Nintendo GOing Forward.

[Syferz]

Is there a breakdown somewhere of the new features/changes in the New 3DS?


I agree with you on the bus width. 256-bit isn't going to happen and 128 might be a stretch as well, going to DDR4 on a 64 or even 96-bit bus they could increase the bandwidth 2-3X. However, I'm not sure of doubling the size of on chip memory buffer to 64MB. It takes up a large amount of die area, they'd have to go with a smaller node than 28nm, but I don't expect them to.

There is the possibility that by 2017 or so, 28nm will be so mature that eDRAM will be much more viable on the node and maybe then they can increase the memory to maybe 48MB.

I think that would be more up to what AMD is offering at the time, it's not like Nintendo is designing the chips or manufacturing them, AMD in 2017 better be beyond 28nm, because everyone else will be. I am not sure why everyone falls back to old nodes with Nintendo, every console they have launched has used the newest node available since at least the gamecube, the only exception is the Wii U, which is because Renesas Electronics didn't produce a 28nm node. We could end all speculation being cynical and saying Nintendo will use Kabini on 28nm with 386ALUs and 1.6GHz quad core. That sounds like it would be more easily accepted than Nintendo is a customer that will pick from AMD's hardware, I mean it's clear that they are waiting on tech since Iwata himself said it would take about 2 years last April, this is most likely for a new node to mature that AMD is planning on using. Heck they have to use a smaller node for their handheld if they go with AMD's APUs anyways because currently ~3watts is too much of a power draw.

And yes I'm a Nintendo fan but I'm also a PC gamer and a hardware enthusiast, I want tech to move forward and I don't see Nintendo making the exact same mistakes with their next hardware cycle because all of their goals are different this time except for a lower power consumption which is relative to whatever number they have in mind.

 

[Fourth Storm]

I agree with Syferz in that Nintendo has not historically been that far behind in manufacturing node used. Wii U's GPU was an exception. What is different this time, however, is that the added complexity of finFETs (and 20nm) will mean that Nintendo do not get a better value by going to a smaller node. In fact, I think the price might increase. There are also huge customers like Apple who might tie up alot of manufacturing lines this year. I think that by late 2016, however, 16nm/14nm will be available to other customers. Nvidia seems to be indicating that is when they will transition and AMD have also indicated that 20nm is more of a stop gap for this year.

Unless I'm missing something, AMD are beginning to integrate ARM cores on their APUs with Project Skybridge, which is designed for the 20nm node. So I would say that 20nm is the baseline, and since the power savings that node offers are marginal at best, Nintendo just might choose to wait, even if it raises their prices a bit. Honestly, hardware price isn't their problem. The market can handle a $199 handheld and $299 home console. Nintendo just haven't offered enough value with their current selection.

 

[liolio]

I hope that the handheld design command at the hone console hardware, the difference in resolution (and so overall hardware requirement) are getting closer and closer. If Nintendo can't get third party support or refuse to break the bank to get it all it is important that publishers see the handheld+home console as similar platform.
As I "bet" in the "new next gen thread prediction" I hope Nintendo goes with the same SOC, the handheld version being a salvaged, low clocked, version of the SOC.
My take/bet is that 28nm should be good enough for the perfs they might aim at.

 

[Syferz]

I hope that the handheld design command at the hone console hardware, the difference in resolution (and so overall hardware requirement) are getting closer and closer. If Nintendo can't get third party support or refuse to break the bank to get it all it is important that publishers see the handheld+home console as similar platform.
As I "bet" in the "new next gen thread prediction" I hope Nintendo goes with the same SOC, the handheld version being a salvaged, low clocked, version of the SOC.
My take/bet is that 28nm should be good enough for the perfs they might aim at.

Why would they need to salvage a 28nm chip? The process is very mature already, it's been used in mainstream gpus since 2011. It is far more likely AMD will have moved on to at least 20nm in time for these devices, since Nintendo themselves said that the technology isn't ready yet and will take another ~16 months from now

 

[liolio]

Why would they need to salvage a 28nm chip? The process is very mature already, it's been used in mainstream gpus since 2011. It is far more likely AMD will have moved on to at least 20nm in time for these devices, since Nintendo themselves said that the technology isn't ready yet and will take another ~16 months from now

My (possibly crazy or misslead) answer is for the sake of designing and produce only one SOC.
In the aforementioned thread (I should have linked my post from scratch) I was not advocating for a massive waste of silicon and good yields, just a "cheap" way to make sure both version of the same SOC deliver consistent performances at the resolution they target.

