I don’t think that’s what was suggested.Why some of you think there is a possibility of having a home console instead of a hybrid system? For the nm? Or for the power consumption ?
Why some of you think there is a possibility of having a home console instead of a hybrid system? For the nm? Or for the power consumption ?
TSMC's N7 process node family can't be completely discounted since Nvidia's currently using that for fabricating several datacentre products (e.g. A100 (here, here, and here), BlueField-3 (here and here), etc.).It’ll either be Samsung 8nm used in Ampere chips or TSMC 4nm used in Ada Lovelace chips.
Not really, since foundry companies don't publicly disclose process node yield rates. Most of the information about process node yield rates come from unofficial sources, which can vary in terms of reliability. This article from Anandtech lists the estimated yield rates of TSMC's and Samsung's 4 nm*/3 nm* process nodes from unofficial source(s), which should be taken with a healthy grain of salt.Are there any concrete statistics on node yields out there? That'd be very helpful in future node debates.
Oh snap yaw! They may actually be taking docked mode up to home console territory like I've been hoping! Fingers crossed we're getting a 20w+ home console mode pushing boundaries of what Drake can achieve!
I know there's no way to know the node until a tear down is performed and we take a look ourselves.
But if they reveal this thing and it's yet again another tiny tablet except this time it's housing an 8 inch screen. Is it safe to assume it's 4nm since 8nm would get really hot with such a form factor? Or does this not matter because it comes down to the clocks as well as the power draw and not just the node?
Same way we will probably guess 8nm if it's on the bulkier side with the Deck and Ally.
Well... that's true. It would be a bit sad if the chip does not reach its maximum power as possible in docked mode.That's not what's being suggested.
The hope is that Nintendo slams the system with power (20w+) to crank the GPU/CPU clocks as high as they can feasibly go, giving an extra boost when using it docked.
Some users argue Nintendo will keep a 2:1 power ratio of docked to handheld, others like myself hope Nintendo allows for a fair bit more docked power as the chip is certainly capable of it.
They could have it in a tiny thin tablet, but they'd have to downclock it so bad it would be a huge disappointment.
don't think that was ever a possibility. hitting limits all the time means lower life span, even if it's rated for itWell... that's true. It would be a bit sad if the chip does not reach its maximum power as possible in docked mode.
Wouldn't it also mean poorer yields?don't think that was ever a possibility. hitting limits all the time means lower life span, even if it's rated for it
It's already been established that the 12SMs on 8N, even if downclocked to hell, would draw double the energy of the 2 Maxwell SMs on 20nm (a la Erista), so having it also be thin would be unfeasible unless cores are being disabled in handheld (be it through FLCG or whatever other means). If GA10F is actually fabbed on 8N, I'm more than likely waging on its SoC being more comparable to the Steam Deck's power draw & accordingly being shaped as such.They could have it in a tiny thin tablet, but they'd have to downclock it so bad it would be a huge disappointment.
TSMC's N7 process node family can't be completely discounted since Nvidia's currently using that for fabricating several datacentre products (e.g. A100 (here, here, and here), BlueField-3 (here and here), etc.).
Not really, since foundry companies don't publicly disclose process node yield rates. Most of the information about process node yield rates come from unofficial sources, which can vary in terms of reliability. This article from Anandtech lists the estimated yield rates of TSMC's and Samsung's 4 nm*/3 nm* process nodes from unofficial source(s), which should be taken with a healthy grain of salt.
* → a marketing nomenclature used by all foundry companies
Here's the actual patent.
Oh snap yaw! They may actually be taking docked mode up to home console territory like I've been hoping! Fingers crossed we're getting a 20w+ home console mode pushing boundaries of what Drake can achieve!
De la Switch 2 va a necesitar más que eso por lo que tengo entendido, ya que creo que lo que hemos visto filtrado fue un test arm+turing sin Tensor y he oído comentar que la versión final será otra con arm+ampere con Tensor, ya que las pruebas con DLSS no fueron bien.
From the Switch 2 it's going to need more than that from what I understand, as I think what we've seen leaked was an arm+turing test without Tensor and I've heard comment that the final version will be another one with arm+ampere with Tensor, as the DLSS tests didn't go well.
