Support NeoGAF

[TheGreatMightyPoo]

How will you play an online game on the go though

"On the go" for me means anywhere in a house with internet.

 

[Cerium]

I'm looking to things and trying to see what's going on. I had a handful of contacts sharing Pascal with me over the summer. Trying to figure out what's going on.

I appreciate your efforts but your reporting on NVidia Pascal in the final system formed the basis of much of your credibility regarding Switch. It's the factoid that we tend to cite the most when we're asked what you've gotten right. It would be very unfortunate if it turned out to be Maxwell. I personally would feel disappointed and misled.

 

[Kimawolf]

You don't seem to know much more "stuff", than the people upon whom you purport to look down.

Think he means maxwell based gpus far surpass ps4 and xb1. And just about all of them do.

 

[Thraktor]

I have a few immediate thoughts after reading through the article:

• Firstly, it's worth noting the difference between Maxwell and Pascal is almost entirely down to the manufacturing process. Maxwell was made on 28nm (and in the case of the TX1, 20nm) whereas Pascal is made on 16nm. The actual architectural difference between the two is minimal, and aside from improved color buffer compression, largely irrelevant for a device like the Switch.
• Despite that, the article never makes any mention of the manufacturing process. I find that extremely strange, as it's obviously the defining difference between the two sets of GPUs.
• In fact, the article gets the difference between the two completely the wrong way around, saying "Nintendo’s box is relatively small, and so it has to fit into the heat profile of a portable device, rather than a set-top box. That’s another reason that explains the older Maxwell technology, as opposed to the Pascal’s state-of-the-art tech." Pascal is literally a more power efficient version of Maxwell, so the incentive would be the other way around.
• The author says "we expect the Nintendo Switch to be more than 1 teraflop in performance", which is notably higher than even those of us who were expecting Pascal were considering (I literally posted earlier today with a 500-750 Gflop estimate). If this is a Maxwell chip, then that would mean at least 4 SMs (512 "CUDA cores") at 1GHz, as they're not going to be able to push much past that on 28/20nm. This is a much larger GPU than most people would have been expecting.

I see a few different scenarios here:

1. The Switch SoC uses Maxwell at 20nm, and simply has a much larger GPU than anticipated to account for the performance.
2. Nintendo looked at the feature-set planned for Pascal when design started, realised that the new features were largely irrelevant, and decided that they would save time and just use a straight-forward die shrink of Maxwell to 16nm instead. That would technically be a Maxwell GPU, but would be almost completely indistinguishable from Pascal in terms of performance.
3. The sources are wrong about Maxwell, the 1 Tflop performance, or both.

Basically, if you're to take the article as being accurate, then the only worthwhile takeaway is this quote:

we expect the Nintendo Switch to be more than 1 teraflop in performance

A Maxwell Tflop is identical to a Pascal Tflop, and it's largely irrelevant to us whether they achieved that by using a larger Maxwell GPU on 20nm/28nm at a lower clock or a smaller Pascal GPU on 16nm at a higher clock.

 

[Josh7289]

Using older, cheaper, and better-understood hardware is Nintendo's modus operandi, though, so I could believe it.

 

[ChrisMahoney]

If the dock rumor is true does it really matter much?
It means you can get the most out of the device without having to make the battery life terrible or the TV mode ugly

Pascal would mean better battery life

 

[Xdrive05]

Ugh. This thread reads sort of like the Wii-U CPU scan thread, where we figured out it was three Gamecube cores on a die, or whatever that case was.

Feels like Nintendo console threads are always like this. People giving "reasonable expectations" and then the truth turns out to be way weaker.

However, in this case, it's a whole lot of fucking nothing to be complaining about. This news doesn't really say much about the performance relative to speculation. Hell. It goes even further to support the "reasonable" speculations we've been making.

Honestly I think the salt in this thread is just people jumping the gun based on getting totally burned the last two Nintendo consoles as far as specs go. Folks, this isn't the same kind of thing... yet. Hang in there.

 

[Inuhanyou]

I would actually be pretty pissed off if true. Why the hell am I spending money on this again if it's not that much better?

