By using this site, you agree to our Privacy Policy and our Terms of Use. Close

Forums - Nintendo Discussion - VGC: Switch 2 Was Shown At Gamescom Running Matrix Awakens UE5 Demo

Soundwave on 21 September 2023
haxxiy said:
Soundwave said:

I won't be posting here much because I find the discourse is just not very good, but I will take some time to respond to this well thought out post. 

The problem that I think arises is there's really no need for 1536 CUDA cores to get that performance. You could get the same performance from 1024 CUDA cores and just clock them higher (which has no effect on the cost) and the chip would be cheaper and have better yields. Having a massive chip like that for no reason just doesn't make sense, your yields will be worse making production more expensive and you're paying for a more complex chip for no reason. 

There is, if you're taking power consumption into account. It scales linearly with frequency but quadratically with voltage, which needs to be higher at higher clocks. A smaller chip at higher clocks would consume more power even if it performs the same.

Also, mind that 5 nm is significantly more expensive than 8 nm. You'll get more dies per wafer in the former, yes, but said wafers are significantly more expensive (I couldn't find the exact figures for Samsung, but TSMC's 10FF process, which is comparable in complexity and size feature to Samsung's 8nm, was ~2.6 times cheaper than N5).

That being said, obviously the bigger node comes with significantly higher power consumption, so lower battery life, more heat, etc. So while I agree it should go for N5, I'm just pointing out that the smaller node would be chosen because of these other considerations, not necessarily SoC cost.

5nm should be significantly cheaper itself in 2024 too ... Apple is moving off 5nm that's a huge reduction in 5nm business on the Pro line of phones and M3 chips right there, and 5nm TSMC business is already down 26% from its peak apparently, that number will probably go north of 40 or 50% next year. 50 series Nvidia cards are going to be on 3nm also so there's another chunk of 5nm business out the door. So there should be open availability. 5N will be 4 years old by the time Switch 2 launches most likely (2nd half 2024), the 20nm process for the Tegra X1 was only 3 years old when the Switch 1 launched. 

I think Nintendo also has to account for what is the overall pricing going to be for a product cycle of 8 years, at least the first 2-3 years in total not just what is cheaper on week 1 as well, over 3-4 years, 5nm could very well over the long term be a most cost efficient choice. 

There are also yield issues with larger chips, a smaller chip is likely (if on a mature node) to have fewer defects than a larger chip, it's simply more mm of space for something to go wrong and you end with an unusable chip that has be binned or thrown out. There's also the issue of Samsung's nodes (especially 8nm) sucking ass and not being able to hit yield/performance targets ... Nvidia did use Samsung 8nm very briefly with the 30 series and then quickly ditched them to move everything to TSMC 7nm ... I'm guessing that Samsung sale's pitch didn't translate in reality to what Nvidia was hoping for. 

A chip way bigger than the Steam Deck APU almost the size of a PS5 SoC inside of a console that's supposed to be smaller than a Steam Deck it's just a really weird idea and I can't recall any console really doing something like that. Certainly not from Nintendo. 

If anything I think the timing actually matches up with a lot of things ... remember when Furukawa in 2020 started making statements that the Switch was only entering the beginning of its middle life cycle? If it is 5nm/4N, they likely knew back then that they were going to wait until the process became more mature and cheaper and Switch 2 was never gonna come out any time soon. It seems awfully convenient the Switch 2 looks to be getting ready to launch in 2024 right as 5nm is getting more open capacity and probably going to be cheaper as it's no longer the cutting edge. 

As an aside (not saying Nintendo planned for this) but TSMC is hurting for business right now, they're experiencing negative growth for the first time in 14 years:

https://english.cw.com.tw/article/article.action?id=3417

If Nintendo has a deal with them, they likely have quite a bit of leverage right now because TSMC really needs their business especially with Apple moving off 5nm (which was projected years ago, everyone knew that Apple would move to 3nm in either 2023 or 2024), that's a lot of open 5nm business that some one has to fill in and generally that's good for a Nintendo/Nvidia. 

The other can of worms is even if it's Samsung, Samsung has 5nm too. 

