• 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.

DF: Nintendo NX Powered By Nvidia Tegra! Initial Spec Analysis

Status
Not open for further replies.

dr_rus

Member
What a ridiculous assertion. Had it not been for NVIDIA'S superior performance, we would have no frame of reference for AMD's drivers being 'crap' by comparison. If NVIDIA was no better at optimizing for their own hardware, NVIDIA drivers would show similar levels of 'crappiness'.

And I was talking about the Wii U's API, not the NX. And I wasn't saying it was much worse AMD's drivers, just that it wasn't automatically better just because it was on fixed hardware.

If by "proactive optimization" you actually meant "having written a much better drivers overall" then you are correct. If you meant that NV specifically optimizes for each new title out there to get to this type of performance advantage then you're mostly wrong - as most Game Ready drivers do not in fact provide any noticeable performance gains compared to whatever drivers were available previously and serve mostly as a compatibility update.

Wii U's API should be better than AMD's drivers as well as it's as thin as any other console API out there. No reason to suspect the opposite.
 

z0m3le

Banned
If by "proactive optimization" you actually meant "having written a much better drivers overall" then you are correct. If you meant that NV specifically optimizes for each new title out there to get to this type of performance advantage then you're mostly wrong - as most Game Ready drivers do not in fact provide any noticeable performance gains compared to whatever drivers were available previously and serve mostly as a compatibility update.

Wii U's API should be better than AMD's drivers as well as it's as thin as any other console API out there. No reason to suspect the opposite.

In practice the 750ti's architecture performs at about 40% faster than the architecture inside Ps4 so not sure the wii u's api would magically make up any ground since GCN is already 30% faster than the architecture inside Wii U.
 
If by "proactive optimization" you actually meant "having written a much better drivers overall" then you are correct. If you meant that NV specifically optimizes for each new title out there to get to this type of performance advantage then you're mostly wrong - as most Game Ready drivers do not in fact provide any noticeable performance gains compared to whatever drivers were available previously and serve mostly as a compatibility update.

Wii U's API should be better than AMD's drivers as well as it's as thin as any other console API out there. No reason to suspect the opposite.

Yeah, I meant NVIDIA's overall driver optimization, not the stability updates. When I said proactive, I mean generally, they are better at plotting out the overall optimization for their new architecture ahead of time than AMD is.

As for Wii U's API, I think you misunderstand my argument. I'm not saying can't be better, I'm saying it's not automatically or inherently better. Some optimization is needed to ensure that it outperforms its driver counterpart.
 
But not entirely. If it were then we wouldn't see the 1.39TFlops 750ti perform so favourably against the 1.84TFlops PS4 (which should be mostly divorced from AMD's driver issues).

That could have something to do with better CPUs in the PCs. The videos i have seen showed some i3 and FX6300 cpus paired with 750ti.
 

Thraktor

Member
Well, you have to draw the line of usefulness somewhere. Surely you can keep those A72s at some 500MHz undocked, particularly if you're conservative with the A53s as well, but why not firmly put those A53s into 2GHz+ territory instead? I know you're really into HH/console CPU parity, but as nice as that sounds, I'm still not seeing it at our current levels of tech.

Fortunately we have actual data to settle our little dispute. Firstly, from single-core floating point results from Geekbench 3 on the Huawei P8 and Mate 8 (which isn't going to be perfect, but should be as good an analog of gaming performance as we can reasonably get) we can work out that an A72 core will outperform an A53 by about 89%, clock for clock.

Then, we can get power curves for a quad-core A53 cluster from Anandtech's look at the Exynos 5433 and for a quad-core A72 cluster from their look at the Kirin 950. Granted the Exynos 5433 is fabbed on Samsung's 14LPE and the Kirin 950 on TSMC's 16FF+, so the latter could potentially have a small fab advantage, but it's the best (and closest to like-for-like) data we have available to us.

If we set the CPU TDP limit at 500 mW in handheld mode, then this data tells us that an A53 cluster on its own would manage slightly over 1.1 GHz. In comparison, with both an A53 cluster and an A72 cluster, you could clock the A53s at 600 MHz and the A72s at 550 MHz within the same 500 mW envelope for 49% higher performance.

If you up the CPU TDP to 1 W, then you get just under 1.5 GHz for a cluster of A53s, but if you use both clusters you could clock the A53s to 800 MHz and the A72s to 1GHz, giving you 79% better performance. Effectively, unless you're running the CPU off an extremely low power draw (like sub 250 mW), you get better performance per Watt by using every CPU core rather than just some of them.

Personally I'd expect an octo-core A53 setup. They take up very little die space, they could run them at ~1GHz in handheld mode for pretty decent perf/W, and if they do want to clock them up in docked mode you could get 2GHz+ out of them without much difficulty.
 
