Nintendo Is Hiring A Machine Learning/DLSS Engineer
Nintendo Is Hiring A Machine Learning/DLSS Engineer
by IHateMyselfButNotYou
8 Comments
LittenPark
“yeah it’ll make ya lunch”
antoni_o_newman
I made a post about Nintendo and this kind of development but got downvoted.
BurnsVeecteem
Judging by the recent hardware leaks I’m surprised they are only hiring a single DLSS engineer now. I thought they’d have a small team on lock as of like a year ago
IntrinsicStarvation
Tensor cores have vastly more throughput for mixed precision than generic graphics shaders like cuda cores.
If your cuda cores get 3 Tflops fp32, tensor cores can get 8x throughout with 24 Tflops fp16.
And yeah data types like int4 are 32x faster that’s 96 Tops.
But this compute power can’t be directly applied to graphics rendering, you have to find ways to leverage ‘ai’ like DLSS or Ray Reconstruction.
Currently tensor cores are vastly under used, they are active at like 90%+ utilization for a fraction of a millisecond, do all the work for stuff like dlss stupid fast, and then mostly idle the rest of the render time.
Yeah I bet nintendo wants to hire more people to come up with ways to use the part of their hardware thats like 8x-32x faster than the cuda cores as much as possible to fill the rest of that render time with work.
QuinSanguine
A as in 1? Sort of like how they had one guy program the online part of Mario Maker 2? Idk about this, lol.
puttje69
That would be nice if that was the real Switch 2 design, looks clean
gmancy
Great. That means they just started making the switch 2
ertaboy356b
Maybe they are porting Machine Learning libraries like pytorch to the switch.
8 Comments
“yeah it’ll make ya lunch”
I made a post about Nintendo and this kind of development but got downvoted.
Judging by the recent hardware leaks I’m surprised they are only hiring a single DLSS engineer now. I thought they’d have a small team on lock as of like a year ago
Tensor cores have vastly more throughput for mixed precision than generic graphics shaders like cuda cores.
If your cuda cores get 3 Tflops fp32, tensor cores can get 8x throughout with 24 Tflops fp16.
https://thebetterparent.com/blog/wp-content/uploads/2023/06/nvidia-cuda-cores-vs-tensor-cores-whats-the-difference.jpg
And yeah data types like int4 are 32x faster that’s 96 Tops.
But this compute power can’t be directly applied to graphics rendering, you have to find ways to leverage ‘ai’ like DLSS or Ray Reconstruction.
Currently tensor cores are vastly under used, they are active at like 90%+ utilization for a fraction of a millisecond, do all the work for stuff like dlss stupid fast, and then mostly idle the rest of the render time.
Yeah I bet nintendo wants to hire more people to come up with ways to use the part of their hardware thats like 8x-32x faster than the cuda cores as much as possible to fill the rest of that render time with work.
A as in 1? Sort of like how they had one guy program the online part of Mario Maker 2? Idk about this, lol.
That would be nice if that was the real Switch 2 design, looks clean
Great. That means they just started making the switch 2
Maybe they are porting Machine Learning libraries like pytorch to the switch.