Nintendo announce: Nintendo NX

Status
Not open for further replies.
I suspect them comparison between the NX and the X1 are completely out of lace, one system is +100Watts the other not.
No TX1 device to date is +100W and neither is the NX expected to be so, since eurogamer's specs point to it being a handheld.
I'm at a loss here...

I would be happy to see Nintendo deploying something like that:
x1 A73 with 2 MB of L2, x4 A35 1 MB of L2. One GPC, one SM, 128 Cuda core able to process FP16 at twice FP32 speed, 1 triangle every two cycles, 4 ROPs and a good amount of L2.
As for the RAM I would be happy with 1GB of fast RAM.
I think (hope?) you're the only one who would be happy with that because those specs sound like it would perform like a piece of crap.
Why spend transistors on a lot of GPU L2 if all other resources are so feeble? Unless "good amount" means something like 128KB..

For a device with a single screen I would stick to 480p / FWVGA as there is a lot of content optimized for that rez more than for qHD /540p or the 600*1024 that appeared on many cheap tablets.
No can do. The original Xbox called and it wants its resolution back.
 
Nvidia will be talking about Tegra Next at Hot Chips next month

Denver cores did not work out so well for Android, would they be better for a gaming console?
 
Why spend transistors on a lot of GPU L2 if all other resources are so feeble? Unless "good amount" means something like 128KB.

Presumably, the idea is to reduce # of off-die accesses for texture ops.

edit: or whatever it is nV does with it.
 
I think (hope?) you're the only one who would be happy with that because those specs sound like it would perform like a piece of crap.
Why spend transistors on a lot of GPU L2 if all other resources are so feeble? Unless "good amount" means something like 128KB..

It's customary for a mobile chip to have more transistor budget than power budget, so spending budget on L2 is not bad.
But for that same reason the chip might as well have two SM, too.
As for GPU L2, perhaps guaranteeing the lowest end Maxwell has 512K or whatever it is something the devs and software would want. For reference GTX 960 has 1MB L2 and GTX 750 has 2MB L2.
Alright, I've looked for it and Tegra X1 has got 256K L2 ( good pdf http://international.download.nvidia.com/pdf/tegra/Tegra-X1-whitepaper-v1.0.pdf )
 
Last edited:
Denver cores did not work out so well for Android, would they be better for a gaming console?
I expect it shouldn't pose an issue for developers making games tailored for the NX, though quick ports may not be as easy.

It's customary for a mobile chip to have more transistor budget than power budget, so spending budget on L2 is not bad.
But for that same reason the chip might as well have two SM, too.
As for GPU L2, perhaps guaranteeing the lowest end Maxwell has 512K or whatever it is something the devs and software would want. For reference GTX 960 has 1MB L2 and GTX 750 has 2MB L2.
Alright, I've looked for it and Tegra X1 has got 256K L2 ( good pdf http://international.download.nvidia.com/pdf/tegra/Tegra-X1-whitepaper-v1.0.pdf )
We will just have to wait until next month to get some solid information as speculation can only take us so far, though I do hope they can increase L2 to at least 512K.
 
If they went with FP16 then this adds even more headaches for developers as it is not just about porting-development HW considerations but also key aspects of the game-rendering engine from FP32 to FP16.
Cheers
I don't think is that much of a headache, it is used a lot in the mobile world. Sounds actually pretty trivial to me.

On the topic of Denver cores, I am not sure it is worth the investment for Nvidia if the architecture can't cope with X86 too. ARM has better cores now than the one Nvidia provides or the in use in the XB1 and to PS4.

I still think that comparison s between the NX and the XB1 are out of place. Nintendo uses Tegra X1 in the devs kit but In don't think they will use that SOC in the final product. The X1 is pushed quite hard in the Shield TV which is not mobile device and so operates within completely different power constraints.

I suspect the Tegra X1 Is the only way for Nintendo to get it's hand on IPs relevant to them: recent Nvidia GPU and low power CPU. I keep as a reference the latest mobile shipping product from Nintendo, the new 3DS. In that context a cluster of A35 on a finfet process should allows a massive jump in sustained CPU performances enough to catch up with their Wii U. Passing on a higher performances CPU cluster (A72 or A73) altogether may free the silicon budget for the second SM Nintendo may need if they really want the system to perform decently on a TV (IMHO a bad idea as it raise s the requirements for the system and that is asking a lot for a mobile Device. There are mobile device that allies for that type of performances and tricks though their form factors is completely different: laptops).

For the purpose of converging their product line, I how Nintendo chose wisely and went with a cheap SOC deployable on other upcoming products.
 
Last edited:
I don't think is that much of a headache, it is used a lot in the mobile world. Sounds actually pretty trivial to me.

On the topic of Denver cores, I am not sure it is worth the investment for Nvidia if the architecture can't cope with X86 too. ARM has better cores now than the one Nvidia provides or the in use in the XB1 and to PS4.

I still think that comparison s between the NX and the XB1 are out of place. Nintendo uses Tegra X1 in the devs kit but In don't think they will use that SOC in the final product. The X1 is pushed quite hard in the Shield TV which is not mobile device and so operates within completely different power constraints.

