Nintendo Switch 2

Technically it has 3TF, but FP32 doesn't have much effect on gaming performance, so cutting it in half probably gets you a more realistic number. If it was 3TF FP16 it could run this game at 4k 120fps :yes:
Ampere is about 20-30% faster than Turning per SM for the doubled FP32. So you can multiply the TF figure by 0.7 or 0.875 to get a rough idea of the comparable RDNA 2/GCN part, respectively.
 
Last edited:
Ampere is about 20-30% faster than Turning per SM for the doubled FP32. So you can multiply the TF figure by 0.7 or 0.875 to get a rough idea of the comparable RDNA 2/GCN part, respectively.
I remember reading an article where they tested if FP32 made a difference and the results were that it was around 10%. Can't find that article right now.
 
I remember reading an article where they tested if FP32 made a difference and the results were that it was around 10%. Can't find that article right now.
You can just compare the 3080 with the 2080 Ti. They have the same number of SMs and the clock difference is marginal, yet on average the 3080 is around 30% faster.
 
Half the power consumption of a console in a handheld isn't great! Hence the lame battery life. Now had Nintendo gone with a smaller process, they'd use less power and get more portable functionality from their portable. So not looking like nonsense to me. ;)

Agree. There is usually no price benefit for being patient with Nintendo. However waiting for the new chip that would presumably come when Switch 2 Lite releases might actually be worth it in terms of battery life.
 
You can just compare the 3080 with the 2080 Ti. They have the same number of SMs and the clock difference is marginal, yet on average the 3080 is around 30% faster.
That doesn't take in to account all the other architectural advancements that are in Ampere. We can't give all the merit to FP32 while ignoring all the rest.
 
That doesn't take in to account all the other architectural advancements that are in Ampere. We can't give all the merit to FP32 while ignoring all the rest.
I thought the main other improvements were concentrated in RT. In any case though it doesn't matter, since the Switch 2 will inherit the performance improvements from Ampere, not just the extra FP32.
 
I thought the main other improvements were concentrated in RT. In any case though it doesn't matter, since the Switch 2 will inherit the performance improvements from Ampere, not just the extra FP32.
The point still stands, that FP32 is a small improvement for performance and that the Switch 2 performs more like a 1,5TF GPU than a 3TF one.
 
The point still stands, that FP32 is a small improvement for performance and that the Switch 2 performs more like a 1,5TF GPU than a 3TF one.
It doesn't, because if, as expected, the Switch 2 performs like the other Ampere parts, then it should be 20%-30% faster per SM than Turing, which is already significantly faster than GCN on a per teraflop basis.
 
Sorry must be missing something here. Did they annnounce the node size? What’s this talk of 8nm over 5nm?
 
It doesn't, because if, as expected, the Switch 2 performs like the other Ampere parts, then it should be 20%-30% faster per SM than Turing, which is already significantly faster than GCN on a per teraflop basis.
Maybe instead of a 5x multiplier it's 6x? Not much changes. Memory bandwidth is also only 4 times higher, so that will probably be the limitation from reaching max throughput.
 
Maybe instead of a 5x multiplier it's 6x? Not much changes. Memory bandwidth is also only 4 times higher, so that will probably be the limitation from reaching max throughput.
At 3.1 TF, it should perform like a 2.7 TF GCN part. So it's expected to be in between the PS4 and PS4 Pro in docked mode, which seems to bear out in the inital footage of games like Cyberpunk. (The memory bandwidth is in line with other Ampere parts).
 
At 3.1 TF, it should perform like a 2.7 TF GCN part. So it's expected to be in between the PS4 and PS4 Pro in docked mode, which seems to bear out in the inital footage of games like Cyberpunk. (The memory bandwidth is in line with other Ampere parts).
That would mean that it is far more powerful than a Steam Deck. I hope so, that would be awesome, but I doubt it.
 
our friend of nVidia Hair technology.

We have the content that was leaked a while ago:

Portable:

CPU: 998.4 MHz
GPU: 561 MHz (~1.72 TFLOPS)
Memory frequency: 4266 MHz
Memory bandwidth: 68.256 GB/s

Docked:

CPU: 1100.8 MHz (assuming swapped)
GPU: 1007.25 MHz (~3.09 TFLOPS)
Memory frequency: 6400 MHz
Memory bandwidth: 102.4 GB/s

CPU is a ARM Cortex A78c - 8 cores
GPU has 1536 CUDA cores - Ampere architecture
12 GB RAM LPDDR5X

It looks very similar to the MX570.

Ampere Architecture
2048 CUDA cores
4.731 TFLOPS
Memory Bus - 64 bit
Bandwidth - 96.00 GB/s
Base Clock - 832 MHz
Boost Clock - 1155 MHz


Elden Ring and Cyberpunk looks similar to Switch 2 versions, in visual and performance.
It obviously suffers more due to the lack of VRAM.

GAMING Switch 2 has '10x the graphics performance' of the Nintendo Switch NVIDIA has posted some information on the GPU found inside the Nintendo Switch 2, and with RT and Tensor Cores and DLSS, it's a massive improvement

NVIDIA confirms the Switch 2 can handle ray-traced reflections, shadows, and 'dynamic lighting'
 
Back
Top