These kinds of assumptions and leaps of logic confuse me. So...even if this information is correct....

How does that even equate to Switch not being way better than Wii U? What is your definition of 'not that much better'? Did you forget literally all the other parts of the console?

 

[Jaded Alyx]

Sooo what does this change? Nothing? Alllright then.

 

[Skittzo0413]

1024 FLOPS/cycle is nonsense. It's literally "floating point operations per second per cycle." wtf is that?

Yeah not sure how she can claim those specs are 90% correct when there's nonsense like that.


Either way this article is essentially nonsense and tells us absolutely nothing we didn't already know. Nintendo would be insane not to use a 16nm process, regardless of Maxwell or Pascal, and since those processes have been available for quite a while it's unlikely they didn't take advantage of that. More performance per watt and lower in price. There's literally no downside.

So even if it is Maxwell at 16nm this article tells us nothing at all.

 

[EloquentM]

This is a Takeda issue, so don't expect anything different from Nintendo hardware wise until he is gone.

You know, your tag doesn't inspire more confidence in your trolling method

 

[Rodin]

Oh, Maxwell over Pascal would definitely be disappointing, but most of the concern trolling is coming from those who don't actually know why. There are obviously knowledgeable, rational folks who have legitimate reasons for being a bit bummed about this rumor.

Yeah i absolutely agree with this.

So the devkit thread says 1024FLOPS/cycle- is that not 1TFlop? That makes no mention of FP16 vs FP32 either. What the hell is going on?

The article largely doesn't make sense, which is why i immediately dismissed it at first, but considering we're talking about mobile hardware it's pretty obvious that the "1tlop/close to xbone" claims are related to fp16 numbers. The comparison with the 6tflops Scorpio is just poor writing by someone who isn't particularly tech savy.

Eh? Takeda didn't design the Switch.

The guy who designed the Switch did that by following the base guideline of the guy in charge of the hardware at Nintendo. When the Switch was originally conceived, that person was still Takeda. Reading about Maxwell and A57 means that the design was locked in very early for some reason.

Sooo what does this change? Nothing? Alllright then.

Battery life and heat, possibly standalone handheld performances.


EDIT: we already have a rumored ballpark for pricing: 249$ for the basic sku and 299 for the version with more storage and Splatoon packed in. These are from LKD.

 

[TheGreatMightyPoo]

You know, your tag doesn't inspire more confidence in your trolling method

lol

 

[eighty(one)]

Fully expecting under $200 price point for launch.

No way.

3DS (I love it) is old and archaic now, still sells for $200. There's no way this things with better components comes at that price.

 

[Hatty]

I would actually be pretty pissed off if true. Why the hell am I spending money on this again if it's not that much better?

Probably for video games you wouldn't have access to otherwise

 

[Shahadan]

"On the go" for me means anywhere in a house with internet.

Nintendo definitely could use "Reclaim your throne" as the Switch slogan

 

[Kouriozan]

Sooo what does this change? Nothing? Alllright then.

Looks to be that way, at least for the consummers.

 

[Mr. Wonderful]

These kinds of assumptions and leaps of logic confuse me. So...even if this information is correct....

How does that even equate to Switch not being way better than Wii U? What is your definition of 'not that much better'? Did you forget literally all the other parts of the console?

No. It just doesn't provide any additional value to me over what my Wii U was for my use cases.

 

[Shahadan]

I have a few immediate thoughts after reading through the article:

• Firstly, it's worth noting the difference between Maxwell and Pascal is almost entirely down to the manufacturing process. Maxwell was made on 28nm (and in the case of the TX1, 20nm) whereas Pascal is made on 16nm. The actual architectural difference between the two is minimal, and aside from improved color buffer compression, largely irrelevant for a device like the Switch.
• Despite that, the article never makes any mention of the manufacturing process. I find that extremely strange, as it's obviously the defining difference between the two sets of GPUs.
• In fact, the article gets the difference between the two completely the wrong way around, saying "Nintendo’s box is relatively small, and so it has to fit into the heat profile of a portable device, rather than a set-top box. That’s another reason that explains the older Maxwell technology, as opposed to the Pascal’s state-of-the-art tech." Pascal is literally a more power efficient version of Maxwell, so the incentive would be the other way around.
• The author says "we expect the Nintendo Switch to be more than 1 teraflop in performance", which is notably higher than even those of us who were expecting Pascal were considering (I literally posted earlier today with a 500-750 Gflop estimate). If this is a Maxwell chip, then that would mean at least 4 SMs (512 "CUDA cores") at 1GHz, as they're not going to be able to push much past that on 28/20nm. This is a much larger GPU than most people would have been expecting.

