-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
Nintendo Switch: Powered by Custom Nvidia Tegra Chip (Official)
- Thread starter ekim
- Start date
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
You like them titbits?
Dakhil
Member
Edit: I think I remember someone on NeoGAF talking about it, but I can't find it. But the rumour's pretty old: https://www.reddit.com/r/NintendoSw...r_someone_who_producing_switch_at_foxconn_is/
Akhe
Member
Oh, thank you so much!
Serenitynow
Member
I'm going to guess that DF's rumor came from a dev who has a devkit. There are people out there who believe the final product will be Pascal.
I still have hopes for Pascal, but there's no way in God's Green Earth we're getting a 4300 mAh battery. I'm sure there is enough space for one, but so did the Wii U's gamepad, and that shipped with something like 2300.
DF have said their information comes from a source that says the specs are directly representative of the final retail units
Yeah but the gamepad didn't need it as much, it wasn't doing much work.
The reality will probably be something in between
Serenitynow
Member
The clock speeds are final. I don't think their sources said, "This Maxwell chip will be in the final retail unit."
lwilliams3
Member
Yeah. They didn't have much more besides them saying that some stuff from Pascal development is tin the chipset.
This is more interesting now that Laura Kate Dale stated that the October dev kits are more powerful than the ones from July. As Blu suggested, though, the kits could have always been Underclocked.
ShadowFox08
Member
ughh the dev kits must in july must have been awfully weak then.
Skittzo0413
Member
The 2x FP16 performance? I think people here have stated that's only a feature of Tegra chips, so no.
That does not necessarily follow. There are any number of specific variables that could be associated with either proposed dev kit that weren't specifically and reliably confirmed that could change the performance of either considerably.
Given that this is a console that is very, very reliant upon the balance between performance and battery life and heat dissipation, I'd wager to say that Nintendo targeted a performance target and then adjusted, through iteration, to find the best way to reach that while keeping all of the aforementioned factors in mind.
That might mean that the process involved a downgrade in one or more areas, but we don't know that for sure. The unofficial info we have just suggests that (which is admittedly historically supported by Nintendo's usual M.O.)
Personally I think that if the platform sucked we would have heard more about it sucking by now. But what do I know?
lwilliams3
Member
*Shrugs*. U untill now, we thought the power was the same. We don't know how much of a difference it is.
Not true. These Tegra Pascal chips are very similar to 2nd generation Maxwell, and it shares the same GPU architecture.
Skittzo0413
Member
Hmm? The question I was responding to was whether or not desktop Pascal chips had 2x FP16 performance, and since that's a feature which is much more important for mobile devices I had thought that only Tegra Pascal (and Tegra Maxwell) chips would have that, so no desktop Pascal chips.
Unless that's incorrect...
You mean Frankenstein's monster. But yes.
To be fair, when using this analogy outside of the story itself, it is generally accepted that "Frankenstein" means the same thing that would otherwise, to those familiar with said story, mean "Frankenstein's monster." It's sort of an accepted bit of easily-communicated imagery.
http://www.anandtech.com/show/10325/the-nvidia-geforce-gtx-1080-and-1070-founders-edition-review/5
Native FP16 compute seems to be a feature only seen on Tegra and Telsa GPUs starting with the Tegra X1. The GTX Titan X is the only exception.
lwilliams3
Member
Oops. I misread the original question you were answering. Sorry about that. I thought the answer was the same, but when I read up on it..
I just finished reading that article before I came back. It is bizarre to find out that the GeForce GTX 1080 is hilariously a lot slower using fp16 than fp32.
The father, the son, and the holy Switch?
djplaeskool
Member
Lot's wife
Ask And Ye Shall Recede.
vial of myrrh oil
dr_rus
Member
It's slower when using native FP16 math which is there for compatibility purposes only (so you could debug the same CUDA code for GP100 system on your laptop with a 1060). Obviously you can use FP32 math instead or emulate FP16 on FP32 ALUs which will give you the same(ish) rate as that of FP32. Do note that native FP16 operations are not exposed in DX and other graphics APIs on GeForce 10 lineup (including Titan X) so it's not like you can even use them in gaming code - unless you're willing to use CUDA in that code.
Romans 3:23
The Last One
Member
Underrated comment of the day.
👍
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
Wait, using mediump and lower GLES declarations on a GP100 will still produce single precision arithmetics?
ed: Bah, Titan X is a GP102 - of course it won't do fp16 proper, CUDA or otherwise.
forknknife
Banned
Don't get it
Don't get it
She was turned into a pillar of salt, and there is likely to be a lot of salt one way or the other come the presentation and subsequent release, especially when looking back at the speculation.
Similar threads
- Locked