Nintendo Switch Technical discussion [SOC = Tegra X1]

[bunnybug]

I didn't say your post were all unreasonable, but you didn't answer my question. Why join a forum only to discuss a product that your obviously not big on? All PS3/360 ports jumped to 1080p 60fps? Assassins Creed 4 anyone? Batman Return to Arkham? Tomb Raider 2013? GTA V? Bomberman is using Unity, and we know how well PS4 gets along with Unity, so that's a mute point. I am Setsuna is a great example of a down port, thanks for bringing it up.

Clearly the Switch version is more optimized than the PS4 version after only three months. LOL Come on, even you don't really believe that.

A great member once said the following:

err i was talking about sony's first party ports vs nintendo first party, all those other ports you mentioned received huge graphics upgrades. I am Setsuna runs 60fps on ps4, and 30fps on switch, plus slightly better graphics on ps4

of course it's more optimized, unless you believe switch is on par with xbone? there no real reason the switch version should be that close to the ps4 version.

 

[Rodéric]

Nice port in a short time, confirming that the Switch is indeed easy to work with.
The Switch isn't PS4 in power, that's just ridiculous to imply it, but it doesn't mean it can't have good and good looking games...
(Besides photorealism is very boring and a waste of time and effort IMO, art style is much more important.)
[I'll support photorealism once we have hardware ray tracers, but until then I think it's the wrong way to go.]

 

[sebbbi]

Nice port in a short time, confirming that the Switch is indeed easy to work with.
The Switch isn't PS4 in power, that's just ridiculous to imply it, but it doesn't mean it can't have good and good looking games...

Nvidia GPUs are easier to work with than AMD. To utilize AMD GPU fully, you need to use async compute a lot. Without async compute, you lose around 20%-30% of your peak perf.

Nvidia Kepler/Maxwell/Pascal have awesome geometry performance. In comparison AMD console GPUs are pretty limited in geometry processing capability. If you want to solve this bottleneck, you need custom GPU-driven rendering pipeline with fine grained compute shader primitive culling. DICEs presentation (http://www.frostbite.com/2016/03/optimizing-the-graphics-pipeline-with-compute/) shows 20%-50% perf gain on PS4 from their fine grained GPU culling technology. Nvidia of course also benefits from techniques like this, but the gains are much smaller.

Generic engines like Unreal and Unity do not have GPU-driven rendering pipelines and do not utilize async compute by default. Unreal supports async compute, but developer needs to program their own compute shaders to use it. I would guess that Snake pass isn't doing much custom compute shader work. Unreal and Unity also favor pixel shader based techniques (for image post processing) instead of compute shaders, since these engines must support wide range of hardware and APIs, starting from mobile decides. Having compute version of every shader would add a lot of maintenance cost, since you can't deprecate the old pixel shader based effects. Thus it is cost efficient to stick with pixel shaders. Nvidia Maxwell/Pascal have better rasterizer than AMD, better delta color compression than AMD and Nvidia rasterizer writes to an unified coherent L2 data cache instead of a separate uncoherent ROP cache. On AMD, if you use pixel shaders for post processing, you need to flush all your caches between each pass. On Nvidia you don't. This is a big difference. Compute shader based post processing doesn't have this problem, and groupshared memory further reduces memory traffic (usually resulting in much bigger gains than delta color compression).

It is certainly possible that Snake Pass suits Nvidia Maxwell hardware better than AMD GCN.

 

[phoenix_chipset]

Not agreeing with any POV, but UE3 was a dog on PS3.

I hated how prominent that engine was last generation. Sub 30fps performance and less than stellar visuals except in something like Gears 3. The DmC reboot actually looked worse than DM4 at less than half the frames. Man, MT framework was a great engine though.

UE4 on the other hand isn't bad for this gen.

 

[Deleted member 13524]

Clearly the Switch version is more optimized than the PS4 version after only three months. LOL Come on, even you don't really believe that.

It's a very simple game by today's standards (is that water just EMBM?). Why is it that hard to acknowledge Sumo spent a lot more resources optimizing for the Switch (that would need to take advantage of every last drop of low-level optimizations just to even run) than for a PS4 who would run it with a very low effort?

Both platforms are able to run the game at 30FPS, and that's what matters for the developer. It's a puzzle/slow-platform game and it doesn't try to win any image quality awards.
I honestly don't see anything in this game that couldn't be done in a PS360 using UE3, in a resolution similar to handheld Switch.

 

[SedentaryJourney]

It is certainly possible that Snake Pass suits Nvidia Maxwell hardware better than AMD GCN.

This would imply that in general AMD hardware is handicapped when it comes to most big commercial game engines, unless the developers licencing them have the resources to implement GPU driven rendering pipelines with heavy use of compute shaders.

Since it's probably the case that larger developers are doing the work of optimising UE4 to console and replicating each others work in some sense, it seems this is something that should be fixed upstream with epic maintaining console optimised builds.