Wii U's GPU downgrade started as the same as NX. Wii U GPU being based around RV770 (HD4870), 1tflop talk, better GPU than PS360, midstep between this and next generation.

In the end, we all know how it turned out.
The documentation never said a "RV770." It states that it was based on the "R700" and it definitely was. Dx10 capabilities and all.

The Wii U's GPU likely designed in consideration of the CPU being stretched beyond its limit, which will not be an issue with tegra X1
 

kami_sama

Member
The documentation never said a "RV770." It states that it was based on the "R700" and it definitely was. Dx10 capabilities and all.

The Wii U's GPU likely designed in consideration of the CPU being stretched beyond its limit, which will not be an issue with tegra X1

While the Wii U's GPU is not as good as we hoped, the CPUs inside are almost criminally underpowered.
 

z0m3le

Banned
I, too, would appreciate this.

PS4/XB1's GCN architecture is about 30% faster than R700 architecture used in Wii U, Pascal is ~40% faster than Polaris which is more or less the same as GCN flop to flop... ROUGH estimate, but Pascal should be about 90% faster than Wii U, flop to flop. (X1 is maxwell based, and again there isn't much difference between performance per flop with pascal)

For illustration purposes, X1 is 512GFLOPs, this is over 5 times faster than Wii U's 176 gflops (nearly 3 times faster than the 352gflops it sometimes gets confused to have) and somewhere around 60% of XB1's performance.

If NX is using Pascal (I don't see Nvidia pushing out another maxwell chip tbh) they can hit much higher clocks when docked, so even in the same configuration as X1 (256 cuda cores) if it is reaching 1.5ghz or 1.6ghz, you'll have 768 to 819GFLOPs (+40%) which gives you 1 to 1.15 TFLOPs, slightly under XB1. If the pascal chip is 3 SM, it would be completely possible to hit 1.228 tflops (+40%) which gives you 1.7tflops, or just under PS4's 1.843TFLOPs.

In order to reach those clocks, NX will have to have a fan in the body of the device, but that fan can be kept off while on the go, and be down clocked to 1/4th it's docked clock, which is perfect for 540p, it is also possible to waste battery and go with 1/2 it's docked clock so that it can have a terrible battery life and hit 720p. The dock can also offer a blower to help move air through the device, as long as the vents allow and the passive cooler fins are designed to be cooled from the side. An active cooler inside a device like this would be very interesting, as they could bridge the gap between devices.

Pay no attention to the speculation if you want, but Pascal is the only architecture that can bridge the gap between a handheld and a modern console.
 
I was thinking about it approaching A9X levels, which is faster than the Pixel C.

You're assuming that the NX and the Pixel C will have the same SoC (they won't). And if this rumor about the NX using a pascal-based Tegra is true, then it's very unlikely that the NX's SoC will be less powerful than the A9X. Do we even have a confirmed FLOP count for A9X? I saw somewhere that it was around 700 GFLOPS, but I'm not sure how accurate that is.
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
Fortunately we have actual data to settle our little dispute. Firstly, from single-core floating point results from Geekbench 3 on the Huawei P8 and Mate 8 (which isn't going to be perfect, but should be as good an analog of gaming performance as we can reasonably get) we can work out that an A72 core will outperform an A53 by about 89%, clock for clock.

Then, we can get power curves for a quad-core A53 cluster from Anandtech's look at the Exynos 5433 and for a quad-core A72 cluster from their look at the Kirin 950. Granted the Exynos 5433 is fabbed on Samsung's 14LPE and the Kirin 950 on TSMC's 16FF+, so the latter could potentially have a small fab advantage, but it's the best (and closest to like-for-like) data we have available to us.

If we set the CPU TDP limit at 500 mW in handheld mode, then this data tells us that an A53 cluster on its own would manage slightly over 1.1 GHz. In comparison, with both an A53 cluster and an A72 cluster, you could clock the A53s at 600 MHz and the A72s at 550 MHz within the same 500 mW envelope for 49% higher performance.

If you up the CPU TDP to 1 W, then you get just under 1.5 GHz for a cluster of A53s, but if you use both clusters you could clock the A53s to 800 MHz and the A72s to 1GHz, giving you 79% better performance. Effectively, unless you're running the CPU off an extremely low power draw (like sub 250 mW), you get better performance per Watt by using every CPU core rather than just some of them.

