There is already a mobile only game coming out so I can imagine a docked only will too.
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
Confirmed: The Nintendo Switch is powered by an Nvidia Tegra X1
- Thread starter FZZ
- Start date
- Status
There is already a mobile only game coming out so I can imagine a docked only will too.
Gotdatmoney
Member
It could. You're allowed. Some dev has a portable only mode for their game. If they want the product out they will give it a shot.
Skittzo0413
Member
I really doubt we'll see a docked only game. It wouldn't surprise me if Nintendo mandated that all games must be mobile at least, and not necessarily the other way around.
For a very intensive game you could probably just target 720p in docked mode and 540p in handheld mode with a few other consolations.
And you miss the fact that the Switch will be able to do both task decently once Netflix and Youtube apps comes available. Can't take a PS4 nor a XBOne out to play on a car trip. Good luck getting your iPad set up to play games on the TV with controllers...The comparisons aren't valid when you want to just cherry pick when the Switch "losses" with ignoring the Switch can do other things to compensate.
That's a fair parallel, but I think something so specific to control input (needing the touchscreen) is different from disallowing the *native* functionality of the system for an engineering workaround to eke out performance.
I suppose time will tell, but I'd put my money on that being very unlikely.
Gotdatmoney
Member
For reference I dont believe any of the publishers of games so demanding they can't run in mobile mode would even bother releasing that game period. So I agree it's unlikely. But as far as we know there is nothing actually stopping it from being done.
Yes indeed.
Lets be glad that the difference between PS4 / XB1 and Switch isn't as vast as high-end 16-bit arcade games of the mid-late 80s and the 8-bit consoles of the same era. -- I remember reading this article on game design in Video Games & Computer Entertainment, feeling like there was little hope of playing my favorite arcade games at home -- even with the then-new 16-bit consoles that had just been released (Genesis, TG-16) and not knowing something like the Neo-Geo would soon be on the horizon.
Downporting to Switch really isn't that bad.
Liabe Brave
Member
I believe I see the problem. I was wrong, but neither of your calculations is correct either; we all have made incorrect simplifying assumptions about the workload.That's no proper math
Think of it this way: imagine that the GPU could render 100fps in FP32 mode. Now, in FP16 mode, 30% of the GPU still do FP32, so that part renders 30fps. The other 70% of the GPU however, now render twice as fast, i.e. 140fps, instead of 70fps. Which means: 170fps overall -> 70% boost.
I assumed that the extra power freed by FP16 would all be used by FP32 tasks. Both of you assumed the extra power would all be used by FP16 tasks--hence 70%. But neither scenario is likely to actually happen.
Maybe it's clearer to think of as timeshare: say the workload completes in 30ms on normal architecture. With Switch, 9ms are used by FP32. The remaining FP16 work completes in 10.5ms instead of 21ms, freeing 10.5ms. What work fills the gap? We have to assume it's distributed just like the previous work, i.e. 30/70 FP32/FP16. So 3.15ms and 3.68ms, leaving 3.68ms free, which is again split 30/70, etc. etc.
This converges toward ~50% gain, not 35% or 70%.
icecold1983
Member
Lol so fp16 and nvidias magic now makes switch half as fast as ps4. Never change nintendo fans
Just_myles
Member
So it's dead stock? No customization?
Skittzo0413
Member
Care to share where anyone said that?
It's likely to be somewhere around half of an XB1 but that's certainly not half a PS4. Nobody has actually claimed that.
ShadowFox08
Member
Or maintain 720p in both handheld and docked, but just increase the fidelity for docked mode.. Or 540p for handheld.
which ever saves more energy in handheld mode while maximizing fidelity and smoothness of docked mode If devs are down for it.
ShadowFox08
Member
Right, especially when PS4 not only has more RAM, but faster(DDR5).
From some ports we've seen so far like I am setsuna and Dragon Quest spin off, PS4 has twice the framerste and better lighting and even texture in some areas(for Dragon Quest at least). PS4 looks like at the very least twice as powerful. GPU and RAM a bit more. I'm guessing 2-2.5x at least.
How much more bandwidth does PS4 have over switch btw?
lyrick
Member
Your DDR is missing a G
Probably to prevent any cannibalization of Nvidia Tesla sales.
SomewhatGroovy
Banned
Realistically, based on what is being discussed here, what can we expect from Skyrim? Are we getting a port of the base game that came out in 2011?
Skittzo0413
Member
...did you actually read that post? He is talking solely about the GPU, and saying that at it's theoretical maximum, 100% optimized, it could be around 50% of the PS4 GPU, depending on whether or not the Nvidia flop advantage occurs in consoles the way it does in PCs.
Nobody is trying to claim the Switch is half as fast as the GPU. Properly optimized games will always use the hardware better than the majority of the games.
