Support NeoGAF

[Mr. Pointy]

How does X1 compare to the graphics cores on the latest iPad Pro/iPhone.

IPhone 6S uses a PowerVR GT7600 which can range from 250-384GFlops depending on the clock speed. The IPad Pro runs a custom PowerVR 7XT which is probably pushing up to 700-800GF.

I don't know how PowerVR's chips compare directly with Tegra.

 

[Schnozberry]

Having close to Wii U Specs in the handheld would be a fantastic jump over 3DS and should still have solid battery life. One thing I'm of two minds about would be the screen. I could see them going with a 1080p screen for streaming games from the console or for streaming video apps, and giving developers the flexibility to render as low as half resolution for games depending on how demanding they are graphically. On the other hand, if they aren't particularly concerned about offering premium experiences for mobile apps and services then they could just go low res and save the money, but it would be nice to have.

The console is somewhat of a mystery in my mind. It could very well be AMD if Nintendo's API is flexible enough, but it seems like an unnecessary complication unless Nintendo just got that much better of a price from AMD for a console part. If it's a Pascal based Tegra part and Nintendo is willing to budget somewhere around 30-40 watts for the console, then it could be quite powerful and efficient, which seems more in line with Nintendo's hardware philosophy. That would allow for 8 A72 Cores and 1TF+ from the GPU, a nice Wifi Radio, and hopefully they could fit it with an internal power supply in a reasonably sized box. In my mind that seems more likely than getting similar performance per watt from AMD, but I don't know what they could offer at 14nm at this point that might fit within Nintendo's budget.

 

[Schnozberry]

IPhone 6S uses a PowerVR GT7600 which can range from 250-384GFlops depending on the clock speed. The IPad Pro runs a custom PowerVR 7XT which is probably pushing up to 700-800GF.

I don't know how PowerVR's chips compare directly with Tegra.

The highest rated PowerVR part is the 16 cluster 7900GT and it's 512 GFlops. The A9X in the iPad Pro is 12 clusters, so you're probably looking at 400GFlops during the very short periods the device is allowed to run at it's maximum clock speed.

 

[cartographer]

It doesn't matter what res the test was under - that's the TDP under the heaviest-registered load of the chip.

Yes, I should have been more direct and not asked about resolution.

For the platform as a whole, right? Clocked to 1000 MHz on the GPU in the Shield TV.

I believe the person was originally referencing Thraktor's post, which referenced the power draw of just the GPU at 500 MHz. I think this is where some of the confusion is coming from: the box and full chip with GPU at 1000 MHz versus just GPU at 500 MHz.

I think he was just parroting Thraktor's post but misunderstood GPU versus the entire chip (or just didn't consider the entire chip in his post) and that lead to Zil's post.

I don't know what Parker would look like with the GPU clocked at 500 MHz.

 

[Bowl0l]

how would you feel if NX reveal was delayed because Nvidia chips have low yields?

imagine a $400 handheld with $5 eshop voucher for every $50 spent in the eshop.

 

[Deleted member 465307]

how would you feel if NX reveal was delayed because Nvidia chips have low yields?

imagine a $400 handheld with $5 eshop voucher for every $50 spent in the eshop.

What I'm imagining in that scenario is the final nail in the coffin of Japanese handheld sales. $400 just seems way too high for a portable.

 

[KingSnake]

how would you feel if NX reveal was delayed because Nvidia chips have low yields?

imagine a $400 handheld with $5 eshop voucher for every $50 spent in the eshop.

Why would it cost $400?

 

[Linkstrikesback]

how would you feel if NX reveal was delayed because Nvidia chips have low yields?

imagine a $400 handheld with $5 eshop voucher for every $50 spent in the eshop.

I'm imagining it. It looks a little like this.

 

[Mr. Pointy]

how would you feel if NX reveal was delayed because Nvidia chips have low yields?

imagine a $400 handheld with $5 eshop voucher for every $50 spent in the eshop.

Call it the Nintendo NX Founder's Edition. That'll make it sell.

 

[KingBroly]

At least if they had the Nvidia President present NX, he could fluff it up like no tomorrow about how awesome it is. People would go apeshit for NX.

 

[KingSnake]

At least if they had the Nvidia President present NX, he could fluff it up like no tomorrow about how awesome it is. People would go apeshit for NX.

Haha, yes. Maybe he could sell a $400 handheld.