I suspect the Tegra X1 Is the only for Nintendo to get it's hand on IPs relevant to them: recent Nvidia GPU and low power CPU. I keep as a reference the latest mobile shipping product from Nintendo, the new 3DS. In that context a cluster of A35 on a finfet process should allows a massive jump in sustained CPU performances enough to catch up with their Wii U. Passing on a higher performances CPU cluster (A72 or A73) altogether may free the silicon budget for the second SM Nintendo may need if they really want the system to perform decently on a TV (IMHO a bad idea as it raise s the requirements for the system and that is asking a lot for a mobile Device. There are mobile device that allies for that type of performances and tricks though their form factors is completely different: laptops).

For the purpose of converging their product line, I how Nintendo chose wisely and went with a cheap SOC deployable on other upcoming products.
I guess it is perspective, I was thinking about it from AAA games with engines developed around FP32 that are not on mobile; I think the hybrid design-concept could be a headache as it is not the best or possibly even ideal mobile or home console for developers unless they HW is a 2-part solution and game development differentiates between them and maybe for certain games even focus on one over the other; depends how much Nintendo commits NX to 'home' console concept or whether they will look at a more dedicated separate model.
I do not know much about that though on mobile (and shows lol), and not sure how many are like the Unreal engine that does seem to be flexible for mobile.
Agree it will not be Tegra X1 Maxwell SOC, but to me it would be some kind of X2 but that is also going beyond what I think your point was with regards to power constraints.

I am not sure how they are going to overcome the power contraints with mobile gaming if they want the hybrid to be diverse, which will be made much worst if they want to use AR; Pokemon Go as an example can have massive power drain for mobile phones (more down to GPS and maps I assume), but still AR could still use a fair amount of power in alternative game styles.

Thanks
 
I would be surprised if Nintendo goes with a Tegra X2, it is clearly a matter of personal perspective but I can't see Nintendo make that big of a step on the hardware. The tegra X2 is likely to be high end SOC, even if it doesn't sell that well Nvidia is better off shipping an over the top SOC even if it hardly fit with the markets requirements the product tries to reach. I guess it is a matter of brand perception for them.
May be they could have a Tegra X2i for Nintendo but I suspect Nintendo would ask for something custom. I can also see Nvidia insist so Denver is in (1 Denver core + x4 A35, could work).

Ultimately it is a matter of selling price.
 
In that context a cluster of A35 on a finfet process should allows a massive jump in sustained CPU performances enough to catch up with their Wii U.
A modern ARM core should completely stomp U-CPU, whose roots are now older than many, if not most high school students today.
 
Eurogamer says the development kit has audible fan noise. But we know that no Nintendo handheld ever utilizes a fan inside it, and NX is probably no exception. Therefore if NX really adopts a new-gen tegra SOC, the peak performance (with the dock) may not exceed current tegra X1 because of lacking active cooling. However for mobile use NX can have better performance at the same power consumption.
 
And x86 goes back to....? ;)
That's not the same thing at all. UCPU is the Gamecube's CPU with three cores jammed together and more cache glued onto there, and GC's CPU was a very lightly modified PowerPC core from the late 1990s.

Today's x86 cores on the other hand are not very like their late '90s versions; a lot has changed under the hood since then.
 
I just want to point out that by the time the NX launches 16nm will not be bleeding edge; SnapDragon 830 10nm early 2017.
I think this is a good strategy to go with 16nm, it's cheap enough currently to be possible, but also offer significantly better performance than 20/28nm SoC's; The price will drop fast.
 
It wouldn't be beyond technical possibly for the NX part to be on 10nm, given the timeframe. Just beyond anything Nintendo has done with process nodes ever. :)

10nm would be a good fit really. It is aimed at high performance mobile chips after all. Apple will probably suck up TSMC's capacity into 2017 though.

Edit: probably 2nd half of 2017 for 10nm A11 so not at all likely for NX
 
Last edited:
I just want to point out that by the time the NX launches 16nm will not be bleeding edge; SnapDragon 830 10nm early 2017.
I think this is a good strategy to go with 16nm, it's cheap enough currently to be possible, but also offer significantly better performance than 20/28nm SoC's; The price will drop fast.
TSMC 16nm FFC or FF+ should both work out nicely.
As liolio reinforced, ultimately it comes down to cost. It may be that it is ultimately cheaper all around to go with nVidias off the shelf product, even though Tottentranz is correct when pointing out that all consoles so far (original xbox?) has shipped with custom silicon. Custom silicon has two things going for it - cost optimisation by cutting out functionality that is redundant for the console application, and the capability of adding dedicated functionality. On the other hand, costs of rolling new silicon are increasing, and thus the cost benefits of increasing production volume of nVidias standard offering may be greater if you are cautious in terms of projected NX console volumes.
It lies in nVidias interests to achieve high volumes of their next Tegra and thus lower costs for all their potential customers. I can definitely see them making Nintendo a sugar coated offer to ensure that competitive edge overall.
It will be interesting to eventually see how it all balanced out.
 
Status
Not open for further replies.
Back
Top