Dagobert Duck
Member
So what would be good codenames in your eyes?
Childish...get it together man, if you want to be full on adult enjoy yourself, yet I think a little bit of "childishness" isnt that bad alltogether...
-
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
So what would be good codenames in your eyes?
Childish...get it together man, if you want to be full on adult enjoy yourself, yet I think a little bit of "childishness" isnt that bad alltogether...
Schnozberry
Member
Seems likely. He can bullshit with the best of them.
Can you theorize the performance of those specs in comparison to X1? Also do you think it will run overclocked when doc?
It'll be upclocked (in comparision to undocked) when docked and downlclocked when undocked. That's what laptops do these days.
Overclocking the chip would lead to more hardware failures.
sir_bumble_bee
Member
I feel like this thread has been going in circles for the past 20 pages...
Skittzo0413
Member
Yeah, well we don't know any more than we did a week ago yet we want to talk tech because it's fun. That's how these threads work when there's no new information coming.
That would be very cool actually. Give us a good indication of what the system is capable of, maybe even some tech demos. I feel like it would be good for both Nvidia and Nintendo.
The full clocked version of X1 has less total memory bandwidth than PS3 (same amount of GPU bandwidth), but that doesn't make it bandwidth limited by comparison. As you mention the chip has quite significant optimisations that improve memory utilization.
Considering its going to be a custom SoC they can use whatever bus width they like. Personally I think its likely they'll have some kind of dedicated video memory.
JoshuaJSlone
Member
They need some manly names like KitKat and Nougat.
The big question I have is: what kind of hardware the NS need to have to be able to run this generation's games at 720p?
Last generation, The PS Vita was what? A 24 GFlops machine that can run games at 544p or less and, sometimes, can archive visuals close to the ps3 and theirs 240Gflops.
I believe that NS will be closer to Xone than Vita was from ps3 (maybe a 500 or 750 GFlops machine in portable mode)
Other thing to consider: when ps4 pro runs games at 4K, these games will have more details than the normal ps4 or all the extra juice will be utilized only to increase resolution?
Last generation, The PS Vita was what? A 24 GFlops machine that can run games at 544p or less and, sometimes, can archive visuals close to the ps3 and theirs 240Gflops.
I believe that NS will be closer to Xone than Vita was from ps3 (maybe a 500 or 750 GFlops machine in portable mode)
Other thing to consider: when ps4 pro runs games at 4K, these games will have more details than the normal ps4 or all the extra juice will be utilized only to increase resolution?
We know nothing. Just PR-talk like "Tegra", "based on" etc.
Remember when reputable sites reported that "Wi U packs the same processor technology found in Watson, the supercomputer."
Some people just want to believe anything they read.
You know, it is possible to read rumors and take part in speculation without automatically believing everything with absolute certainty. We "knew nothing" for sure before the reveal but a substantial number of rumors turned out to be true as is often the case in the video game industry.
darkazcura
Member
I think the major difference here is that one source is from the actual supplier, and the other was speculation from reputable sites. Pretty big difference.
Believing the chip is based on either X1 or Parker isn't exactly wild speculation considering Nvidia's statement.
fredrancour
Member
internal company codenames for computer chips used to video games: too serious and mature to employ a superhero reference.
how do you come to have enough burning hatred for superheroes to think that this makes sense?
Schnozberry
Member
Are you talking about mirroring the display, or Wii U like functionality where the tablet displays a different overlay than the TV?
mirroring. Like a reverse WiiU. So eg you're playing mario maker on the TV but want to use touch to create levels. Not great if you're docked, so streaming to the TV would be good.
No it's not because the information came from IBM. Here's a tweet from the official IBM Watson account.
That turned out to be false just like so many other things.
So for the last time. It's all rumors until we have the retail version in our hands.
I have nothing more to add.
You had nothing to add to begin with. All you're doing is trying to stifle discussion.
tkscz
Member
While it wasn't a Watson, it did use components that the watson used, so they could technically make that claim. That's the difference here. Nvidia isn't claiming to be using a specific chip, only that it's custom and based on their "top Geforce" tech, and right now, Pascal is their top. If rumors are to be believed and history is to be followed, then the Tegra X2 (parker) would be the chip Nintendo would go with. If it saves battery and heat, Nintendo will go for it.
