Where did that pic come from?
Should actually be somewhere around 120-140mm^2 on 5nm+, but I agree, cutting the CPU cores to 8, and dropping the cuda core count from ~1920 to 1280 or 1408 would be fine for a Switch successor, one benefit of using the full chip though is that they can actually come out with a cheaper part, and does sort of lean into what that road map is actually telling us, it seems like Nvidia wants to bin the parts, using the Switch successor to drive production of the SoC, and give higher performing parts to the high end automotive industry and defective parts to the low end AI industry. It could go either way, and the SoC could easily end up the same size as Tegra X1 (121mm^2).
So you find it more likely that the next revision will target budget consummers when the Lite is already doing that work ?
Oh, I didnt know they had a whole page report for this. I guess NeXt is out between April and June to go with the new Jetpack release.
Jetson Roadmap
developer.nvidia.com
Lite is cheaper, but I feel there is a market for a $99-149 super budget sku with no portability. In addition we already have the Calcio firmware entry which sounds very similar.
I am not saying it is useless, it definitely has some purpose but it would be a niche revision.
So, just speculation on my part, but insiders seem to be relying on Tegra X1 profiles to match the Tegra X1 with the new model, however this doesn't rule out modifications to the Tegra X1 chip beyond a die shrink, it means it's still Maxwell/Pascal with A57 cores, but it could double the cuda core count for instance, and it is possibly built on 8nm as there is that Tegra 8nm rumor from earlier in the year. I think if this is the case, we should look at 800 to 900GFLOPs with 256 cuda cores or 1.3-1.6TFLOPs with 512 cuda cores. The reason I think they might have increased the cuda core count is simply because Nano hits that $129 price point by using chips that have disabled parts, such as Tegra X1 using 128 cuda cores instead of 256, or Xavier NX (a nano like design) using 384 cuda cores instead of 512.
I'm asking my self if there is a chance of a $400 premium SKU in 2021 (running the 8 or 7 nm version of the 5 nm chip planned for 2023 lite and hybrid model) alongside two die shrunk TX1 SKUs (lite at $130~150 and hybrid at $250).
Nvidia has 2 commercial SoC on their roadmap, Nano Next for 2021 (almost certainly a cut down Switch Pro SoC) and Orin a 2022 SoC with a 5 watt to 45 watt profile, spanning Jetson price points of $249 (same as Tegra X1/X2 when launched) to $899 (Xavier AGP price point).
It would be really interesting especially if the full Orin chip is really able to scale down to 5W in a Nintendo switch scenario.
I like the idea of Nano Next being a cut down version of whatever Nintendo would get, Originally I thought that if Nano Next was marginally better than Mariko that's what they would settle on
I am just trying to imagine what the Nvidia Nano Next is. I think the extra cuda cores makes sense for a Tegra device, while extra CPU clock speed should be enough to handle regular computations, Tegra Nano series is an entry level AI chip afterall.
The PS3 is capable of 4K. Tegra X1 is also capable of 4K, I think just internal upscaling, and while I think it is possible something like doubling cuda cores could happen, I wouldn't expect it, I'm digging for more information and I'll try to come up with more credible stuff, but in the meantime, it's just about what Nano Next could be, I think we could see it at GTC this October 5th, Nvidia has a keynote and they could talk more about nano next there.
These are all marketing terms that mean nothing. 2K meaning 2000+x1440p is closer to a usable resolution than the ones in that wiki because the ratios are 3:2 not 1:2, or with the case of 2560x1440p, it is 16:9, the standard ratio of a TV.
2K resolution - Wikipedia
en.wikipedia.org
There is a rumor that some developers have at least seen it, but honestly external developers don't really need to see something like this until 5 or 6 months out. Even then, it's not mandatory for 3rd party games to utilize this new hardware, dynamic resolution would still be effected, so many games would still see a huge benefit.
You don't need to deliver a new devkit if your handheld mode is just the OG model max clocks.
Yes, but the Tegra X1+ is using a 64-bit memory bus width instead of a 128-bit memory bus width used in the Tegra X2.
Possibly a handful of key 3rd parties have dev kits by now, but there's much less of a need for early dev kits for a revision like this, since it isn't a full new platform. Unless the dev is working on an exclusive title, they can just continue to target the current Switch for now (since they'll have to do that anyway), then adjust things for Switch Pro once kits become available.
The problem is that the Cortex-A57, unlike the Cortex-A78, doesn't support DynamIQ. The Cortex-A57 supports Arm big.LITTLE, which doesn't allow heterogeneous CPU cores to be used simultaneously, which is the reason why the Cortex-A53 is disabled in the Tegra X1+.
lets not forget as well that those terms in the development world for textures is different as well ...
That's why I mention. They could just use older dev kits and up clocks
Yes.
This is a new one or the old one from march or April 2020?
Yes, I believe there was one a year ago with changes to the SOC and RAM. Besides the 12/16nm die shrink from 20nm, the only thing I saw changed besides smaller CPU cores were the LPDDR4x RAM.
The only thing I can read is the number 1923a1 from the new chip.
HAT002 EDEV Teardown Internal Photos Internal photo_HAT-002(-01)x Nintendo
EDEV 04 Internal Photo details for FCC ID BKEHAT002 made by Nintendo Co Ltd. Document Includes Internal Photos Internal photo_HAT-002(-01)xfccid.io
There are photos. Anyone want to zoom in and play with the contrast to try and read it?
I think the actual components would need to be changed. The revised Nintendo Switch model changed the SoC (going from a 20 nm Tegra X1 to a 16/12 nm Tegra X1+) and the RAM (going from LPDDR4 to LPDDR4X).
so would it be safe to assume that our baseline is another die shrink? (and maybe more ram?)
I think a die shrunk Tegra X1+ is a reasonable assumption. And I guess the new Nintendo Switch model might go for faster RAM (either by using LPDDR5 on a 64-bit memory bus width, or by using LPDDR4X, but with a 128-bit memory bus width, instead of the 64-bit memory bus width on the current Nintendo Switch model) over more RAM.
that would really help the two biggest bottlenecks with switch.
I'm guessing if the chips on the dock have changed, Nintendo would have to mention that to the FCC when making a FCC filing. And given the FCC filing that was made public on 23 August 2020 is very similar to the FCC filing that was made public on 9 July 2019, I think it's reasonable to assume the dock for the new Nintendo Switch model is the same as the dock for the current Nintendo Switch model (the revised model).
If I recall correctly, the AI upscaling on the Nvidia Shield TV only applies to media. And despite the fact that the current Nintendo Switch model and the Nvidia Shield TV share the same SoC, I believe the current Nintendo Switch model doesn't take advantage of AI upscaling when watching media (like on YouTube). So I don't think the AI upscaling on the Nvidia Shield TV will be brought over to the new Nintendo Switch model.
