 4x A57s should indicate that there are no A53s present in the hardware, right?

It could, or it could indicate that none of those cores are or will ever be accessible by developers. I mean GameCube and Wii specs never listed the little ARM cores that assisted in various tasks (starbuck ect), because they were locked away from developers.

Then again the fact that one main core is restricted is the biggest indication that their likely aren't any additonal cores for the OS. Seems very unlikely that they hadn't yet allowed them to handle the OS and were making devs only use 3 cores until that occured. Here's hoping Nintendo can at least release part of the 4th core for gaming if that is the case.
 
Very interesting writeup, thanks. This should bode quite well for first parties and exclusives but like you said, this type of coding requires a fair bit of optimization so I'm a bit concerned that it might be offputting to most big publishers.

Depends, Ubisoft has switched to tile based rendering, vulkan should see more adoption on PC and we know nothing about how NVN works. Even watch dogs 2 was tile based.
 
Yeah, the CLI access may be a bit odd, although as I've never interacted with console development hardware I wouldn't really know whether that's normal or not (however a full featured Linux package manager may be pushing it a little). One thing that did strike me as unusual is that they use Windows folder notation (i.e. \ ) rather than *nix notation ( / ) when running the script. Admittedly I've (as far as I can recall) only ever used *nix machines to remote access other *nix machines, and have never used Powershell, so I'm not sure what the conventions are in that respect.

When I use git bash on my Windows machine, I still have to use *nix notation, so I don't see why this would switch to Windows notation. I'd ask my boss (he uses PowerShell all the time), but he would just ask me why I'm not working instead.
 
Thraktor, why would they need tbr for those games? To hit 1080p @ 60fps with mk8 shouldn't be hard since the switch is much more powerful than the Wii U. Fast Racing RMX might use it just because it seems to be 1080p and 60fps in all docked modes, so probably isn't even maxing out the system.

I am just thinking we'd expect these resolutions and frame rates base on the hardware alone, tbr aside, switch should be capable of playing wii u games at 1080p at the same frames as 720p wii u games, heck they could very likely double the frame rate too.

Not to say your post didn't have insight, it was a great post, but I feel it gives the false impression that these games are definitely using vulkan and tiled base rendering.
 
Thraktor, why would they need tbr for those games? To hit 1080p @ 60fps with mk8 shouldn't be hard since the switch is much more powerful than the Wii U. Fast Racing RMX might use it just because it seems to be 1080p and 60fps in all docked modes, so probably isn't even maxing out the system.

I am just thinking we'd expect these resolutions and frame rates base on the hardware alone, tbr aside, switch should be capable of playing wii u games at 1080p at the same frames as 720p wii u games, heck they could very likely double the frame rate too.

Because without the TBR, isn't the rumored 25.6GB/s RAM bandwidth of the Switch less than Wii U's effective bandwidth?
 
Because without the TBR, isn't the rumored 25.6GB/s RAM bandwidth of the Switch less than Wii U's effective bandwidth?

We don't know what memory bandwidth the switch has, Nintendo's history is high memory bandwidth with a large low memory bandwidth pool. I see no reason for Nintendo to not do that here. I mean did their engineers just take a vacation throughout the design process? To be honest if this is a X1 with no customizations, it would be the first time any major hardware has come to market from the big 3 and not have changes.

Edit: Eurogamer never claimed to have any idea about final hardware beyond clocks, they only knew the July devkits and the clocks that were targeting launch before any final configurations and testing was done.
 
We don't know what memory bandwidth the switch has, Nintendo's history is high memory bandwidth with a large low memory bandwidth pool. I see no reason for Nintendo to not do that here. I mean did their engineers just take a vacation throughout the design process? To be honest if this is a X1 with no customizations, it would be the first time any major hardware has come to market from the big 3 and not have changes.

Right, and I agree with all of that and expect some degree of customization. But the context of the conversation we were having was about how TBR could be used in place of a hardware solution for alleviating the (potential) bandwidth problem.

The RAM chips we see in the OP have been essentially 99% identified as 32bit each, giving us a max bandwidth of 25.6GB/s. Now, we don't know what's in the die itself or if this is a final retail unit, but as of now it doesn't appear that we will get a 128bit bus as some had hoped, so we're discussing the possibility of the Switch indeed being limited to 25.6GB/s and how TBR might be a potential solution to removing that problem.