 

[dampflokfreund]

If the Pascal Tegra draws 1.5 Watt at 500 Gflops, then it's possible to archieve Xbox One Level performance in a handheld. That is really crazy. Think about it: Playstation Vita needs nearly 4 Watts. So a handheld with Pascal and XBox One Level performance (1 TFLOPS) would be even 1 watt more efficient.

Nintendo, do it! A powerful handheld which can be connected to the TV is perfect to get away with Xbox performance in 2017. I believe even many Xbox/PS gamers would buy such a great device.

Then, 1-2 years after PS5/X2 Nintendo could release a powerful console.

There you have it guys, the perfect plan to save Nintendo. But I guess Kimishima already worked that out

 

[KingBroly]

Haha, yes. Maybe he could sell a $400 handheld.

'this is the most powerful handheld of all time. Of all time. Your phone won't be able to do this for at least a decade. A decade. Your phone won't be able to do this.'

 

[dampflokfreund]

'this is the most powerful handheld of all time. Of all time. Your phone won't be able to do this for at least a decade. A decade. Your phone won't be able to do this.'

That kind of marketing is exactly what Nintendo needs though

Wouldn't it still have to be under-clocked for battery purposes?

Yep. On the go. But what if you can plug in the handheld in a powered basestation, which allows the handheld to overclock itself.

 

[KingBroly]

That kind of marketing is exactly what Nintendo needs though

yeah, that's why I suggested they trot the guy out there, lol. He could've sold Wii U, lol

 

[AzaK]

Sorry if I've missed it, but if Tegra is for handheld, does this change what the home console might be made of?

 

[Kathian]

Their not making a handheld launch at the 3DS price any time soon.

 

[KingSnake]

Sorry if I've missed it, but if Tegra is for handheld, does this change what the home console might be made of?

Logic says that Nvidia would want both and it would be also easier for Nintendo. But you never know. Also, we don't really hear Nintendo talk about handheld and console separately, in all the communication they talk about NX as one device, so that also adds to the question.

 

[Possumowner]

'this is the most powerful handheld of all time. Of all time. Your phone won't be able to do this for at least a decade. A decade. Your phone won't be able to do this.'

Well I'm sold...when can I buy it 😃

 

[Deleted member 465307]

Their not making a handheld launch at the 3DS price any time soon.

If the 3DS price drop is coming, I see a max price of $230 (only for some kind of deluxe "XL" model) but really $200 (even for the XL). My hope is that Nintendo shifts the n3DS line down $30-$40 by the time a portable NX launches. If they only drop it by $20 like the 2DS, then perhaps it'll enter higher than $200.

I'm not sure how this Tegra rumor will affect price, but I'm expecting/hoping a Nintendo portable lineup during holiday 2017 to look like $80/$120/$150 (if they can manage it) for the 3DS stuff and NX at $170-$220. And probably an NX console at $300±$50 if they do multiple models ($250/$300?).

 

[blu]

Yes, I should have been more direct and not asked about resolution.

For the platform as a whole, right? Clocked to 1000 MHz on the GPU in the Shield TV.

I believe the person was originally referencing Thraktor's post, which referenced the power draw of just the GPU at 500 MHz. I think this is where some of the confusion is coming from: the box and full chip with GPU at 1000 MHz versus just GPU at 500 MHz.

I think he was just parroting Thraktor's post but misunderstood GPU versus the entire chip (or just didn't consider the entire chip in his post) and that lead to Zil's post.

I don't know what Parker would look like with the GPU clocked at 500 MHz.

Keep in mind what Thraktor posted was the result from one test, namely Manhattan offscreen 1080p, where the test was set up in such a way so that the performance of TX1 would match that of the A8X, ergo the supposed halving of the GPU clock (emphasis on supposed, the clock was never quoted in the article). We don't know if that is the maximum TDP of the GPU at that supposed clock, unless we assume Manhattan 1080p to be the ultimate GPU load. And generally, the test goal was to show Maxwell's efficiency superiority over IMG's GX6850. Which is a bit funny claim to make, as unless apple published the exact pinout of the A8X, NV engineers could only guess what rails they were measuring on the ipad board (notice the funky fluctuations in the supposed GX6850 measurements, compared to the flattish Kepler band for the exact same workload?). In contrast, what NV could have demonstrated without actual doubt would have been the doubled GPU efficiency from Kepler to Maxwell - something they just made a claim on a slide, and something which a rudimentary bulk-SoC check does not confirm for fp32:

$ echo "scale=4;  (365 / 11)" | bc
33.1818
$ echo "scale=4;  (512 / 15)" | bc
34.1333

 

[Toadthemushroom]

The X1 in the Pixel-C was clocked at 850MHz it seems, would it have been the same on the 9.7' iPad Pro if it didn't have a PowerVR 7XT?