 

[Syferz]

My (possibly crazy or misslead) answer is for the sake of designing and produce only one SOC.
In the aforementioned thread (I should have linked my post from scratch) I was not advocating for a massive waste of silicon and good yields, just a "cheap" way to make sure both version of the same SOC deliver consistent performances at the resolution they target.

AMD uses a modular design with it's GCN, each compute unit has 4 SIMDs which compose of 16 ALUs and a scalar unit(iirc) Each CU* (compute unit) can do 4 instructions at once (1 per SIMD) AMD has never made a GPU that consists of a single CU, but here is anandtech's early breakdown of GCN scaling:

http://www.anandtech.com/show/4455/amds-graphics-core-next-preview-amd-architects-for-compute
"While the compute unit is the fundamental unit of computation, it is not a GPU on its own. As with SIMDs in Cayman it’s a configurable building block for making a larger GPU, with a GPU implementing a suitable number of CUs in multiples of 4. Like past GPUs this will be the primary way to scale the GPU to the desired die size, but of course this isn’t the only element of the design that scales."

Now AMD has made CUs in groups of 2, this is much more reasonable than a single SIMD, as the memory controller is built to share across many CUs at once. This is why I think 128 ALUs is the minimum with 256 ALUs being likely for the handheld. For the Console, they can scale by sets of 4, it can be broken up differently but I assume they will maintain this number, XB1 uses 12CUs while PS4 uses 18CUs. I think depending on Nintendo's goals, they will go with a higher clocked 8CUs for the console (the minimum I'd assume) with 16CUs being the more ideal scale to meet PS4 specs while being perfectly scale-able to their handheld.

 

[liolio]

I never alluded to AMD hardware, I'm aware of the rumors though it is unclear to me how your post connects with mine.
I hope Nintendo passes on AMD.

 

[Grall]

It is far more likely AMD will have moved on to at least 20nm in time for these devices

20nm is going to be the shortest silicon lithography generation ever; it's probably not even going to be available anymore by the time Nintendo launches its next console(s). It's going to be as dead an option as 28nm will be.

 

[Shifty Geezer]

Given the latest performance news on nVidia's X1, that plus AMD cores looks like a perfect fit for handheld and console, scaling however high Nintendo want to take it in the home box. Although it would still be much lower battery life than Nintendo's handhelds have aimed for, but that's probably a necessity.

 

[liolio]

Given the latest performance news on nVidia's X1, that plus AMD cores looks like a perfect fit for handheld and console, scaling however high Nintendo want to take it in the home box. Although it would still be much lower battery life than Nintendo's handhelds have aimed for, but that's probably a necessity.

I'm not sure Nintendo wants to scale it high at all.
One just has to look at the new 3Ds which got released mostly as the same time as those late Asus memo pad tablet (me572) for mostly the same price to understand that if Nintendo is under pressure to release good hardware they are resisting it great.

 

[Syferz]

I'm not sure Nintendo wants to scale it high at all.
One just has to look at the new 3Ds which got released mostly as the same time as those late Asus memo pad tablet (me572) for mostly the same price to understand that if Nintendo is under pressure to release good hardware they are resisting it great.

The problem with n3DS is, it's not new hardware, just a refresh of old hardware. Much like the DSi was when it came out in 2009, 2 years before 3DS launched. That is just a problem with the old hardware cycles, you can't make turns in hardware until the cycle is over, you have to stay the path until the end. Another benefit to an account platform will be completely changing hardware specs inside an architecture whenever they need to.

Also my previous post was related to yours simply because I was going over the scaling of GCN, as you were worried about performance when using different SoC.

 

[liolio]

The problem with n3DS is, it's not new hardware, just a refresh of old hardware. Much like the DSi was when it came out in 2009, 2 years before 3DS launched. That is just a problem with the old hardware cycles, you can't make turns in hardware until the cycle is over, you have to stay the path until the end. Another benefit to an account platform will be completely changing hardware specs inside an architecture whenever they need to.