Personally I'd expect an octo-core A53 setup. They take up very little die space, they could run them at ~1GHz in handheld mode for pretty decent perf/W, and if they do want to clock them up in docked mode you could get 2GHz+ out of them without much difficulty.
I commend you for your analysis, but since I'm on literally vapours of internet ATM, and I cannot really verify your sources, I'll just emphasize something I thought I had mentioned in my original post, yet I actually didn't, and something you've overlooked - single thread-performance is still a considerable factor in most such decisions. You're actually looking at a trade-off between throughput (i.e. your optimisation points) and capability to handle individual latency-sensitive tasks (i.e. max sustainable ops one can squeeze out of a core, which apropos does not need to be fp-related at all). Otherwise yes - a bunch of low-clocked cores will always have better power efficiency at throughput, until you start eating into your latency-critical margins.

One could argue that's precisely an argument in favour of having an A72 or two always at hand, and power-govern them at will, but that is not trivial in itself, as such power-level transitions take time and warmup power on their own, so if you need to do that, say, multiple times per frame at the end you still get diminished latency response and wasted power. At the end of the day it really depends what usage scenarios the hw vendor envisioned. But whatever they plan for, they cannot neglect latency-critical response capabilities.
 

Rodin

Member
Well, I'm basing my assessment on that Tegra K1 was the top performing Android soc by a fair margin until the Adreno 820, while being challenged by contemporary Apple phones and tablets. But maybe is an Apple thing more than a PowerVr one? I expect the NX to be a step or two bellow the iPhone 7.

There's absolutely no chance of this happening. Even if the chips were close (and i strongly doubt that the iPhone will have something that can compete with a Pascal-based Tegra SoC designed specifically for a hybrid console... or even the X1 itself), with the iPhone you have to manage heat in a <7mm device, and considering that it's still a phone that has to do other things and run other processes even when you game on it, battery life would also be a huge problem. NX, even in handheld mode, will be a lot thicker and its cooling will be designed with in mind a machine that has to play games. It will also likely have less and slower RAM.

That's without considering that NX could (or could not, we don't know yet) perform vastly better when connected to the dock. Even in portable mode, this thing will trounce the iPhone 7 and the next one without breaking a sweat, and the visuals of the games, between a possibly lower resolution, more/faster RAM, no iOS, better API and a higher budget for games compared to your average mobile title, won't be feasible on the iPhone for... quite a long time.
 

MuchoMalo

Banned
I was thinking about it approaching A9X levels, which is faster than the Pixel C.

Nah, that's not happening on 14/16nmFF. The 6S doesn't even match the A8X. Not even close actually.

ShRlEXo.png
 

Philippo

Member
So, if they're going with the X1 and not a successor (or any kind of custom improvement), what kind of 3rd party game we can expect?
Like, would this and next year multiplaform games like Battlefield 1, Watch Dogs 2, SW: Battlefront etc. be any feasible on the handheld?
 

emag

Member
So, if they're going with the X1 and not a successor (or any kind of custom improvement), what kind of 3rd party game we can expect?
Like, would this and next year multiplaform games like Battlefield 1, Watch Dogs 2, SW: Battlefront etc. be any feasible on the handheld?

It clearly cannot be a standard X1, for the multitude of reasons previously stated.

But, to answer your hypothetical questions directly:
1. Mobile ports and designed for NX games.
2. No, not in any reasonable form.
 

R00bot

Member
So, if they're going with the X1 and not a successor (or any kind of custom improvement), what kind of 3rd party game we can expect?
Like, would this and next year multiplaform games like Battlefield 1, Watch Dogs 2, SW: Battlefront etc. be any feasible on the handheld?

If these are the specs then those games would all be possible. I expect Watch Dogs 2 to come to it, but the others probably not.
 

Thraktor

Member
I commend you for your analysis, but since I'm on literally vapours of internet ATM, and I cannot really verify your sources, I'll just emphasize something I thought I had mentioned in my original post, yet I actually didn't, and something you've overlooked - single thread-performance is still a considerable factor in most such decisions. You're actually looking at a trade-off between throughput (i.e. your optimisation points) and capability to handle individual latency-sensitive tasks (i.e. max sustainable ops one can squeeze out of a core, which apropos does not need to be fp-related at all). Otherwise yes - a bunch of low-clocked cores will always have better power efficiency at throughput, until you start eating into your latency-critical margins.

One could argue that's precisely an argument in favour of having an A72 or two always at hand, and power-govern them at will, but that is not trivial in itself, as such power-level transitions take time and warmup power on their own, so if you need to do that, say, multiple times per frame at the end you still get diminished latency response and wasted power. At the end of the day it really depends what usage scenarios the hw vendor envisioned. But whatever they plan for, they cannot neglect latency-critical response capabilities.

Even taking into account single thread performance evidence would suggest that it's still worth making use of all eight cores. In my first scenario (500 mW) the drop in single thread performance from a 1.1GHz A53 core to a 550MHz A72 core is just 5%, and in the 1W scenario you actually see a 26% jump in single-thread performance from a 1.5GHz A53 core to a 1GHz A72.