If you consider the 2DS and 3DS XL form factors - which are generally thicker and larger slabs (well, a 3DS XL if it was built like a 2DS, as I don't think dual screen is returning next gen), what sort of clock would Nintendo be able to get away with for an X1 or better, given potential throttling and power consumption constraints?

I'm wondering as the Vita-1000 was a pretty close match for the iPad 3 despite the different form factors.

The highest rated PowerVR part is the 16 cluster 7900GT and it's 512 GFlops. The A9X in the iPad Pro is 12 clusters, so you're probably looking at 400GFlops during the very short periods the device is allowed to run at it's maximum clock speed.

I thought one of the key advantages of PowerVR's designs is they don't throttle over sustained periods of loads? Same goes for Apple's CPU designs - the Typhoon Cores in the very-thin iPod Touch 6 didn't throttle at all at 1.1Ghz, and the Twister cores in the 6s/5se/iPad also maintain their performance.

 

[ozfunghi]

Sorry to burst your bubble but tegra X1 draws 20W while gaming. 500 Gflops at 1.5W is a pipe dream.

Yes, I should have been more direct and not asked about resolution.

For the platform as a whole, right? Clocked to 1000 MHz on the GPU in the Shield TV.

I believe the person was originally referencing Thraktor's post, which referenced the power draw of just the GPU at 500 MHz. I think this is where some of the confusion is coming from: the box and full chip with GPU at 1000 MHz versus just GPU at 500 MHz.

I think he was just parroting Thraktor's post but misunderstood GPU versus the entire chip (or just didn't consider the entire chip in his post) and that lead to Zil's post.

I don't know what Parker would look like with the GPU clocked at 500 MHz.

Yes I was only talking about the GPU, which is what the topic is about and what the discussion was about compared to the WiiU GPU and the XBO GPU. Sorry if that wasn't clear. Maybe I used the word chip where I should have said gpu perhaps.

But if the test was rigged or if the numbers don't add up to actual performance, then I guess it doesn't matter what I said.

 

[atbigelow]

I'm wondering as the Vita-1000 was a pretty close match for the iPad 3 despite the different form factors.

PSV is also super duper downclocked.

 

[blu]

Yes I was only talking about the GPU, which is what the topic is about and what the discussion was about compared to the WiiU GPU and the XBO GPU. Sorry if that wasn't clear. Maybe I used the word chip where I should have said gpu perhaps.

But if the test was rigged or if the numbers don't add up to actual performance, then I guess it doesn't matter what I said.

To clarify: I'm not saying the measured numbers for the TX1 were wrong. I'm saying:

1) Manhattan off-screen can be used to get an idea, perhaps a guideline, but to tell the actual TDP one needs much more than a sole Manhattan. The maximum TDP for the GPU alone at the supposed clock of 500MHz could be 2W for all we know. Or even above 2W.

2) The measurements of the A8X GPU cannot be taken without a dose of salt - NV measured something, but till we have authoritative info on the A8X (something only apple and their fab partners know), NV have measured an arbitrary rail on the A8X SoC pinout.

 

[ozfunghi]

To clarify: I'm not saying the measured numbers for the TX1 were wrong. I'm saying:

1) Manhattan off-screen can be used to get an idea, perhaps a guideline, but to tell the actual TDP one needs much more than a sole Manhattan. The maximum TDP for the GPU alone at the supposed clock of 500MHz could be 2W for all we know. Or even above 2W.

2) The measurements of the A8X GPU cannot be taken without a doze of salt - NV measured something, but till we have authoritative info on the A8X (something only apple and their fab partners know), NV have measured an arbitrary rail on the A8X SoC pinout.

Ok thanks. Personally, i wasn't really interested in the comparisson with the PowerVR.

Thraktor said 1.5W for only the GPU puts it in handheld territory. Is that reasonable or on the high end of power consumption? Is 2+W which you aren't rulling out at that clock, still reasonable?