I see a few different scenarios here:

1. The Switch SoC uses Maxwell at 20nm, and simply has a much larger GPU than anticipated to account for the performance.
2. Nintendo looked at the feature-set planned for Pascal when design started, realised that the new features were largely irrelevant, and decided that they would save time and just use a straight-forward die shrink of Maxwell to 16nm instead. That would technically be a Maxwell GPU, but would be almost completely indistinguishable from Pascal in terms of performance.
3. The sources are wrong about Maxwell, the 1 Tflop performance, or both.

Basically, if you're to take the article as being accurate, then the only worthwhile takeaway is this quote:



A Maxwell Tflop is identical to a Pascal Tflop, and it's largely irrelevant to us whether they achieved that by using a larger Maxwell GPU on 20nm/28nm at a lower clock or a smaller Pascal GPU on 16nm at a higher clock.

This should be quoted in the OP tbh. Article is weird and this post is well constructed and argumented.

 

[ultrazilla]

Not worried at all about the Switch's performance. Nintendo learned their lesson with the Wii U. Here's some links to help put the fear to rest if you think this will be Wii U 2.

http://www.ibtimes.com/nintendo-swi...-5-titles-nvidia-gpu-ram-size-details-2439551

https://blogs.nvidia.com/blog/2016/10/20/nintendo-switch/

http://www.ign.com/articles/2016/11/14/nvidia-ceo-calls-nintendo-switch-design-aground-breakinga

http://wccftech.com/nintendo-switch-nvidia-tegra-pascal/

 

[MuchoMalo]

This is a Takeda issue, so don't expect anything different from Nintendo hardware wise until he is gone.

Welp. Sorry, but that man is a coward lol. If Nintendo wants to keep that up, they need to stop pretending to be one of the big boys; otherwise, boot his ass for someone with confidence and vision.

 

[Oersted]

I appreciate your efforts but your reporting on NVidia Pascal in the final system formed the basis of much of your credibility regarding Switch.

What? No. His past is the basis of his credibility, not this.

 

[Kingpin Rogers]

So what's actually being said in this article? From what I'd read on gaf I was expecting switch to be weaker than xb1 but stronger tham Wii U; closer to xb1 than wii u. It sounds like the article is going along with this, so if it's true, what's the issue?

 

[GifGafIsTheBestGaf]

is this bad? someone please explain in DBZ terms

what people were expecting, the What If scenario were Yamcha is a badass
http://www.crunchyroll.com/anime-ne...f-imagines-a-world-where-yamcha-totally-rules

but instead we got the real timeline

ot, it would have been nice if nintendo went pascal but honestly im not surprised at all. that said, in 3-4 years who could deny that they release a pascal or better Switch in future?

 

[Xdrive05]

The time to be upset about Switch specs is when Nintendo down clocks it to something like 250 gflops in portable mode because of battery life concerns. That's a fine reason to be bitter about them using Maxwell.

That's the most likely way Nintendo will fuck up the specs, if anything. But if they can deliver half a tflop in portable mode then we'll have a great thing going for us and should be happy with it.

 

[Captain Europe]

Because Pascal is much more power efficient and us consumers hope for the best possible product?

I mean, if you care that much about battery life/power consumption then sure, but most people are like ''This is a portable Wii U/360, I'm OUT!''.

ITT, people want a $200 handheld PS4.

With Zelda and Splatoon as pack ins no doubt.

 

[Polygonal_Sprite]

Won't be as powerful as PS4 ! Won't do 4k ! Might be as powerful as Xbox One !

The article makes no sense at all and seems click baity.