 

[bunnybug]

It's a very simple game by today's standards (is that water just EMBM?). Why is it that hard to acknowledge Sumo spent a lot more resources optimizing for the Switch (that would need to take advantage of every last drop of low-level optimizations just to even run) than for a PS4 who would run it with a very low effort?

Both platforms are able to run the game at 30FPS, and that's what matters for the developer. It's a puzzle/slow-platform game and it doesn't try to win any image quality awards.
I honestly don't see anything in this game that couldn't be done in a PS360 using UE3, in a resolution similar to handheld Switch.

He thinks snake pass is taxing the hardware, which is insane considering we have games like star wars and doom 4, running 60fps that look a generation ahead of this technically. i think you and sebbi give a good explanation, maxwell seems like a better fit for this game, and for such a low budget title probably were happy with 900/30fps on ps4 with out needing to put much effort .

 

[phoenix_chipset]

He thinks snake pass is taxing the hardware, which is insane considering we have games like star wars and doom 4, running 60fps that look a generation ahead of this technically. i think you and sebbi give a good explanation, maxwell seems like a better fit for this game, and for such a low budget title probably were happy with 900/30fps on ps4 with out needing to put much effort .

I could see them putting more effort into the switch version. Looking at sonic racing transformed, Sumo digital put the most effort into the ps3 version while the Wii U version suffered. So there's precedence of them prioritizing one version. I think they pick a lead platform and then port to the others. Wii was the lead platform for the first sonic racing.

I don't think Snake pass is taxing either console really.

 

[bunnybug]

I could see them putting more effort into the switch version. Looking at sonic racing transformed, Sumo digital put the most effort into the ps3 version while the Wii U version suffered. So there's precedence of them prioritizing one version. I think they pick a lead platform and then port to the others. Wii was the lead platform for the first sonic racing.

I don't think Snake pass is taxing either console really.

I would have to agree with you, but at 1080p on switch it might be one of the best impressive games on the console, on ps4 it's not even close.

 

[Goodtwin]

Thanks @sebbbi

So that makes more sense than Snake Pass being more optimized for Switch than PS4. Unreal 4 inherently gives more bang for the buck with Nvidia hardware than AMD. The less customized a PS4/X1 game is using Unreal 4, the easier it should be to port to Switch. If they created their own custom rendering pipeline, then it becomes more difficult.

 

[sebbbi]

This would imply that in general AMD hardware is handicapped when it comes to most big commercial game engines, unless the developers licencing them have the resources to implement GPU driven rendering pipelines with heavy use of compute shaders.

I am not implying that at all. Most AAA console game engines are very well optimized for AMD GCN: Frostbite (most EA games), big Ubisoft in-house engines (most Ubisoft games). Naughty Dog engine (Uncharted, The Last of Us), Guerilla Games Engine (Killzone, Horizon Zero Fall), etc, etc.

The difference is that general purpose PC & mobile engines need to support wide range of hardware. For example Unity is still supporting DirectX 9.0 SM 3.0 hardware. It is still important for some Chinese developers. Both Unity and Unreal are also supporting mobile phones and tablets (OpenGL ES 3.0). Unity and Unreal both have robust VR support, and that requires scaling up to 90/120 fps. Both engines have robust plugin support to support various kinds of input and output devices. These engines need to be highly configurable in order to meet all these needs. Obviously a dedicated single AAA game series engine is able to extract a bit more from the target hardware, but it is only suitable for that single game series. For example a game centric AAA engine could combine many post processing passes to a single big hardcoded shader, as there's no need to enable/disable effects one by one. Similarly async compute is much easier to implement, as you know exactly your workload. In general purpose engine, the workload and shader passes change drastically based on the game. The engine developer doesn't know in advance what kind of games the developers are creating with their engine. Tuning the engine for their specific game-play and content needs is the responsibility of the developer.

 

[sebbbi]

So that makes more sense than Snake Pass being more optimized for Switch than PS4. Unreal 4 inherently gives more bang for the buck with Nvidia hardware than AMD. The less customized a PS4/X1 game is using Unreal 4, the easier it should be to port to Switch. If they created their own custom rendering pipeline, then it becomes more difficult.

We don't know. I only said that it is certainly possible that Switch version gets more out of the hardware than PS4 version. This is because GCN needs more special purpose code than Maxwell to reach optimal performance.

 

[Goodtwin]

We don't know. I only said that it is certainly possible that Switch version gets more out of the hardware than PS4 version. This is because GCN needs more special purpose code than Maxwell to reach optimal performance.

Certainly, but if your telling me that Unreal 4 would work better with Maxwell than GNC without special purpose code, I think its fair to assume that Sumo probably didn't dig to deep into special purpose code for Snake Pass. This also means that out of the box, Unreal 4 does leave more untapped performance on the PS4/X1 than Switch.

 

[phoenix_chipset]

I would have to agree with you, but at 1080p on switch it might be one of the best impressive games on the console, on ps4 it's not even close.

Eh... it looks like a nice budget title developed in a short amount of time, no more no less. For both Ps4 and switch. I mean yeah the PS4 build doesn't have a big resolution advantage like it should, but it does have a host of other improvements like depth of field, interactive water, more defined lighting etc.