Am i reading your post correctly (2 posts ago) that the Pascal gpu will be 1.36 times more efficient than the Maxwell gpu? So if the handheld gets a Pascal based gpu, it would either draw 1.36 x less, or perform 1.36 x better?
In which case, should TX1 be clocked at 400MHz (due to battery drain), and still manage to reach around 200Gflops, the Pascal based GPU be able to reach +/- 270 gflops?

 

[Toadthemushroom]

PSV is also super duper downclocked.

Mm, yes, both Vita and iPad 3 were clocked at ~200MHz. I think iPad 3 had its SGX-543MP4 clocked a bit higher).

That's actually what I'm getting at, as crazy as it sounds.

If a handheld with a form factor similar to the Vita can produce similar results to a 9.7" tablet, and the Pixel-C has its Tegra X1 clocked at 850MHz, what does that mean for an NX handheld if it had a form factor somewhere close to the 3DS XL (unfolded) and 2DS in size and thickness? Would it be able to do 850MHz, or am I being naïve?

 

[Candescence]

This whole discussion is honestly making me appreciate how we've reached the point where we might actually be able to pack roughly above-last-gen or even current-gen horsepower into DS-esque form factor, give or take size and battery power requirements. That's kind of fucking insane.

 

[cartographer]

Keep in mind what Thraktor posted was the result from one test, namely Manhattan offscreen 1080p, where the test was set up in such a way so that the performance of TX1 would match that of the A8X, ergo the supposed halving of the GPU clock (emphasis on supposed, the clock was never quoted in the article). We don't know if that is the maximum TDP of the GPU at that supposed clock, unless we assume Manhattan 1080p to be the ultimate GPU load. And generally, the test goal was to show Maxwell's efficiency superiority over IMG's GX6850. Which is a bit funny claim to make, as unless apple published the exact pinout of the A8X, NV engineers could only guess what rails they were measuring on the ipad board (notice the funky fluctuations in the supposed GX6850 measurements, compared to the flattish Kepler band for the exact same workload?). In contrast, what NV could have demonstrated without actual doubt would have been the doubled GPU efficiency from Kepler to Maxwell - something they just made a claim on a slide, and something which a rudimentary bulk-SoC check does not confirm for fp32:

$ echo "scale=4;  (365 / 11)" | bc
33.1818
$ echo "scale=4;  (512 / 15)" | bc
34.1333

Awesome. Thank you for the clarification.

 

[James Scott]

This whole discussion is honestly making me appreciate how we've reached the point where we might actually be able to pack roughly above-last-gen or even current-gen horsepower into DS-esque form factor, give or take size and battery power requirements. That's kind of fucking insane.

We probably shouldn't expect more than XBox 360 specs on the handheld. Just because it's possible doesn't mean they'll do it

 

[LewieP]

You can't really directly compare it to the PS4/Xbox One/Wii U without taking into account that it's most likely going to be using a more modest resolution display (I agree with the notion of 540p being the sweet spot), which is less demanding on GPU resources than the respective resolutions the various home consoles output at.

I think it's entirely feasible that a game targeting 1080p/60fps on PS4/Xbox One could also run on a gaming handheld at 540p/30fps with a few other minor compromises in areas such as IQ, effects, lighting, asset quality, texture detail, physics complexity etc. At least from a GPU perspective, cpu I'm not quite so sure about.

 

[Durante]

This whole discussion is honestly making me appreciate how we've reached the point where we might actually be able to pack roughly above-last-gen or even current-gen horsepower into DS-esque form factor, give or take size and battery power requirements. That's kind of fucking insane.

It's not so insane, particularly the "N-times 3DS" comparisons are also a function of 3DS being really fucking slow already when it was released. Especially that CPU. Oh my god.

The lesson being that just because something is possible, don't expect Nintendo to do it
Hopefully the goal of a unified library encourages them to not go all that withered on the handheld this time around.

 

[Mr. Pointy]

Yeah but can the handheld run Crysis?

 

[what-ok]

Is the info "rumors" stating that the NX is a system and or OS that works with a new device to allow for on the go play and some sort of home hub/console? I skimmed the thread and I'm not really that concerned with the tech numbers just interested in the concept.

 

[QaaQer]

SemiAccurate is one of the very worst tech sites on the internet. Even if it stumbles across something which has some nugget of truth in it, if it as much as tangentially involves Nvidia any shred of objectivity flies out of the window in the face of Charlie Demerjian's inexplicable long-term hatred of the company.