Last edited by Soundwave - on 21 September 2023

Around the Network
sc94597 on 21 September 2023
zeldaring said:
sc94597 said:

There is a wide gap between "not being an idiot" and knowing every facet about the hardware one is ostensibly leaking. As he will admit, he gets things wrong about a third of the time (i.e RTX 3090Super, RTX 2080ti Super, etc.)

Notice that he doesn't respond to the many people querying about the apparent contradiction between "Switch = T239" and "8N". 

it looks like a cut down version of t239 so if that happens i win the bet, we don't need to wait for ports lol.

@Bolded Do you mean the T234? 

And no, we'll still stick to the original terms of the bet (which actually favor you if it is a Lovelace chip, you're set free.) 

zeldaring said:
haxxiy said:

There is, if you're taking power consumption into account. It scales linearly with frequency but quadratically with voltage, which needs to be higher at higher clocks. A smaller chip at higher clocks would consume more power even if it performs the same.

Also, mind that 5 nm is significantly more expensive than 8 nm. You'll get more dies per wafer in the former, yes, but said wafers are significantly more expensive (I couldn't find the exact figures for Samsung, but TSMC's 10FF process, which is comparable in complexity and size feature to Samsung's 8nm, was ~2.6 times cheaper than N5).

That being said, obviously the bigger node comes with significantly higher power consumption, so lower battery life, more heat, etc. So while I agree it should go for N5, I'm just pointing out that the smaller node would be chosen because of these other considerations, not necessarily SoC cost.

Really? We had our own vgchartz scientist state that it's  cheaper to make 5nm Then 8nm.

https://www.semianalysis.com/p/ada-lovelace-gpus-shows-how-desperate

SemiAnalysis sources indicate that the wafer cost of TSMC N5/N4 is more than 2.2x that of Samsung 8nm. With that wafer cost increase comes 2.7x higher transistor density. Nvidia’s top-end die went from 45 million transistors per millimeter squared (MTr/mm2) to 125 MTr/mm2. A fantastic density increase that is closer to 2 process node shrinks than 1 process node shrink. 

That was a year ago, when one would expect 5nm to be nearer to its peak cost (due to the current move to 3nm and the fact that the RTX 4000 series was imminently releasing then.) 

Last edited by sc94597 - on 21 September 2023

sc94597 on 21 September 2023
haxxiy said:
Soundwave said:

The problem that I think arises is there's really no need for 1536 CUDA cores to get that performance. You could get the same performance from 1024 CUDA cores and just clock them higher (which has no effect on the cost) and the chip would be cheaper and have better yields. Having a massive chip like that for no reason just doesn't make sense, your yields will be worse making production more expensive and you're paying for a more complex chip for no reason. 

There is, if you're taking power consumption into account. It scales linearly with frequency but quadratically with voltage, which needs to be higher at higher clocks. A smaller chip at higher clocks would consume more power even if it performs the same.

I mean this analysis was done already.

12SM (1536 cores) (assuming T239) just doesn't make sense on Samsung 8nm from a power-per-performance perspective. This is because the voltage-frequency curve flattens at very low frequencies (in this case <470Mhz.)

https://famiboards.com/threads/future-nintendo-hardware-technology-speculation-discussion-st-read-the-staff-posts-before-commenting.55/page-1142#post-683773

There are two things to take from the above. First, as a general point, every chip on a given manufacturing process has a peak efficiency clock, below which you lose power efficiency by reducing clocks. Secondly, we have the data from Orin to know pretty well where this point is for a GPU very similar to T239's on a Samsung 8nm process, which is around 470MHz.

---

That is, if the power budget is 3W for the GPU, and the peak efficiency clock is 470MHz, and the power consumption per SM at 470MHz is 0.5W, then the best possible GPU they could include would be a 6 SM GPU running at 470MHz. Using a smaller GPU would mean higher clocks, and efficiency would drop, but using a larger GPU with lower clocks would also mean efficiency would drop, because we're already at the peak efficiency clock.

In reality, it's rare to see a chip designed to run at exactly that peak efficiency clock, because there's always a financial budget as well as the power budget. Running a smaller GPU at higher clocks means you save money, so the design is going to be a tradeoff between a desire to get as close as possible to the peak efficiency clock, which maximises performance within a fixed power budget, and as small a GPU as possible, which minimises cost. Taking the same example, another option would be to use 4 SMs and clock them at around 640MHz. This would also consume 3W, but would provide around 10% less performance. It would, however, result in a cheaper chip, and many people would view 10% performance as a worthwhile trade-off when reducing the number of SMs by 33%.