All using Maxwell would do is reduce battery life (which is now rumoured to be targeting at least 5 hours) and possible reduce the memory bandwidth (although Nintendo almost always go for a custom solution for this).

A couple of insiders on here have said Switch will run current gen third party games. If the games don't release on it then it won't be because of weak hardware.

 

[Garrett Hawke]

Not worried at all about the Switch's performance. Nintendo learned their lesson with the Wii U. Here's some links to help put the fear to rest if you think this will be Wii U 2.

http://www.ibtimes.com/nintendo-swi...-5-titles-nvidia-gpu-ram-size-details-2439551

https://blogs.nvidia.com/blog/2016/10/20/nintendo-switch/

http://www.ign.com/articles/2016/11/14/nvidia-ceo-calls-nintendo-switch-design-aground-breakinga

http://wccftech.com/nintendo-switch-nvidia-tegra-pascal/

Being maxwell is not going to make this a Wii u 2.

Being pascal is not going to make this a portable PS4.

 

[Schnozberry]

clip

There's also the issue of active cooling and USB 3.1, which weren't present in any standard Tegra X1 design. Whatever is in the Switch, it seems like Nintendo made an earnest attempt to squeeze all the performance out of it that they can, and it is at least partially customized to an unknown extent.

The author of the article seems like a guy who installed a graphics card once and is not convinced he knows everything he needs to write this article. Motley fool was pretty convinced that The Switch would use Parker, and they also invoked sources close to the project.

https://www.fool.com/amp/investing/2016/08/25/report-nvidias-codename-parker-to-power-the-upcomi.aspx

 

[DeuceGamer]

I have a few immediate thoughts after reading through the article:

• Firstly, it's worth noting the difference between Maxwell and Pascal is almost entirely down to the manufacturing process. Maxwell was made on 28nm (and in the case of the TX1, 20nm) whereas Pascal is made on 16nm. The actual architectural difference between the two is minimal, and aside from improved color buffer compression, largely irrelevant for a device like the Switch.
• Despite that, the article never makes any mention of the manufacturing process. I find that extremely strange, as it's obviously the defining difference between the two sets of GPUs.
• In fact, the article gets the difference between the two completely the wrong way around, saying "Nintendo’s box is relatively small, and so it has to fit into the heat profile of a portable device, rather than a set-top box. That’s another reason that explains the older Maxwell technology, as opposed to the Pascal’s state-of-the-art tech." Pascal is literally a more power efficient version of Maxwell, so the incentive would be the other way around.
• The author says "we expect the Nintendo Switch to be more than 1 teraflop in performance", which is notably higher than even those of us who were expecting Pascal were considering (I literally posted earlier today with a 500-750 Gflop estimate). If this is a Maxwell chip, then that would mean at least 4 SMs (512 "CUDA cores") at 1GHz, as they're not going to be able to push much past that on 28/20nm. This is a much larger GPU than most people would have been expecting.

I see a few different scenarios here:

1. The Switch SoC uses Maxwell at 20nm, and simply has a much larger GPU than anticipated to account for the performance.
2. Nintendo looked at the feature-set planned for Pascal when design started, realised that the new features were largely irrelevant, and decided that they would save time and just use a straight-forward die shrink of Maxwell to 16nm instead. That would technically be a Maxwell GPU, but would be almost completely indistinguishable from Pascal in terms of performance.
3. The sources are wrong about Maxwell, the 1 Tflop performance, or both.

Basically, if you're to take the article as being accurate, then the only worthwhile takeaway is this quote:



A Maxwell Tflop is identical to a Pascal Tflop, and it's largely irrelevant to us whether they achieved that by using a larger Maxwell GPU on 20nm/28nm at a lower clock or a smaller Pascal GPU on 16nm at a higher clock.

Should be in OP IMO

 

[LelouchZero]

What the hell is going on here?

Do you guys even know the difference between Maxwell and Pascal? It doesn't seem like many of you do...

Maxwell far surpasses the Xbone and the PS4 and Pascal is only more efficient and has lower power consumption which is why Pascal performs better not because it has more graphical power.

