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 120fpsit's more like 3TF in docked mode no ?

Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
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 120fpsit's more like 3TF in docked mode no ?
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.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![]()
FP16 isn't better than 32. 32 just doesn't make much of a difference for gaming.First Time i read fp16 is better suited than fp32 for gaming, where did you get that info ?
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.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.
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.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.
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.![]()
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.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.
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.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.
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.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.
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.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.
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.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.
It's mostly speculation and some uncertain leaks. And after the showcase I'm not completely sure anymore.Sorry must be missing something here. Did they annnounce the node size? What’s this talk of 8nm over 5nm?
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).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.
That would mean that it is far more powerful than a Steam Deck. I hope so, that would be awesome, but I doubt it.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).
Cyberpunk runs at FSR balanced (536p) on the Steam Deck, while the 7 week old Switch 2 build already runs at 720p+ in docked mode.That would mean that it is far more powerful than a Steam Deck. I hope so, that would be awesome, but I doubt it.
I expect the Switch 2 to be more powerful than a Steam Deck. Just by how much? Native games and a better API can go a long way.Cyberpunk runs at FSR balanced (536p) on the Steam Deck, while the 7 week old Switch 2 build already runs at 720p+ in docked mode.
Yea I can’t see any reason for them not to use 5nm. Why pay more when you can pay less.It's mostly speculation and some uncertain leaks. And after the showcase I'm not completely sure anymore.
our friend of nVidia Hair technology.
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 improvementWe 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
![]()
NVIDIA GeForce MX570 Specs
NVIDIA GA107S, 1155 MHz, 2048 Cores, 64 TMUs, 32 ROPs, 2048 MB GDDR6, 1500 MHz, 64 bitwww.techpowerup.com
Elden Ring and Cyberpunk looks similar to Switch 2 versions, in visual and performance.
It obviously suffers more due to the lack of VRAM.