Basically 512 cores (4SM) or 768 cores (6SM) would give you better performance for the same power target, and less cost than 1536 cores (12SM), if the GPU is a T239 and if it is on Samsung 8nm. 

Last edited by sc94597 - on 21 September 2023

Soundwave on 21 September 2023
sc94597 said:
haxxiy said:

There is, if you're taking power consumption into account. It scales linearly with frequency but quadratically with voltage, which needs to be higher at higher clocks. A smaller chip at higher clocks would consume more power even if it performs the same.

I mean this analysis was done already.

12SM (1536 cores) (assuming T239) just doesn't make sense on Samsung 8nm from a power-per-performance perspective. This is because the voltage-frequency curve flattens at very low frequencies (in this case <470Mhz.)

https://famiboards.com/threads/future-nintendo-hardware-technology-speculation-discussion-st-read-the-staff-posts-before-commenting.55/page-1142#post-683773

There are two things to take from the above. First, as a general point, every chip on a given manufacturing process has a peak efficiency clock, below which you lose power efficiency by reducing clocks. Secondly, we have the data from Orin to know pretty well where this point is for a GPU very similar to T239's on a Samsung 8nm process, which is around 470MHz.

---

That is, if the power budget is 3W for the GPU, and the peak efficiency clock is 470MHz, and the power consumption per SM at 470MHz is 0.5W, then the best possible GPU they could include would be a 6 SM GPU running at 470MHz. Using a smaller GPU would mean higher clocks, and efficiency would drop, but using a larger GPU with lower clocks would also mean efficiency would drop, because we're already at the peak efficiency clock.

In reality, it's rare to see a chip designed to run at exactly that peak efficiency clock, because there's always a financial budget as well as the power budget. Running a smaller GPU at higher clocks means you save money, so the design is going to be a tradeoff between a desire to get as close as possible to the peak efficiency clock, which maximises performance within a fixed power budget, and as small a GPU as possible, which minimises cost. Taking the same example, another option would be to use 4 SMs and clock them at around 640MHz. This would also consume 3W, but would provide around 10% less performance. It would, however, result in a cheaper chip, and many people would view 10% performance as a worthwhile trade-off when reducing the number of SMs by 33%.

Basically 512 cores (4SM) or 768 cores (6SM) would give you better performance for the same power target, and less cost than 1536 cores (12SM), if the GPU is a T239 and if it is on Samsung 8nm. 

Yeah I pretty much agree, it's hard to argue with any of that. 

The 5N process is pretty damn old by 2024 too ... it will be older than the Tegra X1's 20nm was for 2017 (when the Switch 1 launched) and its been known for years that Apple was going to eventually move to 3nm and I'm sure Nintendo also knew that Nvidia themselves would move to 3nm after the 40 series years ago. 

This is not like "OMG! Nintendo never does that! Cutting edge!" ... this is an older node for 2024 than what the Switch was using when it launched. 

Last edited by Soundwave - on 21 September 2023

zeldaring on 21 September 2023
sc94597 said:
haxxiy said:

There is, if you're taking power consumption into account. It scales linearly with frequency but quadratically with voltage, which needs to be higher at higher clocks. A smaller chip at higher clocks would consume more power even if it performs the same.

I mean this analysis was done already.

12SM (1536 cores) (assuming T239) just doesn't make sense on Samsung 8nm from a power-per-performance perspective. This is because the voltage-frequency curve flattens at very low frequencies (in this case <470Mhz.)

https://famiboards.com/threads/future-nintendo-hardware-technology-speculation-discussion-st-read-the-staff-posts-before-commenting.55/page-1142#post-683773

There are two things to take from the above. First, as a general point, every chip on a given manufacturing process has a peak efficiency clock, below which you lose power efficiency by reducing clocks. Secondly, we have the data from Orin to know pretty well where this point is for a GPU very similar to T239's on a Samsung 8nm process, which is around 470MHz.