Actually, calling it a "tech site" does a disservice to real tech sites. It's more like a fanboy blog.

I'm sure you have come to this conclusion after due diligence? Maybe a share a link or two while you are assassinating someone's character.

 

[LewieP]

Is the info "rumors" stating that the NX is a system and or OS that works with a new device to allow for on the go play and some sort of home hub/console? I skimmed the thread and I'm not really that concerned with the tech numbers just interested in the concept.

You are better off asking Nintendo that.

They will probably decline to comment.

 

[KingSnake]

Is the info "rumors" stating that the NX is a system and or OS that works with a new device to allow for on the go play and some sort of home hub/console? I skimmed the thread and I'm not really that concerned with the tech numbers just interested in the concept.

There's no reliable info yet on the device type/types.

 

[Candescence]

It's not so insane, particularly the "N-times 3DS" comparisons are also a function of 3DS being really fucking slow already when it was released. Especially that CPU. Oh my god.

The lesson being that just because something is possible, don't expect Nintendo to do it
Hopefully the goal of a unified library encourages them to not go all that withered on the handheld this time around.

Well, I'm mainly just appreciating how far we've come with tech. Christ, ten years ago the 360 only just came out with the Wii and PS3 not too far behind later that year, and here we are discussing the likely possibility of a Nintendo handheld about the size of the DS Lite easily leapfrogging the 360 in terms of horsepower when the DS itself was basically an N64+.

There are still a lot of unknown variables in this particular equation, but I do imagine Nintendo wants to close the gap between handheld and console as much as possible while keeping the handheld affordable and the console reasonably powerful.

 

[GhostTrick]

It's not so insane, particularly the "N-times 3DS" comparisons are also a function of 3DS being really fucking slow already when it was released. Especially that CPU. Oh my god.

The lesson being that just because something is possible, don't expect Nintendo to do it
Hopefully the goal of a unified library encourages them to not go all that withered on the handheld this time around.

I disagree for the 3DS statement though. Sure it could've been better but back in the days the ARM boom had yet to happen. I think they were stuck with the tech available. Sony were too. Which explains low clocks and such.

But yeah, for this generation, they have the tech and money to offer a handheld performing beyond 360 at 540p. Now will they ?

I hate to be that guy, but yeah, it's Nintendo and who knows what bullshit they can pull. Even though sanity would tell us they'd have to try really hard to not outperform Vita (28gflops) by 3 times (GPD XD for exemple is 72gflops) and the reality of the market should tell us they cant come with soemthing lower than ARM A53. But we're talking about a company who could still pull bullshit with a Cortex A7 CPU and a 40gflops GPU in a handheld.



After all, back in the Wii U days, people were thinking about a 1 Tflops AMD GPU and a newer Power PC design... To end up with a 5 times slower GPU and 3 Wii CPUs ductaped together and overclocked.

 

[Schnozberry]

I thought one of the key advantages of PowerVR's designs is they don't throttle over sustained periods of loads? Same goes for Apple's CPU designs - the Typhoon Cores in the very-thin iPod Touch 6 didn't throttle at all at 1.1Ghz, and the Twister cores in the 6s/5se/iPad also maintain their performance.

Apple's CPU cores hardly throttle at all. PowerVR doesn't throttle as much as the competition, but the chips aren't immune from it.

 

[blu]

Ok thanks. Personally, i wasn't really interested in the comparisson with the PowerVR.

I know, I just went off on a tangent.

Thraktor said 1.5W for only the GPU puts it in handheld territory. Is that reasonable or on the high end of power consumption? Is 2+W which you aren't rulling out at that clock, still reasonable?

I depends what the SoC vendor could cram into 2-3W (which is normally the TDP for such devices) and what nintendo would accept of that, e.g. while a phablet vendor might be fine with a 3W SoC, nintendo might not be. Generally, an A53 cluster can run really well in the sub-W range @ 16FF. So a A53 cluser and the speculated 500MHz Maxwell from TX1 but @16nm could actually do it. It's all up to nintendo what they want.

Am i reading your post correctly (2 posts ago) that the Pascal gpu will be 1.36 times more efficient than the Maxwell gpu? So if the handheld gets a Pascal based gpu, it would either draw 1.36 x less, or perform 1.36 x better?

Correct.