Esto último ya digo que he escuchado y no sé si la fuente sería fiable o no, pero al parecer Nintendo originalmente quería la GPU sin Tensor, pensando que su modelo de reescalado iría bien solo con los CUDA.
NVIDIA parece haber intervenido ofreciendo un segundo SOC con Tensor.
This last I say I've heard and I don't know if the source would be reliable or not, but apparently Nintendo originally wanted the GPU without Tensor, thinking that their rescaling model would go well with just the CUDAs.
NVIDIA seems to have stepped in by offering a second SOC with Tensor.
Thank you for actually linking to the source instead of just a tweet.
Who is he that are communicating with in those tweets? His content is interesting.
That's not what's being suggested.
The hope is that Nintendo slams the system with power (20w+) to crank the GPU/CPU clocks as high as they can feasibly go, giving an extra boost when using it docked.
Some users argue Nintendo will keep a 2:1 power ratio of docked to handheld, others like myself hope Nintendo allows for a fair bit more docked power as the chip is certainly capable of it.
It's a patent for a heat sink, I think people are getting ahead of themselves. There's the matter of the "rule of patents", sure, but this is a patent to place the heat sink directly over the SOC... And that's pretty much it! It's a miniaturised cooling system for a device like Nintendo Switch.
I'm not sure removing the heat pipe will get us 5-10W additional heat dissipation, y'know?
However, yeah, improved cooling is a plus, whether or not they go with it, but this isn't indicative of power consumption.
Given the size of the example motherboard, compared with the existing Switch, this is, in my view, more likely to be related to the various "Switch micro" type patents we've been seeing, like in-line shoulder buttons.
Who is he that are communicating with in those tweets? His content is interesting.
De la Switch 2 va a necesitar más que eso por lo que tengo entendido, ya que creo que lo que hemos visto filtrado fue un test arm+turing sin Tensor y he oído comentar que la versión final será otra con arm+ampere con Tensor, ya que las pruebas con DLSS no fueron bien.
From the Switch 2 it's going to need more than that from what I understand, as I think what we've seen leaked was an arm+turing test without Tensor and I've heard comment that the final version will be another one with arm+ampere with Tensor, as the DLSS tests didn't go well.
Esto último ya digo que he escuchado y no sé si la fuente sería fiable o no, pero al parecer Nintendo originalmente quería la GPU sin Tensor, pensando que su modelo de reescalado iría bien solo con los CUDA.
NVIDIA parece haber intervenido ofreciendo un segundo SOC con Tensor.
This last I say I've heard and I don't know if the source would be reliable or not, but apparently Nintendo originally wanted the GPU without Tensor, thinking that their rescaling model would go well with just the CUDAs.
NVIDIA seems to have stepped in by offering a second SOC with Tensor.
Also have to take into account the other advantages of the tech. The Switch was able to do what it could with its limited RAM bandwidth because of Maxwell's ability to do tile-rendering, which splits the final buffer into small tiles that can fit and be processed/rendered in cache, which is faster than main RAM. How much faster, I don't know (anyone have those numbers?), and one can expect that to be the case with the Switch 2.The memory bandwidth would put something of a limit on how much more powerful you could go, unfortunately. Though it does seem like the RT and DLSS acceleration wouldn't be quite as bandwidth-hungry as the shader rendering, so pushing a chip like this higher may well be a little more efficient.
Maybe what happened is Nintendo was in fact originally planning on a Tegra X2 Switch Pro (not just Mariko with increased clocks) but Nintendo was not liking the upscaling results for Tegra X2 so it, and with it plans for a Switch Pro in general, was scrapped in favor of T239 Drake that NVidia has been pushing as their full on Switch 2. Would likely mean they have had to suddenly spend a lot more time and resources in 2021, 2022, and 2023 to shift to Drake, including figuring out how to get Backwards Compatibility working on Drake.1) There's no Arm based SoC with a Turing based GPU from Nvidia. (There's Xavier, but that has a Volta based GPU.)
2) The only Arm based SoC that doesn't have a GPU with Tensor cores after the Tegra X1 is the Tegra X2. And considering the Tegra X2 doesn't have DP4a instructions support, despite Nvidia advertising the Tegra X2 having a Pascal based GPU, and Nvidia advertising Pascal GPUs being the first to have DP4a instructions support, the Tegra X2's GPU's probably very similar, if not identical, to the Tegra X1's GPU, practically speaking.
