Pemalite said:
Soundwave said:
The current Switch is DLSS 2.0 was possible in 2015 would be basically very close to the XBox One in power straight up.
|
The current Switch does not and can not have DLSS as it lacks Tensor cores that nVidia introduced with Volta.
Soundwave said:
394 GFLOPS to 1.2 TFLOPS yes, but then you have to factor in the Tegra X1 only needing the render 1/4-1/15th the pixel resolution and that gap shrinks very quickly. AMD also routinely claims a TFLOP number and routinely is outperformed by Nvidia GPUs that have lower TF performance as well, so take AMD's teraflop claims with a grain of salt.
|
Flops isn't everything. You are conflating multiple different aspects of hardware here and it is highly erroneous.
AMD's GPU's actually beat nVidia in Tflop benchmarks. A Teraflop for AMD is the same as nVidia, it's single precision floating point. Aka. 32bit FP.
The numbers you see in comparisons however are "theoretical" - But without question, AMD beats nVidia on this front in the real world and has done so for years, especially in Asynchronous compute workloads.
If you are doing any kind of intensive FP32 workload, AMD was the GPU to have because it's theoretical and realworld Teraflop numbers absolutely beat nVidia, which is why AMD's GPU's were the GPU's to have during the crypto-craze era... Crypto mining is entirely compute bound. Aka. Uses those Teraflops and not much else.
But here-in lays the issue... Games aren't all about ALU throughput which AMD focuses far more heavily on than nVidia who typically gives more focus to pixel/texture fillrate/polygon throughput... Which are also needed for games... Ergo, nVidia is able to decisively beat AMD in gaming, despite having less real-world and theoretical Tflop numbers.
So no... Don't take AMD's "Teraflop claims" with a grain of salt... The issue lays with individuals who propagate those numbers without any real understanding of what they mean or what they actually do for gaming.
Soundwave said:
Even shit like SSD, whoopity doo, smartphones already have NVMe drives that are the same thing, Apple's iPhones have had this for 5 years already. UFS 3.1 which is going to be common in Android phones is 3GB/sec by the time Switch 2 is out UFS 4.0 likely is available which would be even faster than that.
|
Not all Smartphones use nVME. Apple has used it since iPhone 6 AFAIK.
Android typically relies on other technologies.
|
Yes of course the current Switch doesn't have DLSS, my point is *if* that technology was available back then for them and they could have implemented it, something that would increase the performance of the Switch significantly. They could render as low as like 512x288 undocked (N64 level resolution) and 540p docked and achieve 720p undocked + full 1080p docked no problem. Games like Zelda BOTW would run even above 1080p, at 900p a DLSS 2.0 could reconstruct that maybe even a full 4K resolution, but certainly 1440p or 1800p would be doable.
AMD's GPUs have for years struggled to match performance of Nvidia GPU's that are 1-2 years older and have or often times run hotter on top of that in many cases. That's all I meant, when people see like a 2070 Super is "only" 9TFLOPS, that may well perform equal to a PS5 (10TF) or even XSX (12TF) ... it wouldn't surprise me. The RDNA2 architecture they have coming now is basically what Nvidia had almost two years ago with Turing.
Apple has used NVMe for about 4-5 years now. Android makers like Samsung favor UFS, UFS 3.1 can get up to 3GB/sec which is basically as fast as an NVMe drive. By the time Switch 2 is out there probably will be UFS 4.0 available if Nintendo wants it and that will probably be even faster.