In which case, should TX1 be clocked at 400MHz (due to battery drain), and still manage to reach around 200Gflops, the Pascal based GPU be able to reach +/- 270 gflops?

Hypothetically, yes.

 

[Peltz]

Is the info "rumors" stating that the NX is a system and or OS that works with a new device to allow for on the go play and some sort of home hub/console? I skimmed the thread and I'm not really that concerned with the tech numbers just interested in the concept.

The concept is far more interesting to me than the specs as well.

I feel like we've reached a point where Nintendo is consistently putting out ageless games on Wii U. The same games with some cleaner IQ isn't really going to excite me at this point.

The same games on the go also will not be exciting to me at all.

I'd love if they really had a brand new concept all over again like the Wii and DS innovations in years passed.

I'd love if I could control games with my brainwaves + eye tracking or something (yes this is totally a pipe dream and likely nothing like what we are getting).

 

[Peru]

The concept is far more interesting to me than the specs as well.

I feel like we've reached a point where Nintendo is consistently putting out ageless games on Wii U. The same games with some cleaner IQ isn't really going to excite me at this .

Nintendo have been reinventing old properties and creating new ones on the wii u better than in decades.

 

[MuchoMalo]

I disagree for the 3DS statement though. Sure it could've been better but back in the days the ARM boom had yet to happen. I think they were stuck with the tech available. Sony were too. Which explains low clocks and such.

But yeah, for this generation, they have the tech and money to offer a handheld performing beyond 360 at 540p. Now will they ?

I hate to be that guy, but yeah, it's Nintendo and who knows what bullshit they can pull. Even though sanity would tell us they'd have to try really hard to not outperform Vita (28gflops) by 3 times (GPD XD for exemple is 72gflops) and the reality of the market should tell us they cant come with soemthing lower than ARM A53. But we're talking about a company who could still pull bullshit with a Cortex A7 CPU and a 40gflops GPU in a handheld.



After all, back in the Wii U days, people were thinking about a 1 Tflops AMD GPU and a newer Power PC design... To end up with a 5 times slower GPU and 3 Wii CPUs ductaped together and overclocked.

Wii U turned out the way it did for backwards compatibility. For them to make NX that weak, they'd either be going for GBA-level battery life or intentionally trying to make it as weak as possible just because, and I'm not even sure if the former can be done due to how low it would need to scale down. Assuming a Tegra K1 (nothing before that would be cost effective of efficient), they'd need to clock it at 100MHz. Would they really go that far? I can understand how you don't want to burned again and choose the remain skeptical though.

Also, keep in mind that 3DS was originally meant to use a Tegra, so it's possible that they just didn't have time to choose a better design after that fell through.

 

[Vena]

I depends what the SoC vendor could cram into 2-3W (which is normally the TDP for such devices) and what nintendo would accept of that, e.g. while a phablet vendor might be fine with a 3W SoC, nintendo might not be. Generally, an A53 cluster can run really well in the sub-W range @ 16FF. So a A53 cluser and the speculated 500MHz Maxwell from TX1 but @16nm could actually do it. It's all up to nintendo what they want.

I wonder how much stock we can put in the Pascal rumor. Seems like such a weird thing to piggy-back off of and specify Pascal. But then again, the NX has seemingly driven people/the internet mad.

 

[MuchoMalo]

I wonder how much stock we can put in the Pascal rumor. Seems like such a weird thing to piggy-back off of and specify Pascal. But then again, the NX has seemingly driven people/the internet mad.

None. Put no stock into it.

 

[Skittzo0413]

I'd love if I could control games with my brainwaves + eye tracking or something (yes this is totally a pipe dream and likely nothing like what we are getting).

I don't know about brainwaves (at least in 2017) but eye tracking is something which Nintendo has been toying with for a while now, seen in the new 3DS as well as at least one patent application. I'm of the opinion that eye/head/face tracking will be a major feature in both the handheld and console (assuming we get both, which to be honest I'm starting to doubt) and it will not be a gimmick which greatly increases the cost of the console. All it really needs is a decent camera and some processing power set aside specifically for the tracking purposes.

That could be a reason why Nintendo might be looking at "industry leading chips" this time. Meaning, if their primary feature uses a decent chunk of processing (face/eye tracking software is fairly computationally intensive) at all times, then they will need a very good CPU to constantly lend its processing power to the non-game specific function. This could also go in line with what LCGeek said about the CPU.