Schnozberry
Member
I don't think anything would technologically prevent them from doing it, but wouldn't an IR pointer and gyro function essentially the same way?
What form of video memory would you expect them to use, though? eDRAM isn't available on any of the plausible manufacturing processes, SRAM would be very expensive if they need 32MB+, and the only off-die memory available which would fit their needs would be HBM2, which would also be extremely expensive, even in a minimal 4GB single-stack configuration (although arguably may be a better choice than SRAM, given the orders of magnitude higher capacity).
It's kind of redundant with tile-based rendering, though, which should give them most of the benefits of dedicated VRAM with just a few MBs of cache, and also make developers' lives a little easier with a single logical memory pool.
dr_rus
Member
X1 isn't a typical Maxwell chip though as it's using 20SoC process and have different SMs. Energy improvements between desktop Maxwell and Pascal parts do not necessarily transfer to X1 and Parker comparison.
Memory controller is abstracted beyond the crossbar and can be any width the product requires. X1 can easily have 128 or even 256 bit memory bus if necessary and Parker can just as easily have the same 64 bit bus as X1. Thus why "based on Parker" isn't saying much without unit numbers / bus widths / clocks.
you need to chill out
for some types of games - like exactly mario maker - you'd want absolute pointer controls (touch immediately in a specific location) rather than a relative pointer control (like a mouse). You could maybe get away with it, but I think some things would be a lot more usable with direct touch.
Although then you'd have to undock it, which might be something Nintendo don't want to make you do..
Probably. LPDDR4 chips come in a maximum I/O width of 64 bits, meaning you'd need four chips for such a configuration, increasing costs, power consumption and physical size. Phones pretty much universally use a single 32 or 64 bit memory chip, and the only consumer product I can think of which uses more than one LPDDR module is the iPad Pro, which uses two 64 bit chips for a 128 bit bus (to allow it to drive an extremely high resolution display).
Grampa Simpson
Banned
I'm not talking about something as large as 32MB+ as it wouldn't be used for rendering anyway (with a tile buffer already being present). Maybe just a very large texture cache. I mean they may not have any separate pool. I'm only basing this on Nintendo's love of fast video memory, they seem to have it on every system they produce. But yeah considering the mode of rendering used maybe this will be the end of that.
And this is where Fujitsu's secret 128-bit bus high density FCRAM will save the day! Or they could incluse 50+ 512Mbits modules for ungodly wide I/O!
Has there been any rumor from dev kit leaks? Is it also possible the first alpha kit were x1 but the final dev kits / specs will be parker?
OK, I thought you meant an actively managed VRAM pool, but I'd definitely be in agreement with you in expecting an increased GPU L2 cache, or maybe a big shared L3.
And this is where Fujitsu's secret 128-bit bus high density FCRAM will save the day! Or they could incluse 50+ 512Mbits modules for ungodly wide I/O!
The issue is that the Wide I/O memory format, which is essentially the successor to the FCRAM used in 3DS, has never really made it into production. Every time they announced a new faster version of the standard it would just be eclipsed by LPDDR, which could provide the same bandwidth at a lower cost, with lower power consumption and on a narrower bus. There don't seem to be nearly as many exotic memory standards around for Nintendo to choose from compared to the good old days of 1T-SRAM and FCRAM.
Schnozberry
Member
Yeah, I was more thinking your second point. I certainly agree that touch is far more ideal in this scenario, but with a capacitive screen you'd need some kind of accessory pen or be forced to deal with finger level accuracy. IR might actually offer greater precision, especially since Mario Maker would need to be reimagined for pinch to zoom to get the kind of pixel precision that was offered with the original.
DeuceGamer
Member
How would that impact performance, or are there ways you expect Nintendo to make up any deficiencies?
It does, but I haven't seen it in any consumer products yet.
I'm certainly not discounting the possibility of a 128 bit bus, but I was just trying to highlight how unlikely 256 bit wide memory would be by showing how rare even 128 bit LPDDR interfaces are.