---

That is, if the power budget is 3W for the GPU, and the peak efficiency clock is 470MHz, and the power consumption per SM at 470MHz is 0.5W, then the best possible GPU they could include would be a 6 SM GPU running at 470MHz. Using a smaller GPU would mean higher clocks, and efficiency would drop, but using a larger GPU with lower clocks would also mean efficiency would drop, because we're already at the peak efficiency clock.

In reality, it's rare to see a chip designed to run at exactly that peak efficiency clock, because there's always a financial budget as well as the power budget. Running a smaller GPU at higher clocks means you save money, so the design is going to be a tradeoff between a desire to get as close as possible to the peak efficiency clock, which maximises performance within a fixed power budget, and as small a GPU as possible, which minimises cost. Taking the same example, another option would be to use 4 SMs and clock them at around 640MHz. This would also consume 3W, but would provide around 10% less performance. It would, however, result in a cheaper chip, and many people would view 10% performance as a worthwhile trade-off when reducing the number of SMs by 33%.

Basically 512 cores (4SM) or 768 cores (6SM) would give you better performance for the same power target, and less cost than 1536 cores (12SM), if the GPU is a T239 and if it is on Samsung 8nm. 

They talk about why Microsoft chose 6nm on the xbox x refresh for 2024 saying it was probably to save money at the 21 minute mark, and it's probably why Nintendo went with 8nm they are all about profits from day 1 they must have got a mind blowing deal.



Around the Network
Soundwave on 21 September 2023
zeldaring said:
sc94597 said:

I mean this analysis was done already.

12SM (1536 cores) (assuming T239) just doesn't make sense on Samsung 8nm from a power-per-performance perspective. This is because the voltage-frequency curve flattens at very low frequencies (in this case <470Mhz.)

https://famiboards.com/threads/future-nintendo-hardware-technology-speculation-discussion-st-read-the-staff-posts-before-commenting.55/page-1142#post-683773

There are two things to take from the above. First, as a general point, every chip on a given manufacturing process has a peak efficiency clock, below which you lose power efficiency by reducing clocks. Secondly, we have the data from Orin to know pretty well where this point is for a GPU very similar to T239's on a Samsung 8nm process, which is around 470MHz.

---

That is, if the power budget is 3W for the GPU, and the peak efficiency clock is 470MHz, and the power consumption per SM at 470MHz is 0.5W, then the best possible GPU they could include would be a 6 SM GPU running at 470MHz. Using a smaller GPU would mean higher clocks, and efficiency would drop, but using a larger GPU with lower clocks would also mean efficiency would drop, because we're already at the peak efficiency clock.

In reality, it's rare to see a chip designed to run at exactly that peak efficiency clock, because there's always a financial budget as well as the power budget. Running a smaller GPU at higher clocks means you save money, so the design is going to be a tradeoff between a desire to get as close as possible to the peak efficiency clock, which maximises performance within a fixed power budget, and as small a GPU as possible, which minimises cost. Taking the same example, another option would be to use 4 SMs and clock them at around 640MHz. This would also consume 3W, but would provide around 10% less performance. It would, however, result in a cheaper chip, and many people would view 10% performance as a worthwhile trade-off when reducing the number of SMs by 33%.

Basically 512 cores (4SM) or 768 cores (6SM) would give you better performance for the same power target, and less cost than 1536 cores (12SM), if the GPU is a T239 and if it is on Samsung 8nm. 

They talk about why Microsoft chose 6nm on the xbox x refresh for 2024 saying it was probably to save money at the 21 minute mark, and it's probably why Nintendo went with 8nm they are all about profits from day 1 they must have got a mind blowing deal.

It still doesn't make sense to use 12SM, you would use 6SM in that case and just clock up because that's even cheaper. There is no charge for clocking up whereas for actual GPU cores you are paying $$$. Microsoft has never made money on any XBox they've sold according to them so I'm not really sure if they're an example of anything. And who's to say Nintendo couldn't use 6nm TSMC either. 



sc94597 on 21 September 2023
zeldaring said:

They talk about why Microsoft chose 6nm on the xbox x refresh for 2024 saying it was probably to save money at the 21 minute mark, and it's probably why Nintendo went with 8nm they are all about profits from day 1 they must have got a mind blowing deal.