Granted this is based on power curves from two different SoCs on different processes, and Geekbench FP isn't a perfect analog for gaming workloads (I've tried to find a suitable integer benchmark for comparison, but most, including Geekbench, have a heavy crypto component, which isn't of much use evaluating the general integer performance of cores with dedicated crypto blocks).

In any case I still believe Nintendo will end up using symmetric cores (except for the possibility of a couple of A35s reserved for OS duties), and they'll all be used both in handheld and docked mode.

So, if they're going with the X1 and not a successor (or any kind of custom improvement), what kind of 3rd party game we can expect?
Like, would this and next year multiplaform games like Battlefield 1, Watch Dogs 2, SW: Battlefront etc. be any feasible on the handheld?

It clearly cannot be a standard X1, for the multitude of reasons previously stated.

But, to answer your hypothetical questions directly:
1. Mobile ports and designed for NX games.
2. No, not in any reasonable form.

I disagree. If we're talking about a Pascal GPU with 2 SMs running at around 600 MHz (which would give use about 300Gflops and would seem a reasonably likely scenario at the moment), then so long as the screen is around 540p, it would be absolutely capable of ports of the games listed. They'd have to be dialled down a little bit to run at native res (which they mostly don't on XBO and PS4), but the major roadblock would be third parties wanting to port the games to a Nintendo platform, not the capabilities of the hardware.
 
I disagree. If we're talking about a Pascal GPU with 2 SMs running at around 600 MHz (which would give use about 300Gflops and would seem a reasonably likely scenario at the moment), then so long as the screen is around 540p, it would be absolutely capable of ports of the games listed. They'd have to be dialled down a little bit to run at native res (which they mostly don't on XBO and PS4), but the major roadblock would be third parties wanting to port the games to a Nintendo platform, not the capabilities of the hardware.

Yeah I asked this in the Pascal thread, that seems to make sense to me. And as far as third parties wanting to port these games, it may actually be a pretty big enticement for them to offer the exact same games that you can get on PS4 or XB1 on a Nintendo portable. I wouldn't bee terribly surprised to see a lot of third party ports around launch to test out this "full console games on the go" idea.
 

Luigiv

Member
That could have something to do with better CPUs in the PCs. The videos i have seen showed some i3 and FX6300 cpus paired with 750ti.

No, not really. A better CPU will help prevent bottlenecking, but it won't make up for lack of graphical grunt. Not on a PC, anyway.

If a PS4 title is targeting 30fps then we can assume it's graphical profile will be set to make the most of the resulting frame time in mind, even if that original decision to target 30 was born of CPU performance concerns. The systems specs are fixed, so if you're being forced into a lower framerate by one component, you may as well increase the load on the other component to match. Most games on the system are maxing out what GPU can do at a given framerate, regardless of that weak CPU. So the fact that so many games on the 750ti can match framerates, with the same graphical setting, means that GPU has to be able to match the PS4's graphical grunt.

I hope that makes sense. That came out rather inelegantly.
 

ArtHands

Thinks buying more servers can fix a bad patch
Yeah I asked this in the Pascal thread, that seems to make sense to me. And as far as third parties wanting to port these games, it may actually be a pretty big enticement for them to offer the exact same games that you can get on PS4 or XB1 on a Nintendo portable. I wouldn't bee terribly surprised to see a lot of third party ports around launch to test out this "full console games on the go" idea.

No they wouldn't. If they did, they will have done so at Vita's launch. And considering that Nintendo's relationship with them are pretty negative, I will be extremely shock if they do so.
 

EDarkness

Member
No they wouldn't. If they did, they will have done so at Vita's launch. And considering that Nintendo's relationship with them are pretty negative, I will be extremely shock if they do so.

I firmly believe that if there's a lot of hype surrounding the NX, then 3rd parties will be willing to try it out...at least initially. I imagine Ubisoft, Warner, and maybe Activision would give it a shot. EA may be on the fence, but I doubt Take 2 would bother. Initial hype would be everything.
 

ArtHands

Thinks buying more servers can fix a bad patch
I firmly believe that if there's a lot of hype surrounding the NX, then 3rd parties will be willing to try it out...at least initially. I imagine Ubisoft, Warner, and maybe Activision would give it a shot. EA may be on the fence, but I doubt Take 2 would bother. Initial hype would be everything.

There's no financial incentive for them to try at all. They will rather be focusing and investing on their next PS4 titles. Or PC/XBox. If we are very lucky, they might just let a bunch of interns create a quick port as a side project or something. Other than that I don't see why we should expect anything from them at all unless you are a masochist.
 