I don't think they really need that kind of bandwidth. With tile-based rendering and a large cache they should be able to get by with a 128 bit bus, or even 64 bit depending on the system's performance.
Schnozberry
Member
Oh I thought parker uses a 128 bit bus.
It does. He was just referring to how many memory chips would be required. For a 128-bit interface, they'd need two 64-bit chips. It's pretty common in the mobile space to use a single chip because of space constraints, so a single 64-bit interface is common. LPDDR4 chips with a 128-bit interface don't exist as of right now. It wouldn't be completely out of the ordinary for Nvidia to use two chips though, because the Shield Android TV has a configuration with two 32-bit chips, as does the Shield Tablet shown below.
It does. He was just referring to how many memory chips would be required. For a 128-bit interface, they'd need two 64bit chips. It's pretty common in the mobile space to use a single chip because of space constraints, so a single 64-bit interface is common. LPDDR4 chips with a 128bit interface don't exist as of right now. It wouldn't be completely out of the ordinary for Nvidia to use 2 chips, though, because the Shield Android TV has a configuration with 2 32-bit chips, as does the Shield Tablet shown below.
Oh gotcha. Sounds like we can at least expect ( or hope for) a 128-bit bus. Which would be quite an improvement over wiiu.
Ahh I see. Thanks for replying.
tkscz
Member
And this is where Fujitsu's secret 128-bit bus high density FCRAM will save the day! Or they could incluse 50+ 512Mbits modules for ungodly wide I/O!
That's something I've been wondering. Do we no they are using LPDDR4? I mean, FCRAM in the 3DS is probably a much faster RAM to go with. Not sure if it's possible or not to get the Tegra to work with it though.
We'll see, but if they do go with 128 bit lpddr4 then even an underclocked X1 could hit PS4 quality at 960x540. That would also easily upscale to 1080p and rumours are that the target resolution for NS is 540p. The only bottleneck at that point would be CPU performance.
It was coming straight out of Nvidia documentation concerning these two Tegra chips.
Schnozberry
Member
Yeah, it really depends on if they plan to use large cache pools or embedded memory. If that's the case, then they wouldn't necessarily need the extra bandwidth. Not that it would hurt.
I should correct myself from my earlier post. 128-bit LPDDR4 chips do exist, but it doesn't appear that they are made in sizes that make sense for the Switch. The largest one I can find is 2GB.
Schnozberry
Member
Yeah, 20nm planar didn't end up being much different than 28nm. Going to 3D Transistors and FinFET is a major improvement regardless.
Skittzo0413
Member
What rumor is that? The Zlatan rumor stating 504p? Hasn't that been thoroughly debunked?
LPDDR4 is far, far faster than FCRAM. On a bandwidth per chip basis it's competitive with GDDR5, and you'd need to jump to HBM or HMC for something "much faster" than it.
That's interesting, especially with it only being available at 2GB or below (wider I/O LPDDR modules are typically reserved for larger capacities). Who's manufacturing them?
Schnozberry
Member
Micron. I should confess that I may be reading this incorrectly.
https://www.micron.com/products/dram/lpdram
Micron. I should confess that I may be reading this incorrectly.
https://www.micron.com/products/dram/lpdram
Hmm, that's interesting, although I suspect the 128 bit I/O parts may be LPDDR2 or LPDDR3. I know some of Intel's laptop CPUs have supported LPDDR3, so perhaps these are for cheap netbooks/chromebooks?
Although, yeah, it's possible we're reading this wrong, as they don't specifically say "I/O width", so it could be relating to the internal organisation of the DRAM cells. (This is definitely stretching past my knowledge of how RAM operates.)
Schnozberry
Member
Looking at the Hynix and Samsung LPDDR4 product lists, I think we are reading it correctly. Although it very well could be 128-bit LPDDR3 for notebook or x86 tablet configurations.
Samsung
Hynix
All of Samsung's 4GB packages are 64-bit, which are organized as 4x16Gb stacked chips to make one module. The Cell organization seems to be directly related to that. I believe that is referred to as x64.
Similar threads
- Locked