Nintendo announce: Nintendo NX

[Prophecy2k]

Actually, engineering can be impressive when it solves a problem, even if the problem is self-inflicted (save when the problem is self-inflected by poor engineering). The management asks for the engineers to solve something, and solving that effectively is where the engineers can be marvels or not. "We want a five wheeled car with a wheel in the middle of the four corners, with three wheel steering and four wheel drive." If the engineers can pull that off well, it's a good job, even though the product is possibly moronic.

Not that I think Wii U is well engineered, but I think it's right to separate the engineering solutions from the product designs where the designs aren't a product of the engineers.

I dunno, I guess in my industry the design of every product is is a product of the engineer. You absolutely cannot devorce the two. Maybe game consoles are different, in which case i stand corrected. Eitherway I would argue that perhaps Nintendo should perhaps let its engineers design it's next console in which case. As the existing folks in charge don't seem to be doing a particularly swell job.

It certainly seemed to work for Sony with the PS4. It's been a runnaway success, and Mark Cerny is now hailed as an effective Geek God (not ot be confused with Greek God; although he could probably pass for one of those too depending on who you ask).

 

[Deleted member 13524]

the choice of going with an MCM however does mean that BC is as simple as adding Espresso as a co processor much the same way they do with their handhelds.

I'm not sure if having the CPU in a different chip makes it easier to integrate later as a coprocessor.
No Nintendo handheld SoC so far has used MCM and that didn't stop them to integrate the main cpu of one console into a coprocessor of the following console.
Or are you suggesting that the NX platforms will bring an external Espresso in all its consoles, just for the sake of BC?

45nm process for the CPU was chosen because Wii's CPU was shrunk to 45nm at the time, and 40nm for the GPU was chosen because Resna (sp?) had to produce the EDRAM inside the GPU die, and this was their smallest process node.

Renesas wasn't the only one capable of producing chips with eDRAM. IBM did it too.
They weren't even the only ones capable of producing chips with eDRAM + AMD GPU combo (and an IBM CPU to top it off). IBM was already doing that too, for the X360.
Somewhere along the R&D cycle they decided it was a good idea to bring the GPU and CPU from different foundries, and I'd bet even that was probably a "let's save 3 pennies here" attitude, just like the resistive vs. capacitive screen.

 

[Syferz]

I'm not sure if having the CPU in a different chip makes it easier to integrate later as a coprocessor.
No Nintendo handheld SoC so far has used MCM and that didn't stop them to integrate the main cpu of one console into a coprocessor of the following console.
Or are you suggesting that the NX platforms will bring an external Espresso in all its consoles, just for the sake of BC?


Renesas wasn't the only one capable of producing chips with eDRAM. IBM did it too.
They weren't even the only ones capable of producing chips with eDRAM + AMD GPU combo (and an IBM CPU to top it off). IBM was already doing that too, for the X360.
Somewhere along the R&D cycle they decided it was a good idea to bring the GPU and CPU from different foundries, and I'd bet even that was probably a "let's save 3 pennies here" attitude, just like the resistive vs. capacitive screen.

Espresso is rumored to be extremely cheap as the original license Nintendo signed didn't have a public expiration date and might still be under effect. (it was worth 1+ Billion USD back in 2000) MCM simplifies the design a bit as they could also place n3DS soc on the MCM as well if need be.

I didn't mean Renesas was the only one who could produce it, but they were the ones Nintendo went with, if it was to save money, I'm not sure how it would be cheaper than a single IBM die, which would reduce power, silicon space and add performance thanks to on die transfer speeds. Not sure what this was but maybe AMD didn't want IBM to have knowledge of Radeon 700? Microsoft had a much older design but who knows what the reasoning was other then NEC was part of the team who created previous consoles and thus Nintendo thought it was a good idea to stick with them, the QA was definitely more difficult coming from 3 different companies rather than 2 though.

 

[Deleted member 13524]

AMD didn't want IBM to have knowledge of the DX10, VLIW5 R700 when they had already gone through that and entire line-ups of DX11 VLIW5, DX11 VLIW4 and had the completely different GCN in the making?
I find that really hard to believe... especially when in 2005 they "gave access" to a GPU that ended up being even more evolved than the desktop parts that released the following year.

And what did you mean that was worth >1B USD? The PowerPC750? I didn't get it.

 

[Syferz]

AMD didn't want IBM to have knowledge of the DX10, VLIW5 R700 when they had already gone through that and entire line-ups of DX11 VLIW5, DX11 VLIW4 and had the completely different GCN in the making?
I find that really hard to believe... especially when in 2005 they "gave access" to a GPU that ended up being even more evolved than the desktop parts that released the following year.

And what did you mean that was worth >1B USD? The PowerPC750? I didn't get it.