It'll likely be much closer to the special edition on PS4/XB1, though with some downgrades. My pet theory though is that it will have upgrades like HD rumble, Nintendo themed mods, and potentially optional motion controls.
Yeah, that's true too. It'll come down to the dev's priorities. The point is, getting a game to work in portable mode if it works in docked mode really shouldn't be that big of a problem.
My guess would be somewhere between, probably closer to last gen versions but run better. Maybe 900p or 1080p.
icecold1983
Member
Yes i read the post and im aware hes talking about gpu performance. Keep drinking that koolaid if you think the tx1 in switch will ever be anywhere close to half the performance of the gpu in ps4
Well its all theoretical of course and depends on the developer and game. But in the theoretical scenario (mentioned by the developer discussed earlier) were 70% of code can be changed to 16bit precision, Switch's GPU would have a max theoretical output of 668gflops. Which is 70% higher than its max at full 32bit precision.
I think in reality its safer to think about a performance increase of 35-50% though. I agree with that because I think 70% seems like a near best case scenario, but people were just talking about theoretical numbers (specs always are, they never work out properly in the real world).
icecold1983
Member
Turning code to fp16 isnt some magic button you press to double performance for that % of code
depends on the code. For some shaders it pretty much is. you just change the precision of your variables.
The issue however is that even if the precision is good enough, not every instruction is available for fp16. So if you're doing more complex stuff the values may get converted back and forth between fp16 and fp32 and you'll loose the benefits because of the conversion overhead if you're not careful. So the actual benefits depends completely on the type of code. Then again FLOPS has always been a bullshit number
Nothing in games development has a magic button, but I wasn't discussing the specifics of implementation.
Tegra processes two 16bit operations at the same speed as one 32bit operations. Which exact operations can be easily changed from 32bit to 16bit and how hard it is to implement is another question entirely. We're discussing a theoretical number based on one specific scenario (a developer saying his game can have 70% of its code run fine in 16bit precision).
Like I said in the post you actually quoted, IMO this seems like a close to best case scenario so realistically would be safer to assume a lesser gain than 70% even for exclusive games that take advantage of the feature. But again within the context of what we're discussing (the developers comments) these numbers of theoretically correct. Worth noting as well BTW that all flop numbers we discuss are theoretical and none of them can be achieved with the clicking of a magic button, or at all even in the real world.
While I like running numbers, no I don't think switch will ever utilize 100% efficiency, and thus it isn't going to be half of ps4. It was just taking the full performance gain and adding Nvidia's performance advantage we see on PC.
I think it really depends on how much time is put into the port. Best case scenario it's a midpoint closer to the old version. Worst case they just do a straight port of the original with some tweaks under the hood.
Polygonal_Sprite
Member
The Splatoon 2 test fire on Friday will be interesting to see if it runs at 900p or 1080p when docked. The app actually boots just now so I wonder if we can already tell or do menus usually run at a different resolution to the actual games?
In terms of best case scenario for Switch I think Xbox One games at 720p when docked with lower quality assets in terms of third party games will be achievable.
In terms of best case scenario for Switch I think Xbox One games at 720p when docked with lower quality assets in terms of third party games will be achievable.
Alebrije
Member
Iam ignorant about Tegra1 and tech hardware. But reading this thread it seems Skyrim will be the worse case scenario you say.
It's clearly Nintendo is a software oriented company and not a hardware one when talking about home consoles. GameCube was thier last effort to have a competitive hardware.
playing Skyrim on the go will be great for mobile enthusiasts but as home console it would be better to play it on PS4 Xbone.
It's interesting to compare what the Swicth can do compared to other APU/GPU's. Did a check on youtube, and it seems like Skyrim on Switch operates at much higher graphical settings than what the A8 7600 is capable of. In-fact, Skyrim on Switch gives us a somewhat decent window into where the Switch's GPU is at in comparison to other GPU/APUs.
A8 7600 Seems to get 30-33 FPS at mid-lowest settings with godrays off, no AA and at a 720p resolution.
https://www.youtube.com/watch?v=YgyBtJGDYyY&t=1126s
Switch version seems to have godrays turned on, AA of some sort enabled, settings mid-high, and the resolution seems higher than 720p (though obviously I can't be sure just by looking, but it seems clearer than a 720p image).
https://www.youtube.com/watch?v=qd-6hEmc2-Y
I also compared it to footage of the GTX 745 running Skyrim SE and it seems it operates at higher graphical settings than what that GPU capable of too (and that is an 800GF GPU for the Kepler architecture).
I know a lot of people weren't sure, but the Switch gameplay shown doesn't seem to be footage of PS4/XB1 or PC, I did a comparison, and it seems draw distance is lower, and it might be a bit... foggier? There's certainly a difference in settings to those versions.
