By using this site, you agree to our Privacy Policy and our Terms of Use. Close
Pyro as Bill said:
Pemalite said:

Nope. There are dependencies in the GPU design that ties the tensor cores to other portions of the chip.

nVidia would need to start from scratch essentially for a "tensor chip". But that won't come cheap... You would be better off grabbing Tegra Xavier and disabling all the junk you don't need... But then you have the bandwidth and latency issue of USB.

Because AMD focuses on price/performance... And AMD can justify the GPU arm due to integration with their CPU products with "APU's". - AMD still doesn't have Ray Tracing yet for example in the PC space... But their GPU's being very compute-centric tends to be lucrative purchase decisions for folding, mining and other activities.

But don't kid yourself, AMD's market share is significantly smaller in the GPU space than they were pre-Graphics Core Next... And mostly because they have been a generation or two behind nVidia in terms of overall efficiency and features.

That doesn't make their products bad though, AMD just plays the pricing game instead.

Thanks Pem. Sounds like they've fallen way behind since the days of the HD4870 if Nvidia can cheat their way to 4K and double the FPS with AI.

Does that mean Sony/MS's next consoles can't do ray tracing and AI-magic or do they use their own tech for that stuff? It sounds like Nvidia is who you'd choose to make your GPU unless you want to mine dogecoin.

Next gen consoles have been confirmed to support hardware Ray Tracing... How effective and extensive that support is... Remains to be seen as AMD has not introduced that technology or even demonstrated it's capabilities in the PC space yet... And thus we are unable to dissect the technology.

Basically we need to wait and see.

As for GPU's and A.I... You don't actually need dedicated FP16/FP32/INT Matrix cores to do the calculations like nVidia has done, you can do it on the regular GPU pipelines, it will just be less efficient as you will be relying on rapid packed math... Whether AMD will have Tensor-like GPU cores remains to be seen, but AMD's GPU's have typically held the edge in pure computational throughput over nVidia so they may not see the need... But games typically need more than just compute, hence why nVidia is faster in games.

Again that will require a wait and see approach... But frame reconstruction to "fake" higher resolutions is a front-and-center focus for -all- platforms going forth, the Xbox One X and Playstation 4 Pro made it a primary issue, we are just in that "trial" phase to see what the best technological approach is, just like with Anti-Aliasing a few years ago after moving from the 7th gens morphological implementations.

If you want the best GPU technology... nVidia is simply where it is at, they are ahead of AMD in every regard... Heck, nVidia still outperforms AMD even with it's GPU's built at 12nm verses AMD's 7nm.

In saying that, if you want the best performance at a cost-sensitive price point, AMD cannot be beaten, which is why it's the only choice for high performance consoles... Especially if you want the integration of high performance CPU cores.

For mobile form factors, nVidia Tegra is definitely the best choice, Nintendo's only other real improvement on that front when the Switch launched would have been to double the Ram and include the Tegra X2 which would have provided 50% more performance at the same power consumption, but I assume nVidia put forth a pretty lucrative deal... And it's paid off for Nintendo either way.



--::{PC Gaming Master Race}::--