I was speculating on AMD not wanting to give a competitor their IP, and they were still using VLIW5 up though 2011 actually, when in 2005 did IBM gain access to Xenos? surely you mean when they started doing an APU? before that it was manufactured by a different company. 360: "The GPU package contains two separate silicon dies, each built on a 90 nm process with a clock speed of 500 MHz; the GPU proper, manufactured by TSMC and a 10 MB eDRAM daughter-die, manufactured by NEC." IBM didn't combined the chips until 2010.

As for the deal between IBM and Nintendo, that would be this one: https://www-03.ibm.com/press/us/en/pressrelease/2181.wss the rumor is that the deal was extended well into the Wii's life and that was part of the reason IBM designed Espresso, it is a partially paid for chip, so Nintendo gets it at a huge discount, at least that is the rumor.

 

[AlNom]

I didn't mean Renesas was the only one who could produce it, but they were the ones Nintendo went with, if it was to save money, I'm not sure how it would be cheaper than a single IBM die, which would reduce power, silicon space and add performance thanks to on die transfer speeds.

It'd have been a larger SOI chip on a somewhat costlier eDRAM process (aside from SOI costing more by default).

The 32MB "eDRAM" is quite likely to be 1T-SRAM, mind you, so wafer costs would have been lower (and local in Japan). I wouldn't be surprised if they already had Hollywood on the latest Renesas node already, with fully integrated 24MB 1T-SRAM, audio & ARM processor. All they'd have to do is bolt on the extra SRAM, add more I/O for bandwidth, port the Radeon IP to Renesas fab library.

 

[Goodtwin]

It'd have been a larger SOI chip on a somewhat costlier eDRAM process (aside from SOI costing more by default).

The 32MB "eDRAM" is quite likely to be 1T-SRAM, mind you, so wafer costs would have been lower (and local in Japan). I wouldn't be surprised if they already had Hollywood on the latest Renesas node already, with fully integrated 24MB 1T-SRAM, audio & ARM processor. All they'd have to do is bolt on the extra SRAM, add more I/O for bandwidth, port the Radeon IP to Renesas fab library.

The 2MB Z-buffer/frame buffer is extremely helpful since we already know from our experimentation with HyperZ and deferred rendering architectures that Z-buffer accesses are very memory bandwidth intensive. This on-die Z-buffer completely removes all of those accesses from hogging the limited amount of main memory bandwidth the Flipper GPU is granted. In terms of specifics, there are 4 1T-SRAM devices that make up this 2MB. There is a 96-bit wide interface to each one of these devices offering a total of 7.8GB/s

Your assumption of 1T-Sram is intriguing, simply because speculation around the edram on Wii U was always clouded in mystery, but perhaps if you simply look a the Gamecube/Wii, its safe to assume that the 32MB is broken down into many units, each with its own bus, and would very likely have similar bandwidth per MB of memory compared to that in the Gamecube/Wii.

 

[AlNom]

Your assumption of 1T-Sram is intriguing, simply because speculation around the edram on Wii U was always clouded in mystery, but perhaps if you simply look a the Gamecube/Wii, its safe to assume that the 32MB is broken down into many units, each with its own bus, and would very likely have similar bandwidth per MB of memory compared to that in the Gamecube/Wii.

Right, well the 24MB pool had much lower I/O bandwidth compared to the texture & framebuffer pools, so they'd have to change that for the 32MB pool if it's going to feed the RBEs/CUs... and thus all the wonderful speculation from @Fourth Storm et al.

 

[Shoujoboy]

Even though it gotba littke off topic I like the discussion my question spawned.

 

[Goodtwin]

Even though it gotba littke off topic I like the discussion my question spawned.

Well it went off track, but its still all relevant to speculation on what Nintendo will likely do in the future. They are odd in some of their choices. Its like building a race car, but then deciding that fuel efficiency is a priority, and we are going to limit of displacement to 2 liters. This is fine in racing when the rules place these restrictions upon everyone, but with Nintendo, they are the only ones creating hardware under these restrictions. Nintendo has its philosophy on creating hardware, and it seems pretty obvious to me that we shouldn't be expecting a 150 watt console the size of a VCR sitting under the tv. Size and power efficiency will still be a thing for Nintendo. That's why I think the NX starts as a portable. I believe they may eventually bring a more powerful version in console form, but the NX portable could easily share games with the Wii U early on, starting the trend of buying Nintendo hardware and games means never having to rebuy them again for new hardware. I think that in itself is a new concept in the console world, and Nintendo could be first to do it.

 

[Rikimaru]

I believe they may eventually bring a more powerful version in console form, but the NX portable could easily share games with the Wii U early on, starting the trend of buying Nintendo hardware and games means never having to rebuy them again for new hardware. I think that in itself is a new concept in the console world, and Nintendo could be first to do it.

Yeah, sure. Like Virtual Console games...
And new portable won't be as powerful as Wii U.

 

[Syferz]

Yeah, sure. Like Virtual Console games...
And new portable won't be as powerful as Wii U.