Yep, you are right, I didn't think of it correctly, it's not that 70% of the GPU doubles, it is that 70% of the code can be ran at fp16, the other 30% has to be ran at fp32, so you end up doing around 180gflops in fp32 and the rest of the gpu can run at fp16 for about ~600gflops, not 680gflops with mixed precision.
lets assume for simplicity that FP32 takes 2 cycles, while FP16 takes only 1 cycle to execute.
if you need to run a program of 100 FP32 operations. it would take 200 cycles.
If you need to run 30 FP32 + 70 FP16, it would take 30*2+70 = 130 cycles.
130/200=0.65 -> using FP16 in 70% of the calculations will reduce the runtime by 35%
(or in the opposite direction 200/130=1.54 -> using only FP32 vs mixed will make it run 54% slower)
The problem here is that if you run FP16 for 70% of the time it will be more than 70% of the code.That's no proper math
Think of it this way: imagine that the GPU could render 100fps in FP32 mode. Now, in FP16 mode, 30% of the GPU still do FP32, so that part renders 30fps. The other 70% of the GPU however, now render twice as fast, i.e. 140fps, instead of 70fps. Which means: 170fps overall -> 70% boost.
So with the numbers from above:
200 cycles for all FP32
running FP32 30% of the time -> 200*0.3(cycles) / 2(cycles/op)=30 operations
running FP16 for 70% of the time -> 200*0.7(cycles) / 1 (cycle/op) = 140 operations
So overall you have 140/(140+30) = 82% FP16 ops
Yeah, Maxwell X1 is from 2015, it has newer features and much more efficient than the AMD GPUs from 2011, used in xb1 and ps4. This is why I said it's not a 1:1, there are times where the A8 7600 might out perform it too, such as the new Doom running in vulkan should run better in A8 7600 than it can on the Switch, but it's hard to know this for sure and it should still be fairly close to that performance with maybe a few missed frames.
Polygonal_Sprite
Member
Reading the past few pages be like
All that matters is what's shown on screen. So far we have seen some nice looking Nintendo games that don't exactly look a generation ahead of Wii U exclusives and a distinct lack of big name third party announcements. I'm expecting Nintendo games that push visuals beyond Wii U to run at 900p when docked and the others that push Wii U like visuals to run at 1080p when docked.
That's fine for me. I bought Switch to play Nintendo and Indie games.
All that matters is what's shown on screen. So far we have seen some nice looking Nintendo games that don't exactly look a generation ahead of Wii U exclusives and a distinct lack of big name third party announcements. I'm expecting Nintendo games that push visuals beyond Wii U to run at 900p when docked and the others that push Wii U like visuals to run at 1080p when docked.
That's fine for me. I bought Switch to play Nintendo and Indie games.
Wait you think a Wii U game on switch looks like a Wii U game? It's fine that you have low expectations, but we do have examples of real world performance from PC in under optimized situations, it's really not a good idea to take a Wii U game ported to the Switch and claim that is what the console can do.
If you look back to the Wii U launch, you saw multiple ports from 360 that ran worse than on the 360, yet we know the Wii U is more capable than the 360 and can even be seen in early games like NFS: most wanted u
callhitman
Member
that will do x1. that will do.
Pitcairn released in March 2012 and Maxwell 2 (which the X1 is based on) in September 2014
Really? X1 is a 20nm chip and the first to add features found in pascal, but somehow it's part of the GTX 980 chip? The AMD HD 7000 series is unified in features afaik, the X1 is not simply like any other maxwell card, but sure you can say march 2012 for XB1 and PS4, you do have to give that same consideration to X1, as there was nothing like it from Nvidia until May 2015.
shouamabane
Member
I'm curious to see what games look like that really push the Switch in a year or so.
Polygonal_Sprite
Member
I'm only going on what we've seen -
Ports -
Zelda - 900p / Has horrendous framerate drops into the teens in certain areas.
Fast Racing Remix - Dynamic 1080p / double framerate for 3/4 players in mp mode / improved lighting.
Mario Kart 8 Deluxe - 1080p / no double framerate for 3/4 players in mp mode / no improved lighting or visuals.
Original games -
Splatoon 2 - 720p / Looks near identical to the Wii U original.
Super Mario Odyssey - 720p / Looks very similar to Mario 3D World in larger levels but with fewer player controller characters / improved lighting.
Fire Emblem Warriors - Looks very similar to Hyrule Warriors visually with better lighting.
I don't think that disproves anything I said in my last post. Switch looks to be outputting games which visually are very similar to Wii U except at times are displaying them at a higher resolution with better lighting. What we're seeing on screen does not support your theory that Switch is 4-5x more powerful than Wii U. Yes the games look better but it's nothing like a generational leap.