Yes, it would make sense to move to 6nm from 7nm, as Sony had done with their newer PS5 revision (it saved them money and alleviated their shortage on 8nm.)

See: https://www.pcmag.com/news/new-ps5-model-uses-more-efficient-oberon-plus-6nm-chip

This detracts from the idea that Nintendo would go 8nm to save money. Both Sony and Microsoft went 8nm, and realized they needed to quickly choose a more recent fab because of how poor 8nm's yields were. Sony in 2021, and Microsoft in 2024. 

Nvidia switched from 8nm Samsung (Ampere) to 5/4nm TMSC. They don't make chips from 6nm Samsung.



Soundwave on 21 September 2023
sc94597 said:
zeldaring said:

They talk about why Microsoft chose 6nm on the xbox x refresh for 2024 saying it was probably to save money at the 21 minute mark, and it's probably why Nintendo went with 8nm they are all about profits from day 1 they must have got a mind blowing deal.

Yes, it would make sense to move to 6nm from 8nm, as Sony had done with their newer PS5 revision (it saved them money and alleviated their shortage on 8nm.)

See: https://www.pcmag.com/news/new-ps5-model-uses-more-efficient-oberon-plus-6nm-chip

This detracts from the idea that Nintendo would go 8nm to save money. Both Sony and Microsoft went 8nm, and realized they needed to quickly choose a more recent fab because of how poor 8nm's yields were. Sony in 2021, and Microsoft in 2024. 

I get the feeling MS just gives zero fucks about the XBox Series S/X at this point, lol. Like Sony went to 6nm years ago ... MS did ... nothing. Sony is releasing a Pro, MS is doing nothing but going a 6nm shrink that Sony already did. 

They know this generation is already a wash out and they won't be able to win marketshare from the PS5, Phil even basically admits as much. 

They just want to get out of this generation with the Activision-Blizzard deal going through, I think they don't want to sink any more money into the XBox Series S/X hardware than they have to because they don't believe it will make a tangible difference in sales. 

They're going to focus probably making a big push against Sony with the next console when they have the next Elder Scrolls game + more Bethesda exclusives + Call of Duty on Game Pass full time. And probably will have bought some more studios in the mean time. I think they're already focusing on the next console. The next few years for MS is just to get all these new studios integrating into making games under their MS leadership. 

Last edited by Soundwave - on 21 September 2023

zeldaring on 21 September 2023
sc94597 said:
zeldaring said:

They talk about why Microsoft chose 6nm on the xbox x refresh for 2024 saying it was probably to save money at the 21 minute mark, and it's probably why Nintendo went with 8nm they are all about profits from day 1 they must have got a mind blowing deal.

Yes, it would make sense to move to 6nm from 7nm, as Sony had done with their newer PS5 revision (it saved them money and alleviated their shortage on 8nm.)

See: https://www.pcmag.com/news/new-ps5-model-uses-more-efficient-oberon-plus-6nm-chip

This detracts from the idea that Nintendo would go 8nm to save money. Both Sony and Microsoft went 8nm, and realized they needed to quickly choose a more recent fab because of how poor 8nm's yields were. Sony in 2021, and Microsoft in 2024. 

Nvidia switched from 8nm Samsung (Ampere) to 5/4nm TMSC. They don't make chips from 6nm Samsung.

My point was that 5/4nm seems   expensive since Microsoft stuck with 6nm, and  between nintendo paying high cost for 5/4nm they probably went 8nm to save money. i mean anything could happen but every souce says 8nm but you right you never know till they open the console up unless nintendo actually announces specs which they haven't for years. 



Chrkeller on 22 September 2023

I suspect the Switch 2 reveal will be much like the iPhone 15. Impressive graphics for mobile but well short of hype.

I'm sticking with my Xbox one to ps4 visual/performance expectations.

The one huge advantage the switch 2 will have compared to iPhone, Nintendo games are cartoon visuals...  ps4 running cartoon aesthetics is going to be nice.  Meaning power isn't as important for some like LM4 as it will be for something like RE9.

I also believe people are underestimating the pst because cross generation games held it back significantly.  The next batch of ps5 games will be a huge step up, especially as developers get better with the hardware.  As an example uncharted 2 smokes uncharted 1.