Basically, GamerBits comes off as very questionable.
none of that sounds likely. the Tegra X2 has even more useless silicon than the TX1 with the denver cores. die shrinking the TX1, which was compatible with the 12nm node, was the right call.Maybe what happened is Nintendo was in fact originally planning on a Tegra X2 Switch Pro (not just Mariko with increased clocks) but Nintendo was not liking the upscaling results for Tegra X2 so it, and with it plans for a Switch Pro in general, was scrapped in favor of T239 Drake that NVidia has been pushing as their full on Switch 2. Would likely mean they have had to suddenly spend a lot more time and resources in 2021, 2022, and 2023 to shift to Drake, including figuring out how to get Backwards Compatibility working on Drake.
Also seems notable that NIntendo has at least been looking at ways to do more cooling. Suggests that whatever they have been working on will likely need much better cooling than the Switch has.
DLSS 1.0 used Tensor cores (here and here). (DLSS 1.9, which used CUDA cores instead of Tensor cores, was only used in Control.)Maybe what happened is Nintendo was in fact originally planning on a Tegra X2 Switch Pro (not just Mariko with increased clocks) but Nintendo was not liking the upscaling results for Tegra X2 so it, and with it plans for a Switch Pro in general, was scrapped in favor of T239 Drake that NVidia has been pushing as their full on Switch 2.
4nm and 8 NM Are most likely because Nvidia already have products there, but we can't completely rule out something in between.I know there's no way to know the node until a tear down is performed and we take a look ourselves.
But if they reveal this thing and it's yet again another tiny tablet except this time it's housing an 8 inch screen. Is it safe to assume it's 4nm since 8nm would get really hot with such a form factor? Or does this not matter because it comes down to the clocks as well as the power draw and not just the node?
Same way we will probably guess 8nm if it's on the bulkier side with the Deck and Ally.
Considering the recent Switch pocket patent, I don't rule out that a second tx1 die shrink has ever been on the table.drake isn't a replacement for a pro. it was always the successor and began earlier than 2021. if there was a Switch Pro, it was gonna be an upclocked Mariko, nothing else
Oh snap yaw! They may actually be taking docked mode up to home console territory like I've been hoping! Fingers crossed we're getting a 20w+ home console mode pushing boundaries of what Drake can achieve!
Yeah it's been talked about a bunch but people keep sending his tweets about the patents here regardless.I thought Nash Weedle was decided to be pretty unreliable? I feel like this thread has covered his track record before.
add it to the cycle I guessYeah it's been talked about a bunch but people keep sending his tweets about the patents here regardless.
That’s why Samsung 8nm completely dosent make senseIt's already been established that the 12SMs on 8N, even if downclocked to hell, would draw double the energy of the 2 Maxwell SMs on 20nm (a la Erista), so having it also be thin would be unfeasible unless cores are being disabled in handheld (be it through FLCG or whatever other means). If GA10B is actually fabbed on 8N, I'm more than likely waging on its SoC being more comparable to the Steam Deck's power draw & accordingly being shaped as such.
Yeah it's been talked about a bunch but people keep sending his tweets about the patents here regardless.
I am not sure I see the issue here. His rumors might be sketchy at best, but the patents he reports are not rumors, they are public information.add it to the cycle I guess
It doesn’t involve collusion at all, just arriving at a similar conclusion. Each manufacturer has the same problem in isolation of one another as there is in aggregate, that higher volume and lower prices can only at best match their current profiteering on these cards. Producing more to earn the same amount of money is a losing proposition, every single time, and that means even if prices go down, they will not do so far enough to be affordable from a cost per GB perspective, because more production will mean these companies will want to make more money, not the same amount.Unless every single manufacturer colludes, at least tacitly, to keep the prices high, it seems pretty unlikely that the current prices could hold with the introduction of a new customer base that is both far larger than the current one and much more price sensitive. Being used in a device like a Switch could even be the tipping point that leads to mass adoption of the format.
Well typically the parents don't mean anything for stuff that will actually release, this thread has discussed this a lot. Nintendo isn't going to show their hand before they play it. So for him to report on the parents isn't wrong or anything, but him connecting them to the Switch 2 is for clicks.I am not sure I see the issue here. His rumors might be sketchy at best, but the patents he reports are not rumors, they are public information.