It seems like they should have known to go with a hardware solution though to make things far easier for third parties. But who knows.
 
Right, and I agree with all of that and expect some degree of customization. But the context of the conversation we were having was about how TBR could be used in place of a hardware solution for alleviating the (potential) bandwidth problem.

The RAM chips we see in the OP have been essentially 99% identified as 32bit each, giving us a max bandwidth of 25.6GB/s. Now, we don't know what's in the die itself or if this is a final retail unit, but as of now it doesn't appear that we will get a 128bit bus as some had hoped, so we're discussing the possibility of the Switch indeed being limited to 25.6GB/s and how TBR might be a potential solution to removing that problem.

It seems like they should have known to go with a hardware solution though to make things far easier for third parties. But who knows.

GPUs have what? 2MB of L2 cache? increasing it to 4? or 8? 11MB cache is all you need for a 720p cache and 3DS has an 6MB or 8MB embedded memory. Besides, 25.6GB/s just isn't enough, the CPU would need to push through that too, and A57 CPUs can use as much as 5GB/s per core. Even if the Switch's CPUs only use half that, you'd be left with 15GB/s, hardly enough for texture bandwidth and GPU communication.

Anyone thinking they pulled a X1 and soldered it onto a board, called it a day, is crazy. I mean why would they even need to do 3 versions of the Switch chip, considering the OP's "A2"
 
Lots of talk about how switch can be stronger than we think, but for me, that game give us a good idea of what is possible:

Snake-Pass-1.jpg


Snake Pass is 1080p 30fps on switch and 1080p 60fps on Xone.

That game beauty, but it's slow, with not much thing on screen and a repetitive scenery.

If Switch can't make a game like that run in the same quality of the Xbox one, then, it is weaker to that machine, don't matter how much FP16 and Nvidia GFlops help it.
 
GPUs have what? 2MB of L2 cache? increasing it to 4? or 8? 11MB cache is all you need for a 720p cache and 3DS has an 6MB or 8MB embedded memory. Besides, 25.6GB/s just isn't enough, the CPU would need to push through that too, and A57 CPUs can use as much as 5GB/s per core. Even if the Switch's CPUs only use half that, you'd be left with 15GB/s, hardly enough for texture bandwidth and GPU communication.

Anyone thinking they pulled a X1 and soldered it onto a board, called it a day, is crazy. I mean why would they even need to do 3 versions of the Switch chip, considering the OP's "A2"

Like I said I agree that there is likely some memory customizations made, but I'm just discussing things in the context of what we curerntly know.

Lots of talk about how switch can be stronger than we think, but for me, that game give us a good idea of what is possible:

Snake-Pass-1.jpg


Snake Pass is 1080p 30fps on switch and 1080p 60fps on Xone.

That game beauty, but it's slow, with not much thing on screen and a repetitive scenery.

If Switch can't make a game like that run in the same quality of the Xbox one, then, it is weaker to that machine, don't matter how much FP16 and Nvidia GFlops help it.

Er, I've only seen the Switch and PS4 versions, and the PS4 version is not a locked 60fps. No idea about the performance of the XB1 version. Regardless, I don't think anyone is expecting the Switch to be more powerful than an XB1.
 
Lots of talk about how switch can be stronger than we think, but for me, that game give us a good idea of what is possible:

Snake-Pass-1.jpg


Snake Pass is 1080p 30fps on switch and 1080p 60fps on Xone.

That game beauty, but it's slow, with not much thing on screen and a repetitive scenery.

If Switch can't make a game like that run in the same quality of the Xbox one, then, it is weaker to that machine, don't matter how much FP16 and Nvidia GFlops help it.
Nobody said Switch was stronger or on par with Xbox One.
In fact we know for awhile it will be closer to Wii U,pushing 1080p its also not easy for any handheld
 
Lots of talk about how switch can be stronger than we think, but for me, that game give us a good idea of what is possible:

Snake-Pass-1.jpg


Snake Pass is 1080p 30fps on switch and 1080p 60fps on Xone.

That game beauty, but it's slow, with not much thing on screen and a repetitive scenery.