The 3DS got release early 2011 as you know, it is soon to be 4 years old Nintendo had no issue breaking compatibility with previous models.
It is not completely new hardware I get what you mean yet that hardware is 4 years old, is significantly outdone by the PS V (not a commercial threat though) not too mention contemporary hardware.
For the ref and those not familiar with the tablet market the Asus Me572 packs: quad core Baytrail + PowerVR G6430 (Moorefield), 2 GB of RAM, 16 GB of storage and an pretty good IPS full HD screen ( the same as the Nexus 2013).
The n3DS is inline with Nintendo practices, that is true. As such it does a good job at showing how things have not changed at NIntendo HQ.

Looking at Erista, now Tegra X1, is a good reference point. Actually I don't expect Nintendo to reach that level of performances with their handheld mostly because they will tune for sustained performances and thermal dissipation. Yet it would already be a gigantic jump forward for NIntendo, pretty much they can deliver the ps360 /WiiU experience on their handheld.

Spoiler

I expect to be let down

Also my previous post was related to yours simply because I was going over the scaling of GCN, as you were worried about performance when using different SoC.

OK, well I speak of a pretty mild difference between the SOC, just enough to accommodate the extra pixels between say 1080p for the TV and something either 720p or slightly below (the handheld screen). I do not speak of scaling say from a PS Vita to a XB1 for example.

I hope they pass on AMD though, they are lagging Nvidia in performance per watts more and more, not too mention others vendors. I'm not impress by their last "mobile" CPU at all.
Overall (in my view obviously) if it is about delivering a potent handheld that serves as a ref point for the home console then AMD is everything but unproven. Their solution as for now within a tablet like TDP are lagging all their serious competitors both CPU and GPU.
I really hope Nintendo passes on their solution as they (imo again) should not aim for performances bracket AMD delivers. AMD solution are still too high power both CPU and GPU, they lack bandwidth efficient of competing mobile designs, etc.
I feel sorry for the red team but AMD have done really few significant advancements to GCN since its release, Nv is a lot more active so is PowerVR.

 

[Syferz]

I hope they pass on AMD though, they are lagging Nvidia in performance per watts more and more, not too mention others vendors. I'm not impress by their last "mobile" CPU at all.
Overall (in my view obviously) if it is about delivering a potent handheld that serves as a ref point for the home console then AMD is everything but unproven. Their solution as for now within a tablet like TDP are lagging all their serious competitors both CPU and GPU.
I really hope Nintendo passes on their solution as they (imo again) should not aim for performances bracket AMD delivers. AMD solution are still too high power both CPU and GPU, they lack bandwidth efficient of competing mobile designs, etc.
I feel sorry for the red team but AMD have done really few significant advancements to GCN since its release, Nv is a lot more active so is PowerVR.

I'm not sure how far behind AMD truly is here... X1 is only 512GFLOPs (single precision) clocked at 1GHz @ 10w on the 20nm process. (the 1tflop is half precision or fp16 "peak") mullins is 128gflops clocked at 500mhz @ 2.8w on the 28nm process. I am unsure how much wattage 20nm process saves, but if power/performance was linear; mullins scaled to 7cu @ 500mhz would be 10w and 448GFLOPs on the 28nm process, I think AMD could come fairly close with an updated architecture on their APU (arm or x86) while moving down to 20nm, Nintendo of course would not use a 10w part in their handheld which will use ~4w for the entire system (I think 3DS is 4.1w)

 

[liolio]

I'm not sure how far behind AMD truly is here... X1 is only 512GFLOPs (single precision) clocked at 1GHz @ 10w on the 20nm process. (the 1tflop is half precision or fp16 "peak") mullins is 128gflops clocked at 500mhz @ 2.8w on the 28nm process. I am unsure how much wattage 20nm process saves, but if power/performance was linear; mullins scaled to 7cu @ 500mhz would be 10w and 448GFLOPs on the 28nm process, I think AMD could come fairly close with an updated architecture on their APU (arm or x86) while moving down to 20nm, Nintendo of course would not use a 10w part in their handheld which will use ~4w for the entire system (I think 3DS is 4.1w)

Well it's never been about FLOPS figures, Nvidia used to offer more performances per "paper FLOPS" than AMD, it got significantly worse since Maxwell.