Oregano

Member
I firmly believe that if there's a lot of hype surrounding the NX, then 3rd parties will be willing to try it out...at least initially. I imagine Ubisoft, Warner, and maybe Activision would give it a shot. EA may be on the fence, but I doubt Take 2 would bother. Initial hype would be everything.

I wouldn't even rule Take 2 out completely. They've been saying some positive stuff and they put out X-Com on the Vita just a few months ago. NX could definitely get some stuff, even if not AAA stuff.
 
No they wouldn't. If they did, they will have done so at Vita's launch. And considering that Nintendo's relationship with them are pretty negative, I will be extremely shock if they do so.

Vita could not play straight ports of PS3/PS4 games, while the NX, based on what is heavily rumored, can play straight ports of XB1/PS4 games at 540p resolution. That's the main difference. Instead of being marketed as a companion game or a down port it's literally the exact same experience on a smaller screen.
 

dr_rus

Member
Alright yall, explain the Pascal and what it gon do

The only thing which can be said with 100% certainty for now is that it will bring the same feature set as what the current desktop GeForce 10 series cards have. Meaning that it will support whatever is there in DX12 which for a handheld is pretty cool.

Otherwise expect low power consumption (thanks to 16nm process) and overall performance somewhere between GTX 730 and GTX 750Ti most likely.
 

MuchoMalo

Banned
Vita could not play straight ports of PS3/PS4 games, while the NX, based on what is heavily rumored, can play straight ports of XB1/PS4 games at 540p resolution. That's the main difference. Instead of being marketed as a companion game or a down port it's literally the exact same experience on a smaller screen.

You're making a huge assumption here, that it'll have a 540p screen and and will simply upscale from 540p when in console mode. Didn't Matt say that the resolution would be lower than some would want, but higher than most expect? That sounds a bit more like 720p, especially for a tablet.

The only thing which can be said with 100% certainty for now is that it will bring the same feature set as what the current desktop GeForce 10 series cards have. Meaning that it will support whatever is there in DX12 which for a handheld is pretty cool.

Otherwise expect low power consumption (thanks to 16nm process) and overall performance somewhere between GTX 730 and GTX 750Ti most likely.

Vulkan is more relevant here. It's extremely likely that it'll be using a customized version of Vulkan.
 

Plasmids

Member
The only thing which can be said with 100% certainty for now is that it will bring the same feature set as what the current desktop GeForce 10 series cards have. Meaning that it will support whatever is there in DX12 which for a handheld is pretty cool.

Otherwise expect low power consumption (thanks to 16nm process) and overall performance somewhere between GTX 730 and GTX 750Ti most likely.

I'm happy with that for a handheld bring it on ;)
 

EDarkness

Member
There's no financial incentive for them to try at all. They will rather be focusing and investing on their next PS4 titles. Or PC/XBox. If we are very lucky, they might just let a bunch of interns create a quick port as a side project or something. Other than that I don't see why we should expect anything from them at all unless you are a masochist.

Companies will jump in if there's hype and try to capitalize on that. They did with the Wii and they'll do it again if the NX has a lot of hype for it when it's announced. The Wii U didn't have any hype at all and it was majorly underpowered for what companies planned on doing. However, if they can port their games over without too much trouble and they'll run on both the TV and portable, they may be willing to give it a shot provided there's enough excitement behind the system. General excitement drives sales and builds the initial userbase.
 

Kathian

Banned
People thinking it's going to run as well as a PS4 - it's built to run at 1080p as well. It's going to be

Id be surprised if for Zelda it was anything more than a res bump when docked for example.

Nintendo is making a console for their own games first and foremost. Plus there is the issue of memory which is not going to be close to a consoles.
 
You're making a huge assumption here, that it'll have a 540p screen and and will simply upscale from 540p when in console mode. Didn't Matt say that the resolution would be lower than some would want, but higher than most expect? That sounds a bit more like 720p, especially for a tablet.

The post that I'm referring to already discusses that assumption, so I thought it was clear that for the purposes of that particular discussion we're assuming 540p.

Also I don't know why people are assuming tablet, and Matt said that in a thread in which the context was people expecting 480 and hoping for 720, so 540p fits better with what he said.

Granted that was over a year ago and it's not the clearest leak so 540p is by no means certain, but that's why Thraktor's post mentions he's assuming 540p.
 

kami_sama

Member
You're making a huge assumption here, that it'll have a 540p screen and and will simply upscale from 540p when in console mode. Didn't Matt say that the resolution would be lower than some would want, but higher than most expect? That sounds a bit more like 720p, especially for a tablet.

I hope for a 720p screen. More would be a waste of resources for a portable, I think.

Vulkan is more relevant here. It's extremely likely that it'll be using a customized version of Vulkan.

Don't be so sure. If Nvidia's past actions are anything to go by, it'll be proprietary. But I think all consoles do that, so it wouldn't be a surprise.
 