Let's just agree it's not good game to judge either console's capabilities and appreciate it for what it is.

 

[Goodtwin]

Eh... it looks like a nice budget title developed in a short amount of time, no more no less. For both Ps4 and switch. I mean yeah the PS4 build doesn't have a big resolution advantage like it should, but it does have a host of other improvements like depth of field, interactive water, more defined lighting etc.

Let's just agree it's not good game to judge either console's capabilities and appreciate it for what it is.

The majority of multiplatform games are not the technical showpieces for consoles in general. What Snake Pass does do is give "some" insight into how cross platform games will scale to Switch. Inevitably, multi platform games that come to Switch will be compared to other consoles, and this will be the criteria used to determine the viability and likelihood of future multi platform games being ported over. Retro Studios could develop a new Metroid that blows us away, exceeding what we though a TX1 powered device is capable of, but that really doesn't tell us much about the viability of porting PS4/X1 games to Switch. Until we have a good variety of multi platform games on Switch, its hard to really nail down exactly how many compromises need to be made to transition from PS4/X1. Right now we know a much lower resolution and lower settings are obvious, but are ports of games like COD out of the question? I don't think so, but E3 should be a good indicator.

 

[Deleted member 13524]

The majority of multiplatform games are not the technical showpieces for consoles in general.

Most high profile multiplatform games are.

 

[Goodtwin]

Most high profile multiplatform games are.

Really? So a multi play was the PS3's most impressive title? How about 360? It's always exclusives that squeeze the most from hardware.

Let me clarify, I am not saying multi plat titles are not technically impressive, they often are. Battlefield and Assassins Creed are very impressive, and do make tremendous use of the hardware. I am just saying that multi plat titles typically leave a little performance left untapped because they are not targeting a single platform. Whenever you ask someone what the most technically impressive game is for a given platform, its almost always going to be a first party exclusive. Uncharted and the Last of Us for PS3, and Halo for 360.

Sent from my SM-G360V using Tapatalk

 

[bunnybug]

Really? So a multi play was the PS3's most impressive title? How about 360? It's always exclusives that squeeze the most from hardware.

Let me clarify, I am not saying multi plat titles are not technically impressive, they often are. Battlefield and Assassins Creed are very impressive, and do make tremendous use of the hardware. I am just saying that multi plat titles typically leave a little performance left untapped because they are not targeting a single platform. Whenever you ask someone what the most technically impressive game is for a given platform, its almost always going to be a first party exclusive. Uncharted and the Last of Us for PS3, and Halo for 360.

Sent from my SM-G360V using Tapatalk

well idk i think GTAV and RDR(360), are the most technically impressive games lastgen. you're missing his point he's trying to make. snake pass is not even close to being a technically impressive game on ps4, even compared to high profile third party games, it's a slow platformer with nothing going on screen, the graphics look like something from lastgen, and it's the second lowest resolution game on ps4 in nearly 4 years.

 

[sebbbi]

Nvidia GPUs are easier to work with than AMD. To utilize AMD GPU fully...

Just wanted to clarify that I meant AMD GCN2 (consoles) vs Nvidia's latest (Maxwell/Pascal). AMD PC GPUs have also improved since GCN2.

Improvements for general performance:
- GCN3 introduced delta color compression. Including ability to sample/load compressed textures without decompress step.
- GCN3 improved geometry tessellation performance
- GCN4 improved geometry performance in general (including fast strips, primitive discard, etc).
- GCN4 improved delta color compression.
- GCN4 added instruction prefetch (reduces pipeline latency, again helps with geom bottleneck).
- GCN4 improved async compute scheduling (GPU side)

GCN5 (Vega) adds these general performance improvements:
- L2 cache includes L2 ROP cache (L1 ROP caches under L2). Don't need to flush caches between pixel shader passes.
- Tiled rasterizer. Reduces overdraw, bandwidth and makes ROPs more efficient in general.
- Improved geometry pipeline (including proper load balancing, up to 2x higher peak throughput)
- General purpose memory paging system

(I didn't list features that don't bring performance improvements without programmer intervention)

All of these improvements mean that GCN5 should run general purpose pixel/vertex shader code much better than GCN2. GCN5 has most of the same tricks that are seen in modern Nvidia GPUs. There are nice compute improvements as well, but they need special programmer support (DPP, SDWA, FP16). We will see the real impact of these improvements when DX12 SM 6.0 becomes available. Doom is already using these features with Vulkan, resulting in nice gains.

 

[HMBR]

on the PC you can also find games with fairly absurd relative performance when it comes to Unreal 4, Maxwell vs AMD like: GTX 960 = R9 290!
http://gamegpu.com/images/stories/Test_GPU/Action/ABZU/abzu_1920.jpg
but on the PC it's easier to justify with how the drivers work and so on.

in any case, this is a very positive outcome for the Switch, the resolution difference is obvious on my monitor, but the rest, not really.