Support NeoGAF

[naguanatak]

Otherwise, why did you respond to me?

Asking myself that same question right now.
Most of all that is subjective. I can go into the Steam Deck OT right now and preach how I think the NS2 is much better designed. Sure, I can do more with a Steam Deck, but the Switch 2 is - in my case - the better device. I reported that after having purchased a used Steam Deck, tested it and resold it to buy a Switch 2. My reporting that the Steam Deck is too much hassle was met with disdain and a lack of comprehension. All good, I'm happy with my purchase having tested both and deciding what is best in my use case.
It's like eating steak all day and talking about it... it's okay for people. But answer a question why you don't eat it and they will pounce on you because 'you HAVE to tell everyone ALL THE TIME' because people get triggered for different reasons.
I guess I was triggered by reading the same shit on here all the time. Didn't even check who wrote the stuff this time, honestly. Have a couple of free minutes at the desk at work and thought I respond...
No worries though. Everyone has the right to be upset and complain and come into topics to praise theirs... have at it.

 

[naguanatak]

Rodrigo is a long history of coming in every switch 2 threads to promote steamdexk

You think his post is all innocent but when the guy made an argument that steamdeck is better because you can run cyberpunk 2077 with path tracing on it (hint it's a fucking mess, but he's all for it!) and the irony of calling 30 fps shitshow while his suggestion looks like pixel soup….then my threshold of trust that he's posting with good will is gone down the drain. Want to white knight the troll? Go for it

Fun times. I remember that.

 

[Panajev2001a]

Slapping 10nm here like it's a bad thing shows how much people don't understand these technologies and put too much emphasis on a node number

Call it 8nm vs TSMC 5nm, but even the same 7nm nominal process from both in no metric really beats their competition. It is a bit cheaper to build on, sure. It is fair to care about this but as price of the HW increase the more we can care if the vendor smooths some corners.

 

[Panajev2001a]

Nvidia does not push high density silicon for their mobile tech

Repeat after me, they will if paid to do so.

Tapeout is 2021, revisions happened afterwards. Tapeout marks the end of design phase, not that it's ready to produce and error free. What does that matter?

AMD allegedly proposed their biggest deal for an APU and it was refused. There's a lot of factors here than a wishlist. True engineers know what I'm talking about.

True engineers apparently are only allowed to praise Nintendo for clearly punching above their weight with a perfect HW design that does not cut corners.

Come on man, nobody is insulting or being aggressive but you keep losing patience (I get it, it happens to everyone) and treat people blatantly like idiots beneath you.

 

[Panajev2001a]

You want to produce, MASS produce a chipset and remain low cost

I am not asking them to steal nodes from Apple and compete at that level, not sure why the only choice is either one extreme or the status quo.

What about the console screams low cost to you though? Do we really think they are losing massively over every piece of HW sold and not building upon the great job done with Switch and capitalising on the money potential there?

 

[FireFly]

nVIDIA had Ampere ported to TSMC 7nm and Blackwell now is on their optimised 5nm (4LP) process. I am not finding sources that do not generally show Samsung processes being behind the "equivalent" TSMC ones and optimised 10nm vs 5nm TSMC mmmh… I find it hard to believe it is as a wash as you say (otherwise Apple would not spend so much money to be on the very latest process from TSMC for their A series SoCs either ):

I believe for 7nm vs 10nm TSMC advertised a 20% higher clock speed at the same power, or 40% less power at the same clocks. So for 8nm to 7nm, we could be talking about a 15% clock speed improvement. If we look at Ampere on 8nm vs Ada on 5nm, we get around a 50% improvement in performance per watt. However we would expect the improvement to be less here, since T239 already incorporates Ada's clock gating.

 

[Emedan]

You don't need to because of how they nailed the design. Children to adults can comfortably play on them without issue.

Nintendo handhelds always have that flat shape that makes the hands cramp up with any extended play for anyone who doesn't have the small hands of a child.

P o r t a b i l i t y.

There's always a trade off. Nintendo chose portability above performance and according to you comfortability, nothing wrong with that - you as a consumer can choose what's most important for you and buy your products with that in mind. Seeing how Switch 2 already have outsold the Steam Deck (I think?) I'd say Nintendo hit right aligning the design of the Switch 2 with consumers preferences.

I wonder how many consoles Nintendo really gotta sell before this nonsens talking point about high performance hardware being the end all be all for gaming, it obviously isn't. Fun gaming experiences are what matters, not Ghz, resolutions and framerates (unless it's so disruptive it effects the enjoyment of the game to a too large degree).

 

[Buggy Loop]

Repeat after me, they will if paid to do so.

I think they know how to design and advice a customer

Also, generally not for GPUs

Look at any Nvidia GPU

They never tap anywhere near the density of a node. Including TSMC made ones.

Stability is king, low density is king also for cooling.

These density gains evaporate for wide lower frequency platforms.

While it's generally true that lower lithography nodes (smaller nm) benefit chip performance, especially for high-frequency applications, low-frequency chipsets don't see the same degree of improvement. This is because the primary advantages of smaller nodes are reduced power consumption and increased transistor density, which directly impact speed and efficiency in high-frequency circuits. Low-frequency chips may still see some benefits from lower power consumption, but the performance gains from increased clock speed are less significant

True engineers apparently are only allowed to praise Nintendo for clearly punching above their weight with a perfect HW design that does not cut corners.

That's your interpretation? Lol

Just before that I said there's a lot of factors here outside of just a wishlist. How did you interpret that as perfect?

But even with a lot of decisions that lead to mass production and likely package deal for other components, Nvidia engineers optimized accordingly.

Is this chipset not efficient?