MuchoMalo

Banned
There's no financial incentive for them to try at all. They will rather be focusing and investing on their next PS4 titles. Or PC/XBox. If we are very lucky, they might just let a bunch of interns create a quick port as a side project or something. Other than that I don't see why we should expect anything from them at all unless you are a masochist.

Actually, this isn't true. It's a bigger risk to not at least try at launch just in case it turns out to be a hit out of the gate. Of course, the attempts are such low effort that they flop anyway.
 
I wish someone would just ask Matt instead of speculating in his words for 2 years :p

Here's the opening line of that thread BTW

I've seen way too many times people asserting that they will be in shock if Nintendo's next handled will not have a 1080p screen and find this fact inexcusable because even mid-end phones have these kind of screens and yadda yadda.
 

MuchoMalo

Banned
The post that I'm referring to already discusses that assumption, so I thought it was clear that for the purposes of that particular discussion we're assuming 540p.

Also I don't know why people are assuming tablet, and Matt said that in a thread in which the context was people expecting 480 and hoping for 720, so 540p fits better with what he said.

Granted that was over a year ago and it's not the clearest leak so 540p is by no means certain, but that's why Thraktor's post mentions he's assuming 540p.

Eurogamer's mockup looks like a tablet.

I hope for a 720p screen. More would be a waste of resources for a portable, I think.



Don't be so sure. If Nvidia's past actions are anything to go by, it'll be proprietary. But I think all consoles do that, so it wouldn't be a surprise.

I assume you mean Nintendo. There are two reasons to suspect they'll base the API on Vulkan:

1.) Both Nintendo and Sony have based past APIs on OpenGL, and Vulkan is the new evolution of that.
2.) Nintendo is a Vulkan Working Group member.

Sony is one as well, which indicates that they both likely joined since it's pretty much their future. There's a chance that they won't support it in time for launch, though.

I wish someone would just ask Matt instead of speculating in his words for 2 years :p

Here's the opening line of that thread BTW

This supports what I said, I'd say.
 

dr_rus

Member
Vulkan is more relevant here. It's extremely likely that it'll be using a customized version of Vulkan.

I'm talking about features, using DX12 feature set as a reference is easier than specifying something out of the zoo of Vulkan extensions. I don't think that they'll use Vulkan (no real reason to on a fixed h/w platform) but it will be something rather similar in design most likely, as with all consoles since couple of gens ago.
 

Baseballjones

Neo Member
This will be the most powerful portable ever mass produced, I don’t understand why people are so disappointed with the rumored specs! If they have any merit to them a 540p or 720p pocket size gaming machine sounds amazing!! I love Nintendo and would certainly fork over $450 for a Scorpio killer console…. but that isn’t how Nintendo operates. Plus, this is a portable device from what we can tell and won’t have a primary focus on HDTV gaming. As long as the image is upscale to 1080P while docked I’m a day one!! Shoot might even get two!
 
Even taking into account single thread performance evidence would suggest that it's still worth making use of all eight cores. In my first scenario (500 mW) the drop in single thread performance from a 1.1GHz A53 core to a 550MHz A72 core is just 5%, and in the 1W scenario you actually see a 26% jump in single-thread performance from a 1.5GHz A53 core to a 1GHz A72.

Granted this is based on power curves from two different SoCs on different processes, and Geekbench FP isn't a perfect analog for gaming workloads (I've tried to find a suitable integer benchmark for comparison, but most, including Geekbench, have a heavy crypto component, which isn't of much use evaluating the general integer performance of cores with dedicated crypto blocks).

In any case I still believe Nintendo will end up using symmetric cores (except for the possibility of a couple of A35s reserved for OS duties), and they'll all be used both in handheld and docked mode.

I disagree. If we're talking about a Pascal GPU with 2 SMs running at around 600 MHz (which would give use about 300Gflops and would seem a reasonably likely scenario at the moment), then so long as the screen is around 540p, it would be absolutely capable of ports of the games listed. They'd have to be dialled down a little bit to run at native res (which they mostly don't on XBO and PS4), but the major roadblock would be third parties wanting to port the games to a Nintendo platform, not the capabilities of the hardware.

Games are being scaled up to target the XBO's performance at a minimum, there's no way they're going to scale them down further to put them on the NX if it's significantly weaker, if at all. If in the docked/TV mode it's at least on par with the XBO it probably has a better chance of happening.


This will be the most powerful portable ever mass produced, I don&#8217;t understand why people are so disappointed with the rumored specs! If they have any merit to them a 540p or 720p pocket size gaming machine sounds amazing!! I love Nintendo and would certainly fork over $450 for a Scorpio killer console&#8230;. but that isn&#8217;t how Nintendo operates. Plus, this is a portable device from what we can tell and won&#8217;t have a primary focus on HDTV gaming. As long as the image is upscale to 1080P while docked I&#8217;m a day one!! Shoot might even get two!