On top of that there is very little third party port announcements (which is imo a clear indication of what third parties think of the hardware). It's clear that Nintendo were aiming for a Wii U that could be taken on the go with Switch (probably the original vision of Wii U).
All the talk of Nvidia flops being 30% more than AMD flops / fp16 / miracle working API's etc is well and good but until we actually see games that support any of those theories it's utterly baseless speculation. If Switch was 4-5x Wii U it would not be stuttering around at 15fps in Zelda no matter how shoddy the port job.
https://www.youtube.com/watch?v=P2dvhoBbqw8
Fast RMX runs at (dynamic) 1080p 60fps while Fast racing neo ran at 720i (yes interlaced) less effects, this is a 4 to 5 times better performance, but whatever, just enjoy the video I posted. You can be in the wait and see, but no one should be spending much time on your speculation about performance because it's based on 'Wii U is only powerful enough to render NSMBU at 720p and Nintendoland is a pretty simple game, this console is definitely weaker than 360'
icecold1983
Member
what features found in pascal? its identical to a 980 with the exception of limited support for 2x fp16
I have no idea what a lot of these features do but these were added to the second-gen Maxwell architecture (found on the page for Maxwell on wiki):
Dynamic Super Resolution,Third Generation Delta Color Compression, Multi-Pixel Programming Sampling, Nvidia VXGI (Real-Time-Voxel-Global Illumination), VR Direct, Multi-Projection Acceleration, Multi-Frame Sampled Anti-Aliasing(MFAA) and Direct3D12 API at Feature Level 12_1. HDMI 2.0
Polygonal_Sprite
Member
Dude you're putting words in my mouth. I never said "this is all Switch will ever be capable off". I'm simply saying from what we've seen of the games so far Switch looks to be on the same level visually as Wii U with better lighting and at times running games at a higher resolution.
In terms of geometrical complexity, lighting, materials, character models and textures it isn't a generational leap and that's fine because it isn't what Nintendo set out to make. Had they wanted that they would have built a PS4 level console for $250 and called it a day.
Switch is very impressive for its form factor but imo it's nowhere near as powerful as you're suggesting. Time will tell I guess.
Edit - I don't know what your video is meant to prove. Zelda was Nintendo's five year in the making flagship launch title not a third party remaster...
I agree, but even if the Switch can do better, people never think about the production value target. Nintendo changes its graphical generation every two generations. They barely started with WiiU, and it was a massive leap into the unknown, with modern shaders etc.. They've been really good at it, but they won't jump to their next generation of graphic that soon. Even if they have more power, it'll be used to make game run better, at better rez etc.. But the general level of assets won't be drastically different suddenly.
And let's not forget, the Switch is also a 3ds successor, and in that case the jump in budget / production value is really scary. Game Freak are looking for a 3D artist and aiming at Vita/WiiU level of models.
Polygonal_Sprite
Member
I completely agree. One of the reasons I think they went with a modest hardware leap was development costs. Creating every Nintendo game looking like Ratchet PS4 would not be good for the balance books.
I've never suggested that it was a generational leap, it is noticeably faster than Wii U and has more modern features than the PS4/XB1, it also is on par with the AMD APU I listed, that is not surprising really and is a fairly low bar for Switch to hit. This isn't some great leap in performance and it doesn't somehow mean that it is on par with current gen consoles.
As for your edit, Zelda isn't a 5 year game, it wasn't produced as a launch title for the Switch, it was produced as a swan song for the Wii U and ported over about a year ago to unfinished Switch hardware, it's not going to be optimized because it was rushed to meet launch and only went gold a couple weeks before the Switch launched.
Zelda breath of the wild hit full development in January 2013, and went gold in February 2017 or about 4 years with most of that time spent on the Wii U.
Polygonal_Sprite
Member
Ok 4 years not 5. Just because it was originally a Wii U game doesn't mean it shouldn't run and look significantly better if Switch is really 4-5x more powerful even with a year of porting (which isn't a rush job by any means).
BotW would be 1080p when docked with better shadows, better texture filtering, better AA and a solid 30fps were Switch as powerful as you say. Just because they're getting 1080p on Fast Racing and Mario Kart doesn't mean they will hit that improvement in every game because as you know different genres require different hardware needs.
If Zelda was rushed for launch where is the framerate optimisation patch almost a month later? I don't think it's a matter of a rushed port at all but rather Switch being a modest hardware leap over Wii U which struggles to run anything other than a closed circuit racing game at 1080p.
As I said earlier Splatoon should run at 1080p in docked mode. If it doesn't it will add further fuel to the fire that Switch isn't much more powerful than Wii U.
- Status