You guys need to know your stuff if you're going to be dealing with technicalities.

That's dependent on the number of cores and the clock-speed of the GPU.

The GTX 750 Ti is fairly close to the Xbox One's GPU in power and is roughly 45.9% more powerful than the speculated specs for Switch's GPU, but this is without taking the bandwidth constraints into consideration.

However if this machine is rendering at 720p using FP16 in handheld mode the bandwidth constraints will be more tolerable in comparison to 1080p or 720p with FP32, unfortunately it will still be limited by bandwidth when rendering higher resolutions with 25.6GB/s of memory bandwidth, hopefully the bandwidth will see a notable increase to cope with higher resolutions.

 

[TheGreatMightyPoo]

I trust that Thraktor guy because his posts are immense and over my head.

 

[Smiles and Cries]

I have a few immediate thoughts after reading through the article:

• Firstly, it's worth noting the difference between Maxwell and Pascal is almost entirely down to the manufacturing process. Maxwell was made on 28nm (and in the case of the TX1, 20nm) whereas Pascal is made on 16nm. The actual architectural difference between the two is minimal, and aside from improved color buffer compression, largely irrelevant for a device like the Switch.
• Despite that, the article never makes any mention of the manufacturing process. I find that extremely strange, as it's obviously the defining difference between the two sets of GPUs.
• In fact, the article gets the difference between the two completely the wrong way around, saying "Nintendo’s box is relatively small, and so it has to fit into the heat profile of a portable device, rather than a set-top box. That’s another reason that explains the older Maxwell technology, as opposed to the Pascal’s state-of-the-art tech." Pascal is literally a more power efficient version of Maxwell, so the incentive would be the other way around.
• The author says "we expect the Nintendo Switch to be more than 1 teraflop in performance", which is notably higher than even those of us who were expecting Pascal were considering (I literally posted earlier today with a 500-750 Gflop estimate). If this is a Maxwell chip, then that would mean at least 4 SMs (512 "CUDA cores") at 1GHz, as they're not going to be able to push much past that on 28/20nm. This is a much larger GPU than most people would have been expecting.

I see a few different scenarios here:

1. The Switch SoC uses Maxwell at 20nm, and simply has a much larger GPU than anticipated to account for the performance.
2. Nintendo looked at the feature-set planned for Pascal when design started, realised that the new features were largely irrelevant, and decided that they would save time and just use a straight-forward die shrink of Maxwell to 16nm instead. That would technically be a Maxwell GPU, but would be almost completely indistinguishable from Pascal in terms of performance.
3. The sources are wrong about Maxwell, the 1 Tflop performance, or both.

Basically, if you're to take the article as being accurate, then the only worthwhile takeaway is this quote:



A Maxwell Tflop is identical to a Pascal Tflop, and it's largely irrelevant to us whether they achieved that by using a larger Maxwell GPU on 20nm/28nm at a lower clock or a smaller Pascal GPU on 16nm at a higher clock.

great post thanks for clearing this mess up

 

[MuchoMalo]

I have a few immediate thoughts after reading through the article:

• Firstly, it's worth noting the difference between Maxwell and Pascal is almost entirely down to the manufacturing process. Maxwell was made on 28nm (and in the case of the TX1, 20nm) whereas Pascal is made on 16nm. The actual architectural difference between the two is minimal, and aside from improved color buffer compression, largely irrelevant for a device like the Switch.
• Despite that, the article never makes any mention of the manufacturing process. I find that extremely strange, as it's obviously the defining difference between the two sets of GPUs.
• In fact, the article gets the difference between the two completely the wrong way around, saying "Nintendo’s box is relatively small, and so it has to fit into the heat profile of a portable device, rather than a set-top box. That’s another reason that explains the older Maxwell technology, as opposed to the Pascal’s state-of-the-art tech." Pascal is literally a more power efficient version of Maxwell, so the incentive would be the other way around.
• The author says "we expect the Nintendo Switch to be more than 1 teraflop in performance", which is notably higher than even those of us who were expecting Pascal were considering (I literally posted earlier today with a 500-750 Gflop estimate). If this is a Maxwell chip, then that would mean at least 4 SMs (512 "CUDA cores") at 1GHz, as they're not going to be able to push much past that on 28/20nm. This is a much larger GPU than most people would have been expecting.