Because it's supposed to also be a gaming console and is looking grossly under powered for 2016 in that regard.
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
Even taking into account single thread performance evidence would suggest that it's still worth making use of all eight cores. In my first scenario (500 mW) the drop in single thread performance from a 1.1GHz A53 core to a 550MHz A72 core is just 5%, and in the 1W scenario you actually see a 26% jump in single-thread performance from a 1.5GHz A53 core to a 1GHz A72.
But a 1.5GHz A53 draws nowhere close to the power a 1GHz A72 does. Here's a same fab-node comparison (sorry about the random urls - still on a semblance of an internet) - the power/clock ratio between the two processors is approx 1/3: http://www.phonearena.com/image.php?m=Articles.Images&f=name&id=177342&popup=1 (just the 1st slide, disregard the rest).

The power drawn by a 1.5GHz A53 would roughly match that of a 500MHz A72, and the latter's IPC would clearly not be able to compensate that.
 
But a 1.5GHz A53 draws nowhere close to the power a 1GHz A72 does. Here's a same fab-node comparison (sorry about the random urls - still on a semblance of an internet) - the power/clock ratio between the two processors is approx 1/3: http://www.phonearena.com/image.php?m=Articles.Images&f=name&id=177342&popup=1
(just the 1st slide, disregard the rest).

The power drawn by a 1.5GHz A53 would roughly match that of a 500MHz A72, and the latter's IPC would clearly not be able to compensate that.



http://www.anandtech.com/show/9878/the-huawei-mate-8-review/3

You should read that before crossing off A72.
 

MuchoMalo

Banned
I'm talking about features, using DX12 feature set as a reference is easier than specifying something out of the zoo of Vulkan extensions. I don't think that they'll use Vulkan (no real reason to on a fixed h/w platform) but it will be something rather similar in design most likely, as with all consoles since couple of gens ago.

No console uses completely custom APIs. Sony and Nintendo have both based past APIs on OpenGL. Vulkan just happens to be the successor to OpenGL, and both companies just happen to be members of Vulkan's working group.
 

Thraktor

Member
But a 1.5GHz A53 draws nowhere close to the power a 1GHz A72 does. Here's a same fab-node comparison (sorry about the random urls - still on a semblance of an internet) - the power/clock ratio between the two processors is approx 1/3: http://www.phonearena.com/image.php?m=Articles.Images&f=name&id=177342&popup=1 (just the 1st slide, disregard the rest).

The power drawn by a 1.5GHz A53 would roughly match that of a 500MHz A72, and the latter's IPC would clearly not be able to compensate that.

I'm struggling to replicate your extrapolation from that graph. Even if we ignore that, given the cluster sizes, it seems to be a 28nm comparison (and relative power consumption at specific frequencies could change considerably with the jump down to finfet), we only seem to be given a single data point for each processor, at 1 GHz for the A53 and 1.2 GHz for the A72. Giving us a mW/GHz and Pstatic for each is all well and good, but they're only valid if you keep voltage constant, and as we don't know the voltage increase required to push the A53 to 1.5 GHz, or the voltage reduction possible by reducing the A72 to 1GHz, I can't see how we can definitively judge their relative power consumption even on 28nm, let alone 16nm.

I can actually get proper reliable like-for-like figures for both on 16FF+, though, by doing a little extrapolation of some of the data in Anandtech's Kirin 950 article. Although they don't explicitly give a power curve for the A53s on the Kirin 950, they do include the cores on a (somewhat confusing) graphic that compares perf/W to absolute performance in specint2000. The Y axis uses an arbitrary scale, but by referencing some of the other graphs in the article, I've been able to decipher it and recreate the power curve it would have been based on, here next to the A72 for comparison:

a72_a53_powercurves.png


The A72 curve is based on Anandtech's "target" trend line, as Huawei overvolt the A72s at 1.5GHz and under, and I don't see any reason to assume Nintendo would do the same.

My first response to this is that power consumption is down dramatically from the same cores on 14LPE, far more so than I would have assumed for two fabrication processes within the same generation, but this also could be down to implementation differences between Samsung and Hisilicon. In any case, taking the same metric as before (Geekbench FP), we get these results:

- Below 500mW, there's fairly meagre gain to using both clusters, so you might as well stick with the A53s.
- From 500mW to 850mW, it depends on how much you value aggregate multi-threaded performance versus peak single threaded performance. Benefit of using all cores is between 18% to 38% over this range, but you're dropping about 30% of peak single-thread performance.
- At 850mW, you get to the point where you're better off with just the A72 cluster over the A53 cluster (this is 1.1GHz A72 vs 2.1GHz A53).
- Above 850mW you get a fairly constant 40% overall performance boost from using all cores rather than just the A72s, and the loss of peak performance is around 25% (gradually declining as total power draw increases).

