By using this site, you agree to our Privacy Policy and our Terms of Use. Close
curl-6 said:
Pemalite said:

7nm fabs are up and running now, nVidia just needs to work with TSMC and bring the design over. (Not as simple as it sounds sadly.)

Yeah. Easy doubling in performance with lowered clocks+voltages+binning over the X1 at the same TDP.

On the bandwidth front, I would assume Nintendo would make a few cut backs there, Half of 137GB/s is still a respectable 68.5GB/s of bandwidth, which is still a substantial increase over the current Switch.

A slight-more-than-doubling of horsepower and bandwidth sounds like a pretty good deal for a mid-gen Switch Pro; higher res in games with adaptive resolution, less slowdown in GPU/bandwidth bound games.

Would there be compatibility issues though, I mean Tegra Xavier is Volta right, would base Switch games have any Maxwell-specific stuff that wouldn't translate directly?

Also would a 2020 release for such a revision be viable or is that too early?

Nope, should translate fine.
At the end of the day... Pascal wasn't a big deviation from Maxwell and Volta builds upon the foundations from Maxwell.

Plus games are not generally built to the metal anymore, they target the low levels API's at most. 2020 would probably be a good time frame... 7nm would be ramped up and pretty economical by then.

Bofferbrauer2 said:
Pemalite said:

Yeah it is. But the same held true to the Tegra X1 (Tegra Drive CX/PX) and X2 (Tegra Drive PX2).
Tegra can scale upwards and downwards fairly well in power consumption... Tegra X1 and Tegra X2 were 20w parts in the "drive' configuration, where-as Xavier is 30w... And Tegra X1 was still forced to scale downwards for the Switch anyway.

But... In saying that, it is a massive chip, so it would likely benefit from being fabbed at 7nm which will solve the bulk of issues anyway... But the real benefit would be the big uptick in bandwidth plus a more modern and efficient GPU architecture.

Well those had a maximum TDP of 7.5-15W. And Nintendo had to drastically lower the clock speed to not suck the battery dry in an instant on the 15W X1 already, and not even docked does it run at full speed.

The X2 possibly could run at full speed (or near full speed at least) as it's less consuming that the X1. In fact, if it can run at full speed docked (Denver cores deactivated), then the Switch would be at more than half performance of the XBO.

But Xavier would need to be clocked down so much I don't know if that's even worth it anymore. Unless most of those 30W would come from the Tensor cores (can be deactivated) or NVidia wants to Die-shrink Xavier (which I both seriously doubt), I fear Xavier doesn't make too much sense.

The Car variants of the Tegra X1 and X2 had higher TDP's than 7.5w or 15w. They were 20w. Xavier is 30w, but you can scale all these chips upwards and downwards to hit various performance/power targets, all comes down to how much voltage you are willing to dump into it to drive up those clocks.

Xavier can also have some crap turned off to save power like the Tensor cores and one of the Quad-core CPU partitions.

Volta based Tegra does have higher IPC than Pascal as well remember, so even at lower clocks, she can do more work.




www.youtube.com/@Pemalite