More important than performances is power consumption. We have hands on experience with Nvidia SOC, they are on the pretty hot side and consume a bit too much yet it works in tablets, throttling happens but it is reasonable.
AMD chips well let just say that if AMD were delivering the PR figures you are quoting market and OEM reaction toward Puma/Mullins would completely different so would be my pov on the matter.

 

[Syferz]

Well it's never been about FLOPS figures, Nvidia used to offer more performances per "paper FLOPS" than AMD, it got significantly worse since Maxwell.

More important than performances is power consumption. We have hands on experience with Nvidia SOC, they are on the pretty hot side and consume a bit too much yet it works in tablets, throttling happens but it is reasonable.
AMD chips well let just say that if AMD were delivering the PR figures you are quoting market and OEM reaction toward Puma/Mullins would completely different so would be my pov on the matter.

Yeah, no design wins is a huge disappointment, but I think this has more to do with Intel pricing AMD out of the tablet market all last year (they were selling their atom chips below anything amd could offer and there was a heavy push for the $99 tablet) Intel has just recently ended their heavy discounts on these chips however so maybe we will see AMD's procs in something, however HP Stream 7 from CES is only $119 and is an atom chip, so maybe we won't.

 

[Butta]

I personally hope that Nintendo will not release a new version of their console until late 2017. I just purchased a Wii U and find it quite impressive. I personally find the graphics to be acceptable and wish Nintendo would do the following to increase sales:

- Release a device to allow remote play everywhere in the house (perhaps this would only work in a wired network with some kind of special access points) or extend the range (allow 2.4 GHz for remote play - may cause slight lag)
- Drop the price by $50 on the black Wii U and ditch the white one
- Allow people to activate bluray, dvd, DLNA playback for a reasonable fee
- Announce Animal Crossing or Pokemon game that truly utilizes the assymetrical gameplay in an innovative way

 

[Deleted member 11852]

Out of curiosity, why do you care if Nintendo sell a white console?

 

[Butta]

Out of curiosity, why do you care if Nintendo sell a white console?

I just find they skimped on the flash memory too much in the white one... 8 GB is too small. Also, having only 1 to choose from is less confusing for a casual

 

[Megadrive1988]

This is what I'm expecting / hoping for Nintendo as they go forward.

2015 - Nintendo introduces their Quality of Life platform, whatever that is, and continues to release their big titles on Wii U (Zelda, Star Fox, etc).

2016 - (Fall) - They release their next generation handheld / tablet / portable platform with graphical capabilities at least in the same ballpark as Xbox 360, PS3, WiiU. It's well beyond the performance of PS VITA but that won't matter since that platform is dead.

2017 - (Fall) Next generation console, using the same silicon building blocks as the handheld, but more and faster CPU cores, GPU shader / stream processors, TMUs, ROPs, 256-bit bus, 3D stacked memory, etc. this console needs to be in the ~100+ TDP range and show a clear difference beyond PS4 as it would be releasing 4 years after PS4 and only perhaps 2 years before PS5.

By Fall 2017, Nintendo's chip partners would be able to at least have 16nm FinFET+ at worst, perhaps 14nm FinFET. The console should not launch in 2016 but rather 2017 for sure, giving Wii U a not-shortened lifecycle of 4 years, but 5.

Keep in mind this next console from Nintendo could be their last if it does badly like Wii U or even just reaches GameCube level of sales. Nintendo needs to go reasonably impressive with hardware, or not go at all.

 

[McHuj]

I think Nintendo needs a new device announcement this year for a new handheld. The 3DS was announced in March of 2010, it will be 5 years.

But here's the thing, it doesn't have to be a 3ds replacement. It could be a additional platform like the DS was to the Gameboy early on. I'm thinking a tablet with buttons, but not a WiiU tablet. More subtle and simpler. I don't know what exactly, but if there's a way to do this,Nintendo could.

I don't think an updated DS is enough, I want the same reaction when I saw the WiiMote for the first time.