D
Deleted member 13524
Guest
How?
Its chips would still need a substrate, packaging and a high-end node from TSMC or Samsung.
"New" Nintendo Switch (OLED Model) [2021-07-06]
- Thread starter Goodtwin
- Start date
Its chips would still need a substrate, packaging and a high-end node from TSMC or Samsung.
That's my understanding of the production limitations as well.
Like my gut feel on all this is I doubt they have those lower level ingredients available but sitting idle for older mfg lines that couldn't somehow be used in the newer more financially beneficial smaller nodes. So even if someone could get time scheduled on older nodes they're still bottleneck by those items.
Megadrive1988
Veteran
Reading that sends RV770 shivers down my spine.
Yet I feel in my gut that a 1 Tflops (fp32) handheld console almost certainly isn't happening until Nintendo's next-gen hardware--Whatever that is, I don't expect it to be released before Q1 2025. Probably using one of the smaller nodes, one that is more mature by then, like one of TSMC's "N5" family, but where there is also a great deal more capacity..As TSMC's investments in new "5nm" factories are more than just built, but actively helping to turn the global semiconductor shortage into semi-boom.
Last edited:
The most probable outcome is that they want to retain backwards compatibility ...
D
Deleted member 13524
Guest
You're saying it'll take Nintendo 8 years to get 2.5x the GPU throughput that is present in the 2017 Switch?
I guess I found that one person who won't ever be disappointed by Nintendo's hardware specs.
D
Deleted member 11852
Guest
And Nintendo's record in volume manufacture is woefully behind that of Microsoft and Sony.
tongue_of_colicab
Veteran
Why not? Hardware and software are still selling strong, why should Nintendo rush to release anything other than a refresh? With chip shortages being what they are and there not really being anything in mobile land that is affordable and fast enough to get ps5/xsx ports either, there is no rush. Wouldn't be surprised if the switch isn't replaced for at least another 2 years.
Not that I mind, maybe they can finally start releasing some new games... I don't think I've bought a game made for switch since Link's Awakening came out...
D
Deleted member 13524
Guest
1) another 2 years is 2023, not 2025 as mentioned in the post I replied to.
2) there's finally some real competition coming to the Switch in the form of SteamPal (especially if the rumored price of $400 comes through), and perhaps even Qualcomm's platform becomes a good competitor (doubtful if it's just another Android platform without any real game dev effort though).
Given how well Steam machine did, I doubt it will even make a dent on Switch sales lol.
D
Deleted member 13524
Guest
Given the current state of the semiconductor industry, nothing will make a dent on Switch sales. Not even the Switch Pro.
And unless it's a terrible device, I bet the SteamPal will follow the same route as every other gaming hardware of 2021, though: sell out in a couple of seconds to scalper bots and then re-sell out at attrocious prices on ebay.
I tend to agree.
Switch will continue to sell out the rest of the year and the pro will sell all they can make of it. Steampal will sell out also but i think its more to do with the limited quantities they will have vs the switch pro
DavidGraham
Veteran
A well known leaker claims Nintendo will use a custom version of Orin S, codenamed T239
A switch with less bezel is being used as a promo image again. This time for kingdom come delivered
Rootax
Veteran
That would be a huge improvement. I don't believe it...
Would it though?
I’ve refrained from speculating for a long time, and while kopite7kimi throws out a juicy proposition, it’s still claims ”modified” for instance.
How?
What process and clocks?
And perhaps most critical of all, what memory solution?
I have a fair bit of respect for Thraktor keeping things sane at resetera, but even if we accept these rumors as Truth, they still leave as large a span for performance characteristics as reasonably informed napkin math.
I'll add this for memory subsystem reference. Currently the Switch uses LPDDR4 at a maximum of 3200MT per second docked for a maximum nominal bandwidth of 25.6GB/s.
If they use the same 64-bit interface but at LPDDR4x speeds: 4266MT/s => 34.1GB/s
64-bit interface updated to LPDDR5: 6400MT/s => 51.2GB/s
Using a 64-bit interface to LPDDR5 brings some LPDDR5 inherent advantages to the table as well, so the improvement would arguably be somewhat more than what the numbers suggest.
8GB LPDDR5 would be a single package.
Of course, Nintendo could be ambitious and double the memory interface. In which case:
LPDDR4x: 4266MT/s => 68.2GB/s nominal
LPDDR5: 6400MT/s => 102.4GB/s nominal
But without even getting into the thorny discussion about what is a good idea or indeed possible at 4k resolution and 10W at Samsung 8nm, we can see that the memory subsystem variance is HUGE!