AMD doesn't make anything portable that is weaker than Wii U, closest is "mullins" but unless they go with a high resolution screen, the 128GFLOPs would be enough to compete just fine with Wii U on a 480p-540p screen, and that is a 28nm chip with puma+ cores and below 3 watt power draw at the 2ghz cpu clock.

 

[AlNom]

AMD doesn't make anything portable that is weaker than Wii U, closest is "mullins" but unless they go with a high resolution screen, the 128GFLOPs would be enough to compete just fine with Wii U on a 480p-540p screen, and that is a 28nm chip with puma+ cores and below 3 watt power draw at the 2ghz cpu clock.

The bandwidth would be anaemic, no? Even with half the ROPs, the WiiU still has a dedicated GPU-bandwidth pool along with the main DDR3 pool (achieve same as DDR4 on half the bus width).

2GHz for Mullins would be turbo. Pretty sure the 3W SDP is for the low power/clock scenario.

 

[Goodtwin]

Yeah, sure. Like Virtual Console games...
And new portable won't be as powerful as Wii U.

So you don't think a portable could rival the Wii U in late 2016? I think AMD's recent mobile APU's paint a different picture. Your already seeing GPU performance that rivals the Wii U, and seeing as how I see little value in a portable having a higher than 720p screen, the resolution target can still be 720p. Memory setup could still incorporate edram, or a small pool of fast sram to handle GPU operations and reduce the memory bandwidth requirements to the main memory. Seeing as how there is such a shift towards mobile processor development, tablets and phones are replacing PC's and laptops for many people, so it only makes sense for us to see a surge in R&D for these low power consuming chips. Im not saying the first iteration of NX will blow Wii U away, but that it can be very comparable in performance. I also think Nintendo will unify accounts, and that will begin with the transition from Wii U to NX. I have a feeling what you bought on Wii U, will be playable on NX.

 

[Rikimaru]

So you don't think a portable could rival the Wii U in late 2016?

Could you see a portable in 2011 rivaling a portable from 2004? And have worse resolution?
Next portable will be 480p and have worse than Vita specs most likely (except RAM).

I have a feeling what you bought on Wii U, will be playable on NX.

No way this could happen (assuming NX is a portable).
And Mullins is too power hungry for a nintendo portable.

 

[east of eastside]

AMD Styx is probably Nintendo's NX DS portable's APU confirming by extension that NX console will also share the same K12 CPU and Next-gen GCN cores.

http://www.androidheadlines.com/page/2

At this juncture, it cannot be said for certain what’s caused AMD to decide to sell into the mobile processor market. Perhaps the Amur and Styx processors have something new (and undisclosed) to excite and interest the market?

Answer:

http://metro.co.uk/2014/07/16/amd-hints-at-deal-for-3ds-successor-4799938/

AMD hints at deal for 3DS successor

‘Another area of interest for semi-custom is handheld gaming, believe it or not,’ said Moshkelani. ‘Everyone thinks it is dead, but the 3DS is still selling.’

AMD is one of the few electronics companies that hasn’t produced their own Android microconsole recently and nothing Moshkelani says seems to suggest that’s what he’s talking about.

‘[It] has to be at least $100 million annual revenue for us to go for it,’ he said of AMD projects in general. ‘That’s a minimum, not a target’.

Translation: AMD has come up with a new mobile processor for a minimum $100 million revenue handheld gaming deal... it can only be NX DS.

 

[BRiT]

You really think Nintendo will be buying the latest and greatest ultra bleeding edge for mobile processors from a fabrication standpoint?

 

[Shifty Geezer]

We can but hope!

 

[AlNom]

Start off with 20nm, then switch to finfets later and overclock for BigNX mobile. Meanwhile, the phone companies are sucking up all the advanced node wafer production for $$$.

 

[east of eastside]

Another journalist asking the same question:

http://www.extremetech.com/gaming/204567-amds-desktop-mobile-roadmaps-for-2016-may-have-just-leaked

If these slides are true, they raise as many questions as they answer, particularly concerning why AMD would launch an ultra-low power ARM CPU. True, it could theoretically make an ARM play in Android, but AMD has never shown much interest or ability to comprehensively attack the Android market.

NX DS!

You really think Nintendo will be buying the latest and greatest ultra bleeding edge for mobile processors from a fabrication standpoint?

You can buy 14nm products today. Bleeding edge? 10nm? 7nm? Samsung will have cranked millions of Galaxy S6 processors (and later this year iPad Air 3 and iPhone 7) before NX ever goes into manufacturing. AMD is using the same 14nm fab whether it is Glofo or Samsung manufacturing it. The process will be mature by the time NX launches and the economics will support it (cost vs benefit) over 28nm.

Do you know what other 2016 mystery mobile products AMD is making when they have already stated 2016 gaming design win using semi-custom parts introduced in 2016?

PS Vita 2?!