I see a few different scenarios here:

1. The Switch SoC uses Maxwell at 20nm, and simply has a much larger GPU than anticipated to account for the performance.
2. Nintendo looked at the feature-set planned for Pascal when design started, realised that the new features were largely irrelevant, and decided that they would save time and just use a straight-forward die shrink of Maxwell to 16nm instead. That would technically be a Maxwell GPU, but would be almost completely indistinguishable from Pascal in terms of performance.
3. The sources are wrong about Maxwell, the 1 Tflop performance, or both.

Basically, if you're to take the article as being accurate, then the only worthwhile takeaway is this quote:



A Maxwell Tflop is identical to a Pascal Tflop, and it's largely irrelevant to us whether they achieved that by using a larger Maxwell GPU on 20nm/28nm at a lower clock or a smaller Pascal GPU on 16nm at a higher clock.

You know, another funny thing is that they're expecting something almost on-par with Xbone yet at the same time say that it's too weak for Xbone games.

What I can say for certain is that they only heard about the architecture and nothing else, and everything else is just uninformed guessing.

Also, I want to remind you that 20nm and 16nmFF are identical in terms of transistor density.

 

[J@hranimo]

Just preparing you for disappointment.

No, that's literally dumb. It's obviously not going to be such a huge bump in graphical fidelity, but just enough to handle 3rd party ports.

My expectations have been set for quite some time now.

 

[Nanashrew]

I have a few immediate thoughts after reading through the article:

• Firstly, it's worth noting the difference between Maxwell and Pascal is almost entirely down to the manufacturing process. Maxwell was made on 28nm (and in the case of the TX1, 20nm) whereas Pascal is made on 16nm. The actual architectural difference between the two is minimal, and aside from improved color buffer compression, largely irrelevant for a device like the Switch.
• Despite that, the article never makes any mention of the manufacturing process. I find that extremely strange, as it's obviously the defining difference between the two sets of GPUs.
• In fact, the article gets the difference between the two completely the wrong way around, saying "Nintendo’s box is relatively small, and so it has to fit into the heat profile of a portable device, rather than a set-top box. That’s another reason that explains the older Maxwell technology, as opposed to the Pascal’s state-of-the-art tech." Pascal is literally a more power efficient version of Maxwell, so the incentive would be the other way around.
• The author says "we expect the Nintendo Switch to be more than 1 teraflop in performance", which is notably higher than even those of us who were expecting Pascal were considering (I literally posted earlier today with a 500-750 Gflop estimate). If this is a Maxwell chip, then that would mean at least 4 SMs (512 "CUDA cores") at 1GHz, as they're not going to be able to push much past that on 28/20nm. This is a much larger GPU than most people would have been expecting.

I see a few different scenarios here:

1. The Switch SoC uses Maxwell at 20nm, and simply has a much larger GPU than anticipated to account for the performance.
2. Nintendo looked at the feature-set planned for Pascal when design started, realised that the new features were largely irrelevant, and decided that they would save time and just use a straight-forward die shrink of Maxwell to 16nm instead. That would technically be a Maxwell GPU, but would be almost completely indistinguishable from Pascal in terms of performance.
3. The sources are wrong about Maxwell, the 1 Tflop performance, or both.

Basically, if you're to take the article as being accurate, then the only worthwhile takeaway is this quote:



A Maxwell Tflop is identical to a Pascal Tflop, and it's largely irrelevant to us whether they achieved that by using a larger Maxwell GPU on 20nm/28nm at a lower clock or a smaller Pascal GPU on 16nm at a higher clock.

You always make the most insightful technical posts. You're the best, Thraktor!

Also I agree with everyone else, should be in the OP.

 

[Vic]

You know, another funny thing is that they're expecting something almost on-par with Xbone yet at the same time say that it's too weak for Xbone games.

WHat I can say for certain is that they only heard about the architecture and nothing else, and everything else is just uninformed guessing.