Of course, the bandwidth needs are dependent on what you try to achieve, but it has been pointed out as a main bottleneck for many Switch titles.
On the other hand, how much can happen performance wise within the current Switch power envelope? (Should we use the original Switch model instead?)
Then we proceed to compare process technologies. Under optimal circumstances TSMC 10nm (which is similar to Samsung 8nm) is twice as dense, but doesn't offer twice the power efficiency vs. Mariko 16nm. Lets assume Samsung 8nm is better, and can offer both twice the density and twice the power efficiency. That would imply that if we could afford roughly the same die size, we could squeeze in twice as many transistors, which, when clocked the same as current models would consume roughly the same amount of energy. Assuming a bit of increased efficiency, we would also need to provide at least twice the bandwidth for the SoC not to be more hamstrung by bandwidth than the current models. Which fits neatly with a single 8GB LPDDR5 module.
This kind of Napkin Math always pegged an updated Switch at roughly 2.5x the performance of the current model. But without actually knowing neither the memory subsystem choices nor (yet) the process technology, the span is huge. Nintendo may go really cheap with the memory and opt for 64-bit LPDDR4x (ack! pfft!), or they may be really ambitious and feed it via 128bits worth of LPDDR5, and fab on Samsung 5nm. And I'm starting to feel that these critical parameters won't be known until after the device is launched making literally writing a Bible sized thread filled with speculation is a wonderful commentary on how public tech forums work. It is actually to the credit for B3D not to have engaged in this, but honestly, I mostly feel that this may be because the more prolific residents here would prefer to deny the Switches existence outright.
If they use the same 64-bit interface but at LPDDR4x speeds: 4266MT/s => 34.1GB/s
64-bit interface updated to LPDDR5: 6400MT/s => 51.2GB/s
Using a 64-bit interface to LPDDR5 brings some LPDDR5 inherent advantages to the table as well, so the improvement would arguably be somewhat more than what the numbers suggest.
8GB LPDDR5 would be a single package.
Of course, Nintendo could be ambitious and double the memory interface. In which case:
LPDDR4x: 4266MT/s => 68.2GB/s nominal
LPDDR5: 6400MT/s => 102.4GB/s nominal
But without even getting into the thorny discussion about what is a good idea or indeed possible at 4k resolution and 10W at Samsung 8nm, we can see that the memory subsystem variance is HUGE!
Of course, the bandwidth needs are dependent on what you try to achieve, but it has been pointed out as a main bottleneck for many Switch titles.
On the other hand, how much can happen performance wise within the current Switch power envelope? (Should we use the original Switch model instead?)
Then we proceed to compare process technologies. Under optimal circumstances TSMC 10nm (which is similar to Samsung 8nm) is twice as dense, but doesn't offer twice the power efficiency vs. Mariko 16nm. Lets assume Samsung 8nm is better, and can offer both twice the density and twice the power efficiency. That would imply that if we could afford roughly the same die size, we could squeeze in twice as many transistors, which, when clocked the same as current models would consume roughly the same amount of energy. Assuming a bit of increased efficiency, we would also need to provide at least twice the bandwidth for the SoC not to be more hamstrung by bandwidth than the current models. Which fits neatly with a single 8GB LPDDR5 module.
This kind of Napkin Math always pegged an updated Switch at roughly 2.5x the performance of the current model. But without actually knowing neither the memory subsystem choices nor (yet) the process technology, the span is huge. Nintendo may go really cheap with the memory and opt for 64-bit LPDDR4x (ack! pfft!), or they may be really ambitious and feed it via 128bits worth of LPDDR5, and fab on Samsung 5nm. And I'm starting to feel that these critical parameters won't be known until after the device is launched making literally writing a Bible sized thread filled with speculation is a wonderful commentary on how public tech forums work. It is actually to the credit for B3D not to have engaged in this, but honestly, I mostly feel that this may be because the more prolific residents here would prefer to deny the Switches existence outright.
Well, I have 0 inside information but I fully expect an upgraded Switch to be announced soon because I just picked up 2 more Switch Lites for my youngest 2 kids for doing well in school this year. This is just how the universe works. I spend money on something, better stuff or a price drop gets announced. So anyone waiting on the Switch Pro... You're welcome.
If that pans out... Do you have a PS5? If not, please buy one. I want PS5 slim at lower price.
Similar threads