If Switch can't make a game like that run in the same quality of the Xbox one, then, it is weaker to that machine, don't matter how much FP16 and Nvidia GFlops help it.

jimcarreywhoareyoutalkingto.gif
 
Like I said I agree that there is likely some memory customizations made, but I'm just discussing things in the context of what we curerntly know.

You are speculating on what the X1 can do with Eurogamer's clocks, which is fine. Assuming that a power of a Wii U game needed to be brought over to Vulkan and given TBR is a stretch. To believe that Nintendo saw this limitation and decided that they would not customize memory is a further stretch.

My point is that it is more reasonable to assume there is cache on the die, I mean you've already assumed that they removed the A53 cores, so they are willing to make those customization but lack the desire for fast memory when the head of Nintendo's hardware development said that they wanted to make sure that they give their young developers the legacy of fast memory in future platforms just a year or 2 ago?

Nothing lines up with these assumptions, multiple sources are telling us that this is the easiest platform to develop for. How could that be if there is a memory bottleneck?
 
Given the above, and taking into account the potential constraints you mentioned earlier in the thread, what are some current-gen AAA titles that would likely represent a "soft" ceiling for what's economically feasible in terms of Switch ports?

Honestly I don't really know. I'm less concerned about bandwidth than a lot of people seem to be, just as the software we've seen suggests it's not that big of a problem. I wouldn't expect games hitting XBO/PS4 levels, but at the same time there shouldn't really be much stopping any given XBO/PS4 game from running on Switch with some compromises here and there.

Very interesting writeup, thanks. This should bode quite well for first parties and exclusives but like you said, this type of coding requires a fair bit of optimization so I'm a bit concerned that it might be offputting to most big publishers.

Well, it depends on the developer/publisher. For any studio which already has a reasonably well-implemented Vulkan renderer (e.g. id, Epic, Croteam), there shouldn't be too much difficulty involved in tweaking it to run efficiently on Switch. Some other teams may also have the inclination, resources, and/or the expertise to put together a good Switch port, and I wouldn't be surprised to see, for example, Ubisoft, and the larger Japanese third parties, to put in a bit of effort on this front. Then of course there are going to be some publishers who just want to squeeze out a quick port regardless of quality, but that's going to be largely unrelated to the system's architecture.

Part of the issue is that PS4 and XBO are architecturally virtually identical, so it makes Switch seem a lot more exotic than it is. I have no doubt that Switch is much closer to them than, say, PS3 was to XB360, or GC and Xbox were to PS2, or pretty much any other gaming hardware was to its contemporaries. That's not to say producing well optimised Switch ports would be trivial, but I don't see it being a massive issue for any developer that's given a reasonable amount of time and resources to do the job.

When I use git bash on my Windows machine, I still have to use *nix notation, so I don't see why this would switch to Windows notation. I'd ask my boss (he uses PowerShell all the time), but he would just ask me why I'm not working instead.

Yeah, it's an odd one. I've never used Powershell, though, so I wouldn't know if it "translates" to *nix notation or something like that.

Depends, Ubisoft has switched to tile based rendering, vulkan should see more adoption on PC and we know nothing about how NVN works. Even watch dogs 2 was tile based.

Thraktor, why would they need tbr for those games? To hit 1080p @ 60fps with mk8 shouldn't be hard since the switch is much more powerful than the Wii U. Fast Racing RMX might use it just because it seems to be 1080p and 60fps in all docked modes, so probably isn't even maxing out the system.

I am just thinking we'd expect these resolutions and frame rates base on the hardware alone, tbr aside, switch should be capable of playing wii u games at 1080p at the same frames as 720p wii u games, heck they could very likely double the frame rate too.

Not to say your post didn't have insight, it was a great post, but I feel it gives the false impression that these games are definitely using vulkan and tiled base rendering.

Are you mistaking tile-based rendering (TBR) with checkerboard rendering? Checkerboard rendering is a software technique that has recently become popular to render at half-resolution, alternating the pixels used each frame in a checkerboard arrangement, and reconstructing the image to produce full-resolution output. It's up to developers to implement, and Ubisoft have used it in several titles, and Sony have been encouraging it as a way to get "4K" resolution from PS4 Pro.