This and at this point, we don't even know if they're referencing the dev kit or the final hardware.

 

[Skittzo0413]

I have a few immediate thoughts after reading through the article:

• Firstly, it's worth noting the difference between Maxwell and Pascal is almost entirely down to the manufacturing process. Maxwell was made on 28nm (and in the case of the TX1, 20nm) whereas Pascal is made on 16nm. The actual architectural difference between the two is minimal, and aside from improved color buffer compression, largely irrelevant for a device like the Switch.
• Despite that, the article never makes any mention of the manufacturing process. I find that extremely strange, as it's obviously the defining difference between the two sets of GPUs.
• In fact, the article gets the difference between the two completely the wrong way around, saying "Nintendo’s box is relatively small, and so it has to fit into the heat profile of a portable device, rather than a set-top box. That’s another reason that explains the older Maxwell technology, as opposed to the Pascal’s state-of-the-art tech." Pascal is literally a more power efficient version of Maxwell, so the incentive would be the other way around.
• The author says "we expect the Nintendo Switch to be more than 1 teraflop in performance", which is notably higher than even those of us who were expecting Pascal were considering (I literally posted earlier today with a 500-750 Gflop estimate). If this is a Maxwell chip, then that would mean at least 4 SMs (512 "CUDA cores") at 1GHz, as they're not going to be able to push much past that on 28/20nm. This is a much larger GPU than most people would have been expecting.

I see a few different scenarios here:

1. The Switch SoC uses Maxwell at 20nm, and simply has a much larger GPU than anticipated to account for the performance.
2. Nintendo looked at the feature-set planned for Pascal when design started, realised that the new features were largely irrelevant, and decided that they would save time and just use a straight-forward die shrink of Maxwell to 16nm instead. That would technically be a Maxwell GPU, but would be almost completely indistinguishable from Pascal in terms of performance.
3. The sources are wrong about Maxwell, the 1 Tflop performance, or both.

Basically, if you're to take the article as being accurate, then the only worthwhile takeaway is this quote:



A Maxwell Tflop is identical to a Pascal Tflop, and it's largely irrelevant to us whether they achieved that by using a larger Maxwell GPU on 20nm/28nm at a lower clock or a smaller Pascal GPU on 16nm at a higher clock.

Yeah I agree with all of this, though I'm assuming the 1TFlop quote from the article is referencing FP16 rather than FP32 and the comparisons from there are just the author misunderstanding those differences.

It'll be nice to hear some clarification if/when NateDrake hears back from his sources.

 

[PinkyMonkey]

Nintendo tends to use oldet generation hardware, this shouldn't be a surprise to anyone by now.

 

[Speely]

I have a few immediate thoughts after reading through the article:

*reason and logic*

Thanks, Thraktor. You are always an appreciated voice of reason (and knowledge.)

Guys, I don't think we should be freaking out about Maxwell vs Pascal. If it truly is a custom chip that both Nintendo and Nvidia are pinning their hopes on, it will be a considerable re-work either way, and either way we can reasonably expect to see what we were always expecting to see: Better than Wii U and around/slightly below Xbone performance.

 

[Schnozberry]

The guy who designed the Switch did that by following the base guideline of the guy in charge of the hardware at Nintendo. When the Switch was originally conceived, that person was still Takeda. Reading about Maxwell and A57 means that the design was locked in very early for some reason.

Look up Sudha Sudharsanan. He's been at Nintendo for years, was a former Nvidia Engineer, and now heads their Tech Development. Takeda didn't drive the Switch concept. It was Nintendo pivoting towards their rising stars.

 

[Xellos]

Even with Pascal, Switch was never going to be in XB1/PS4 territory. Pascal is not so efficient that it can hit that kind of performance inside of a 6" tablet.

With Maxwell, 400-500 GFLOP GPU is the most likely target for Switch. Pascal would have been better but not by a lot. In either case Switch is a significant improvement over Wii U but less powerful than Xbox One, so whether it's Maxwell or Pascal is not a huge deal to me.

 

[Polygonal_Sprite]

Should be in OP IMO

Definitely, great post as usual from Thraktor.