Basically, as power draw goes up, it makes more sense to use all cores (which is kind of obvious when I think of it, but at least now I have numbers). In any case it actually makes me more inclined towards an octo-core A53 setup, knowing how high it could clock even with an extremely low power allocation.
 

jdstorm

Banned
So, if they're going with the X1 and not a successor (or any kind of custom improvement), what kind of 3rd party game we can expect?
Like, would this and next year multiplaform games like Battlefield 1, Watch Dogs 2, SW: Battlefront etc. be any feasible on the handheld?

If these are the specs then those games would all be possible. I expect Watch Dogs 2 to come to it, but the others probably not.

I firmly believe that if there's a lot of hype surrounding the NX, then 3rd parties will be willing to try it out...at least initially. I imagine Ubisoft, Warner, and maybe Activision would give it a shot. EA may be on the fence, but I doubt Take 2 would bother. Initial hype would be everything.

There's no financial incentive for them to try at all. They will rather be focusing and investing on their next PS4 titles. Or PC/XBox. If we are very lucky, they might just let a bunch of interns create a quick port as a side project or something. Other than that I don't see why we should expect anything from them at all unless you are a masochist.

The great thing about EA's use of the Frostbyte engine is that it's used for every game and it is very scaleable. I'd imagine they'd want to release a few titles at launch if it was easy enough to port to the WiiU. Battlefield 1 definitely won't happen, and Titanfall is extremely unlikely, but Star Wars games have done well on Nintendo consoles in the past, and Xenoblade Chronicles X sold enough copies that a port of Mass Effect:Andromeda could be justified if it was cheap enough. Activision will probably bring Destiny, but apart from that I can't see much 3rd party support there initially.
 

Renekton

Member
Basically, as power draw goes up, it makes more sense to use all cores (which is kind of obvious when I think of it, but at least now I have numbers). In any case it actually makes me more inclined towards an octo-core A53 setup, knowing how high it could clock even with an extremely low power allocation.
Nintendo tends to prioritize in-house devs so it depends if their core guys can multi-thread that wide.
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
I'm struggling to replicate your extrapolation from that graph. Even if we ignore that, given the cluster sizes, it seems to be a 28nm comparison (and relative power consumption at specific frequencies could change considerably with the jump down to finfet), we only seem to be given a single data point for each processor, at 1 GHz for the A53 and 1.2 GHz for the A72. Giving us a mW/GHz and Pstatic for each is all well and good, but they're only valid if you keep voltage constant, and as we don't know the voltage increase required to push the A53 to 1.5 GHz, or the voltage reduction possible by reducing the A72 to 1GHz, I can't see how we can definitively judge their relative power consumption even on 28nm, let alone 16nm.
Curiously enough, your graph would confirm my rough extrapolation for the A53@1.5GHz vs A72@500MHz if you bothered to extrapolate the A72 plot down into the 500MHz range ; )

My first response to this is that power consumption is down dramatically from the same cores on 14LPE, far more so than I would have assumed for two fabrication processes within the same generation, but this also could be down to implementation differences between Samsung and Hisilicon. In any case, taking the same metric as before (Geekbench FP), we get these results:

- Below 500mW, there's fairly meagre gain to using both clusters, so you might as well stick with the A53s.
- From 500mW to 850mW, it depends on how much you value aggregate multi-threaded performance versus peak single threaded performance. Benefit of using all cores is between 18% to 38% over this range, but you're dropping about 30% of peak single-thread performance.
- At 850mW, you get to the point where you're better off with just the A72 cluster over the A53 cluster (this is 1.1GHz A72 vs 2.1GHz A53).
- Above 850mW you get a fairly constant 40% overall performance boost from using all cores rather than just the A72s, and the loss of peak performance is around 25% (gradually declining as total power draw increases).

Basically, as power draw goes up, it makes more sense to use all cores (which is kind of obvious when I think of it, but at least now I have numbers). In any case it actually makes me more inclined towards an octo-core A53 setup, knowing how high it could clock even with an extremely low power allocation.
The bolded part is what I've been talking about. First off, this is the wattage range I'd expect them to allot to CPU purposes, and second, my gut feeling is nintendo would value the guaranteed 30% single-thread performance more than the hypothetical 18%-38% multi-thread performance. At least I know I'd have done that if the decision was mine. But of course that's highly speculative, and this new nintendo could decide they would run their new CPUs at 1W just as well.
 
Status
Not open for further replies.
Top Bottom