Stop focusing on a node number for a minute

Is it not efficient?



Beating a TSMC 6nm OLED model with +25% battery mAh

Uh

 

[rnlval]

I understand why it'd be running low in handheld mode, to save battery life.

But c'mon, 1 GHz is so low you can't expect good performnce out of the CPU at these clock speeds, even on power efficient ARM.

Curiously enough, the clock speeds in docked mode are actually lower than in handheld mode, you would think it'd be the other way around. Now I also understand why you want them to be similar, because graphics can be scaled more easily between the modes but the CPU performance is a more fixed target. Basically you won't want to have a situation where a game could run in docked mode but couldn't in handheld mode.

However, take a look how Cyberpunk shipped:




It's obvious some developers don't care about abysmal performance.

In my opinion, Nintendo should add a mode where they heavily overclock the CPU in docked mode, to atleast 2 GHz or higher for titles like that, that would help immensely. Handheld will continue to run shitty like that but atleast in docked mode it would run decently.

Of course I don't know the thermal headroom they have in docked mode but I'd assume the temperatures are pretty low due to the additional fan. I think the charger also delivers up to 60W which should be more than enough.

NVIDIA hasn't design market leading smart phone SoC to save thier ass.

 

[Zathalus]

Beating a TSMC 6nm OLED model with +25% battery mAh

Uh

It's a bit more impressive than that. The battery of the Steam Deck OLED has 2.6x the capacity over the Switch 2.

The eventual refresh will likely have further significant battery life gains when not on Samsung 8nm. I do wonder if they will change to TSMC 4 or Samsung 4. 3+ hours on that battery capacity would be seriously impressive.

 

[rnlval]

Good for you! Buy it.

Slim enough... lol



They balanced everything for the form factor and LPDDR5x bandwidth limitations

Which clearly every other PC handhelds outside Steam deck fucked up big time. Doesn't matter those high clocks, they're starving on bandwidth. Super inefficient. They're engineering jokes, totally unbalanced, a shitload of computational power left idle. Too much compute for what it can feed. There's no "Steam deck 2" until memory has a huge boost, or you have APUs with tons of cache.

AMD hasn't integrated the Strix Halo GPUs' L3 infinity cache design in mainstream APUs. Besides GPU scaling, Strix Halo's iGPU has one major additional improvement over Strix Point's iGPU i.e. large L3 cache. AMD would need to evolve Strix Point scale APUs in addition to porting RDNA 4 or UDNA/RDNA 5 improvements into mobile IGPU.

Per CU, both RDNA 3.5 and RDNA 4 have double-texture sampling fetch improvements.

It's a bit more impressive than that. The battery of the Steam Deck OLED has 2.6x the capacity over the Switch 2.

The eventual refresh will likely have further significant battery life gains when not on Samsung 8nm. I do wonder if they will change to TSMC 4 or Samsung 4. 3+ hours on that battery capacity would be seriously impressive.

Samsung has 7 nm (7LPP), 5 nm (5LPE), 4 nm (4LPP, 4LPP+, 4LPE), 3 nm (SF3).

 

[Xeela_sunston]

That's why I'm not going to raise a pitchfork over this. If this was across the board I would lambast Nintendo, and I wouldn't feel bad doing it. If a company screws up, they should be called out for the screwup. I do think they could have mitigated this concern when they designed the console, but unless we see a widespread issue specifically revolving around this issue I don't think this is worth getting that upset over. Realistically, I'm not sure what the real-world difference would be with a difference of 103MHz. It is weird that the docked version has a (marginally) lower CPU clock speed than the OG Nintendo Switch, but at this point I don't know what the reasonings were for this.

CDPR not fully using the 6c/6t threads of the A78c since their porting a game that expects x86 CPU's for what arch ARM uses.

 

[rodrigolfp]

And he'll note that XeSS is gonna fix all that up

Who said XesS would gonna FIX it, freak??? Make it better than FSR =/= fix. Not even DLSS quality is a perfect fix for IQ.

Rodrigo is a long history of coming in every switch 2 threads to promote steamdeck

Pointing clowns comparing SW2 Docked mode vs Deck "handheld mode" using shittier FSR instead os XesS that is better =/= promoting. Keep crying in blind defense of SW2.

 

[rnlval]

CDPR not fully using the 6c/6t threads of the A78c since their porting a game that expects x86 CPU's for what arch ARM uses.

ARM Cortex A78
Integer pipelines
P0: Branch
P1: Branch
P2: ALU
P3: ALU
P4: ALU / MUL
P5: ALU / MAC /DIV

FP / SIMD pipelines, each pipeline is up to 128 bits wide.
P6: FPU / FADD / FDIV / ALU / IMAC
P7: FMUL / FADD / ALU

When compared to AMD Zen 2, ARM Cortex A78 is weaker with SIMD resources.
From https://fuse.wikichip.org/news/3536/arm-unveils-the-cortex-a78-when-less-is-more/
VS

AMD Zen 2
Integer pipelines
P0: ALU
P1: ALU
P2: ALU
P3: ALU
P4: AGU (Ld/St)
P5: AGU (Ld/St)
P6: AGU (St)

FP / SIMD pipelines, each pipeline is up to 256 bits wide.
P7: FADD
P8: FMA / FMUL
P9: FADD
F10: FMA / FMUL

X86 fused load/store with ALU operations instruction set. AVX2 supports GPU-style gather instruction.

https://en.wikichip.org/w/images/f/f2/zen_2_core_diagram.svg

For mobile phone SoC market, there's a problem with using the ARM Cortex X series license with a 3rd party GPU from 2025. https://semianalysis.com/2022/10/28/arm-changes-business-model-oem-partners/