 

[CronoShot]

Using cheaper, older parts is a very Nintendo thing to do. I don't know why anyone expected differently.

 

[Uno Venova]

Since my only expectations are a traditional, cheap console that plays Nintendo games, something I haven't gotten in like a decade, these news about power doesn't phase me much. Just need the console to run games well.

 

[kIdMuScLe]

It doesn't even make sense because the switch console is too thin to house a maxwell chipset. Maxwell is too big

 

[BitStyle]

I have a few immediate thoughts after reading through the article:

• Firstly, it's worth noting the difference between Maxwell and Pascal is almost entirely down to the manufacturing process. Maxwell was made on 28nm (and in the case of the TX1, 20nm) whereas Pascal is made on 16nm. The actual architectural difference between the two is minimal, and aside from improved color buffer compression, largely irrelevant for a device like the Switch.
• Despite that, the article never makes any mention of the manufacturing process. I find that extremely strange, as it's obviously the defining difference between the two sets of GPUs.
• In fact, the article gets the difference between the two completely the wrong way around, saying "Nintendo’s box is relatively small, and so it has to fit into the heat profile of a portable device, rather than a set-top box. That’s another reason that explains the older Maxwell technology, as opposed to the Pascal’s state-of-the-art tech." Pascal is literally a more power efficient version of Maxwell, so the incentive would be the other way around.
• The author says "we expect the Nintendo Switch to be more than 1 teraflop in performance", which is notably higher than even those of us who were expecting Pascal were considering (I literally posted earlier today with a 500-750 Gflop estimate). If this is a Maxwell chip, then that would mean at least 4 SMs (512 "CUDA cores") at 1GHz, as they're not going to be able to push much past that on 28/20nm. This is a much larger GPU than most people would have been expecting.

I see a few different scenarios here:

1. The Switch SoC uses Maxwell at 20nm, and simply has a much larger GPU than anticipated to account for the performance.
2. Nintendo looked at the feature-set planned for Pascal when design started, realised that the new features were largely irrelevant, and decided that they would save time and just use a straight-forward die shrink of Maxwell to 16nm instead. That would technically be a Maxwell GPU, but would be almost completely indistinguishable from Pascal in terms of performance.
3. The sources are wrong about Maxwell, the 1 Tflop performance, or both.

Basically, if you're to take the article as being accurate, then the only worthwhile takeaway is this quote:



A Maxwell Tflop is identical to a Pascal Tflop, and it's largely irrelevant to us whether they achieved that by using a larger Maxwell GPU on 20nm/28nm at a lower clock or a smaller Pascal GPU on 16nm at a higher clock.

Thanks for this post. I actually didn't even pick up on the article mentioning 1 Tflop. All I can hope for is Pascal for better battery, as that would probably be the most noticeable difference, correct?

 

[ChrisMahoney]

Even with Pascal, Switch was never going to be in XB1/PS4 territory. Pascal is not so efficient that it can hit that kind of performance inside of a 6" tablet.

With Maxwell, 400-500 GFLOP GPU is the most likely target for Switch. Pascal would have been better but not by a lot. In either case Switch is a significant improvement over Wii U but less powerful than Xbox One, so whether it's Maxwell or Pascal is not a huge deal to me.

Pascal could have reached close to XB1 territory when docked... Maxwell not so much

 

[Shikamaru Ninja]

Look up Sudha Sudharsanan. He's been at Nintendo for years, was a former Nvidia Engineer, and now heads their Tech Development. Takeda didn't drive the Switch concept. It was Nintendo pivoting towards their rising stars.

NTD operates in conjunction with Nintendo PTD. So while Sudharsan and the Redmond team research and develop a lot of technology, the Japan arm (Platform Technology & Development Division) is still the lead developer.

Ko Shiota is the head of PTD, and was the right hand man of Takeda. But even before all that, the big decisions of the hardware were probably still made at the end of the Iwata/Takeda/Miyamoto reign back in early 2015?

 

[TheGreatMightyPoo]

It doesn't even make sense because the switch console is too thin to house a maxwell chipset. Maxwell is too big

Ridley Maxwell.