Tile-based rendering, however, is a hardware technique where the screen is divided up into some number of tiles and each are rendered one by one, with the intention being that the tiles can sit in the GPU's cache while being operated on, reducing the amount of bandwidth required. This still renders every pixel as with a more traditional immediate-mode approach, it's really just optimising the order a frame is rendered to save bandwidth. This also isn't something developers would enable or disable for a particular game, it would simply automatically happen on the GPU. In fact, Maxwell GPUs were using TBR for about a year before anyone outside Nvidia even realised it.

We don't know what memory bandwidth the switch has, Nintendo's history is high memory bandwidth with a large low memory bandwidth pool. I see no reason for Nintendo to not do that here. I mean did their engineers just take a vacation throughout the design process? To be honest if this is a X1 with no customizations, it would be the first time any major hardware has come to market from the big 3 and not have changes.

Nintendo's recent history is a small, high-bandwidth pool of memory on-die, specifically to keep buffer accesses from hitting main memory. Nvidia's TBR implementation is actually a continuation of this, only instead of keeping the entire frame on-die, which would require 32MB+ and take up a huge amount of die area, it keeps only a small portion of the frame on-die at a time, cycling between them. This gives them most of the benefits of their preferred approach while being smaller, cheaper and simpler for developers (as they only have a single memory pool to manage).
 
Lots of talk about how switch can be stronger than we think, but for me, that game give us a good idea of what is possible:

Snake-Pass-1.jpg


Snake Pass is 1080p 30fps on switch and 1080p 60fps on Xone.

That game beauty, but it's slow, with not much thing on screen and a repetitive scenery.

If Switch can't make a game like that run in the same quality of the Xbox one, then, it is weaker to that machine, don't matter how much FP16 and Nvidia GFlops help it.

Apparently the snake itself is extremely demanding on the CPU, the devs have themselves said that. Who was saying this thing is stronger than an Xbox One anyway?
 
Er, I've only seen the Switch and PS4 versions, and the PS4 version is not a locked 60fps. No idea about the performance of the XB1 version. Regardless, I don't think anyone is expecting the Switch to be more powerful than an XB1.

Nobody said Switch was stronger or on par with Xbox One.
In fact we know for awhile it will be closer to Wii U,pushing 1080p its also not easy for any handheld


Sorry, but it is not what it looks like.
OK, there is people that assuming the clocks and other configurations that DF leak as the worst case, when switch is a ~400 GFlops machine with 3 under clocked A57 chips and 25Gb/s bandwidth.

But there is there other side too.
There are people that think the switch use a 16mm pascal custom Tegra. That gives numbers close to 600 GFlops. But they talking more how Nvidia GFlops is roughly 30% better than AMD (what let switch with a performance close to a 780 GFlops AMD in theory), plus the advantage of it utilize FP16, what can increase that number roughly by 50%, what let it at 900 GFlops with mixed precision, or equivalent to a 1170 Gflop GPU from AMD, that is, basically, a Xbox one.
They sum it with the possibility of it use A72 on CPU, what, even with 4 cores, can put switch in the same ballpark as Ps4 in that area.

So yes, there is people that are hoping to a "almost" Xbox one machine.
 
Sorry, but it is not what it looks like.
OK, there is people that assuming the clocks and other configurations that DF leak as the worst case, when switch is a ~400 GFlops machine with 3 under clocked A57 chips and 25Gb/s bandwidth.

But there is there other side too.
There are people that think the switch use a 16mm pascal custom Tegra. That gives numbers close to 600 GFlops. But they talking more how Nvidia GFlops is roughly 30% better than AMD (what let switch with a performance close to a 780 GFlops AMD in theory), plus the advantage of it utilize FP16, what can increase that number roughly by 50%, what let it at 900 GFlops with mixed precision, or equivalent to a 1170 Gflop GPU from AMD, that is, basically, a Xbox one.
They sum it with the possibility of it use A72 on CPU, what, even with 4 cores, can put switch in the same ballpark as Ps4 in that area.

So yes, there is people that are hoping to a "almost" Xbox one machine.

Umm. An "almost X1" machine isn't a X1 machine. Everyone's keeping this between Wii U and X1, the former more likely than the latter. You''re misreading folks.
 
Top Bottom