Xbox doing a weird semi proprietary thing makes it not the best comparison, since competition is only barely extant, and I don't think we know whether or not MS is taking a cut.It doesn’t involve collusion at all, just arriving at a similar conclusion. Each manufacturer has the same problem in isolation of one another as there is in aggregate, that higher volume and lower prices can only at best match their current profiteering on these cards. Producing more to earn the same amount of money is a losing proposition, every single time, and that means even if prices go down, they will not do so far enough to be affordable from a cost per GB perspective, because more production will mean these companies will want to make more money, not the same amount.
Like, even despite CFe cards being physically incompatible without jury-rigging a solution and therefore not able to influence the regular CF Express market with lower pricing, Xbox Series using CF Express for storage upgrades hasn't made them cheaper (despite being produced in a higher volume than the entire pre-existing CF Express card market as was mentioned), 1TB is USD$150. And it's not like we can say it's locked behind the platform holder like Vita cards, either, WD and Seagate are competing brands offering the same storage and connection method.
Well... Patents are objective data, not clues from insiders.Yeah it's been talked about a bunch but people keep sending his tweets about the patents here regardless.
they're also mostly useless as you won't be able to glean much from them regarding hardware prior to Nintendo officially announcing the device.Well... Patents are objective data, not clues from insiders.
There are exceptions to that rule. PS Portal was patented years ago, for the PS4.they're also mostly useless as you won't be able to glean much from them regarding hardware prior to Nintendo officially announcing the device.
doubt it'll apply to anything regarding Switch 2 which is the main subject.There are exceptions to that rule. PS Portal was patented years ago, for the PS4.
Yes, they are... Like 90% (more or less) of the contents of this topicthey're also mostly useless as you won't be able to glean much from them regarding hardware prior to Nintendo officially announcing the device.
Nintendo Labo had its patents published before reveal, of course, but I doubt we see something like that anytime soon.There are exceptions to that rule. PS Portal was patented years ago, for the PS4.
This is as much a revelation as Nintendo Switch Has Games.The patent lets us know, without a doubt, that Nintendo is working on novel cooling solutions for portable consoles. That is interesting information.
Can it be done? YesIs 350MHz portable, ~700Mhz docked a unrealistic expectation under 8nm drake?
For portable? Absolutely. VCmin of 8N based on Orin is around 420MHz. Below that, you're not saving energy because of parasitic/static power. And based on Thraktor + others power curves extracted from T234, RTX 2050, etc, even at 420MHz GPU clock for HH, we're looking at 6W for GPU power budget Handheld. Tegra X1 Erista (20nm) used around 3W for the GPU iirc:Is 350MHz portable, ~700Mhz docked a unrealistic expectation under 8nm drake?
Code:
Code:Clock W per TPC 0.42075 0.96 0.52275 1.14 0.62475 1.45 0.72675 1.82 0.82875 2.21 0.93075 2.73 1.03275 3.32 1.23675 4.89 1.30050 5.58
I believe in 4nm simply because it's going so save Nintendo some money throughout ReDraketeds whole lifetime compared to some savings early on with 8nm.
John Carmack said back on PS4/XOne days ( And Mark Cerny also said similar with PS5): A bigger but slower GPU might be lesser performant in real world scenario because it's harder for developers to leverage its performance. No GPU is 100% utilized. A smaller, but faster clocked GPU might win in real world due to being more easy to leverage performance and the higher clocks meaning that everything inside the GPU is running faster. It's the biggest reason why we're questioning SEC 8N for T239. Why would Nintendo design a GPU so big just for it to be clocked at Vmin clocks? Sure, there are concerns due to power efficiency, but a smaller but faster clocked GPU would be much more suitable on 8N and have better yields.
If they designed such a big GPU, they might have done so because they know they're fabbing it on a node which they can clock it at decent clocks and take advantage of the GPU performance. Otherwise, it makes zero sense for them to design and fab a 12 SM GPU on SEC8N. Unless they discovered some profound power and area optimizations and/or Switch 2 will be much more similar in size and weight to SteamDeck, ROG Ally, Legion Go, etc.
If is on 4N what make the biggest sense they are not doomedI'll copy what I said elsewhere:
That being said, Nintendo always had some problematic and somewhat puzzling decisions with hardware design, so who knows. Nintendo is doomed anyway...