This is a GPU thread though so I digress.

 

[Graphics Horse]

I depends what the SoC vendor could cram into 2-3W (which is normally the TDP for such devices) and what nintendo would accept of that, e.g. while a phablet vendor might be fine with a 3W SoC, nintendo might not be. Generally, an A53 cluster can run really well in the sub-W range @ 16FF. So a A53 cluser and the speculated 500MHz Maxwell from TX1 but @16nm could actually do it. It's all up to nintendo what they want.
.

Is ninendo's stance on that a heat dissipation issue or a battery cost issue? I'm just interested if they could be considering something larger than normal, also how much can thickness of the device improve heat dissipation?

 

[Thraktor]

It won't be. According to this table:

Maxwell's cores * clock / watt is 3072 * 948 / 250, whereas
Pascal's cores * clock / watt is 3584 * 1328 / 300, so

$ echo "scale=4; (3584 * 1328 / 300) / (3072 * 948 / 250)" | bc
1.3619

Apparently the necessity for fp64 functionality (read: corresponding transistors) is highly questionable in a handheld, but unless NV already have a reduced-fp64 version of the design for TP1, I wouldn't expect them to customize the architecture for nintendo to such an extent (read: semi-custom vs completely custom).


It doesn't matter what res the test was under - that's the TDP under the heaviest-registered load of the chip.

Keep in mind what Thraktor posted was the result from one test, namely Manhattan offscreen 1080p, where the test was set up in such a way so that the performance of TX1 would match that of the A8X, ergo the supposed halving of the GPU clock (emphasis on supposed, the clock was never quoted in the article). We don't know if that is the maximum TDP of the GPU at that supposed clock, unless we assume Manhattan 1080p to be the ultimate GPU load. And generally, the test goal was to show Maxwell's efficiency superiority over IMG's GX6850. Which is a bit funny claim to make, as unless apple published the exact pinout of the A8X, NV engineers could only guess what rails they were measuring on the ipad board (notice the funky fluctuations in the supposed GX6850 measurements, compared to the flattish Kepler band for the exact same workload?). In contrast, what NV could have demonstrated without actual doubt would have been the doubled GPU efficiency from Kepler to Maxwell - something they just made a claim on a slide, and something which a rudimentary bulk-SoC check does not confirm for fp32:

$ echo "scale=4;  (365 / 11)" | bc
33.1818
$ echo "scale=4;  (512 / 15)" | bc
34.1333

It's worth keeping in mind the difference between perf/W at the highest achievable clocks and perf/W at the kind of clocks you would be likely to see in a handheld. When Nvidia were claiming a 2x perf/W increase from TK1 to TX1, they were of course cherry-picking the test, but they also would have cherry-picked the point at which the perf/W gap was the greatest, and that point is likely to be a lot closer to the clocks we'd be looking at for a handheld than the max achievable clocks. This isn't to say that I think the real-world perf/W ratio between the TX1 and TK1 in a handheld would be anywhere close to 2, but I do think it's at least somewhere above 1.

Ditto for the difference between Maxwell and Pascal, comparing the performance of GM200 versus GP100 at 250W+ envelopes won't necessarily translate to the benefits when moving from one architecture to the other in a 2W TDP, which could be much higher or much lower for all we know. Incidentally, on the subject of the transistor implication of Pascal's increased FP64 support, it appears that GP100 is the only Pascal die that's actually making the shift to 64 ALU SMs with half-rate FP64 support, and GP104 (and presumably all smaller chips) are apparently sticking to the 128 ALU Maxwell SMs with very limited FP64 support. This seems to be borne out with GP104 having a similar ALU density per billion transistors to Maxwell, and with GTX 1080 benchmarks seeing a performance boost roughly in line with raw floating point performance growth (i.e. indicating few architectural changes). This could actually be a good thing for Nintendo, as they could in theory get a 28nm "Maxwell" chip for a home console and a 16nm "Pascal" chip for the handheld with the same SMs across both.

 

[ozfunghi]

awesomeness

thanks

I wonder how much stock we can put in the Pascal rumor. Seems like such a weird thing to piggy-back off of and specify Pascal. But then again, the NX has seemingly driven people/the internet mad.

With Nintendo's track record, under normal circumstances i'd never consider them going for it. But the "industry leading chip" and "modern custom chip" lend more credibility to it being an option.