• 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.

DF: Nintendo NX Powered By Nvidia Tegra! Initial Spec Analysis

Status
Not open for further replies.

Neoxon

Junior Member
What if it's just tegra x1, would it be ok? does it matters.

People likes to talk about tech, but when comparing games from 3ds vs wii/wiiu. It doesn't really matter that much. If you have a base station that can increase the power to make it equal to xbox1 it would be ok.
  1. It's most likely gonna be some customization of Tegra, be it X1 or something else.
  2. We have no idea if there's gonna be extra power in the dock.
 

Thraktor

Member
Not sure what you mean as it handles benchmarks in the same way it handles real world applications, there is not difference. The offloading to A57 partition is there for tasks which benefit from multithreading more than from a high single thread performance. I would assume that a thread which needs high performance processing will be mapped to Denver2 core while lighter threads will go on A57 which are worse in ILP but are likely better in perf/watt.

Denver executes ARM instructions in one of two ways. The first is a simple binary translation layer and the second is dynamic code optimisation. The former obviously offers worse performance than the latter (otherwise there wouldn't be any need for the latter), and the DCO is only applied once a segment of code has looped a specific number of times through the binary translation layer. This means than Denver will perform better with code which contains loops with large numbers of iterations than it will with code that has few loops or loops which tend to iterate a small number of times. Benchmarks are inherently just loops with extremely high iteration counts, so basically present pretty much the best case scenario for Denver. Real world code will therefore perform similarly to benchmarks only in the best case, and worse in the average case, as a greater proportion of code is executed through the binary translation layer.

While the extent to which "real-world" code will underperform benchmarks is hard to judge (particularly without any Denver cores handy to test on), the only Denver device in the wild currently is the Nexus 9, and reviews of that pretty much universally criticised the sluggish real-world performance (despite excellent benchmark results), which would indicate that the theoretical failings of Denver do carry across to reality. With Nvidia dropping Denver from TX1, and only bringing it back for Parker now that they have a HMP-capable interconnect between it and traditional ARM cores, it seems that Nvidia understands this, and feel that their best approach for varied workloads is a mix of Denver and "big" ARM cores and, in their own words, to "Schedule the task on the right CPU core".
 

Durante

Member
To be fair, every long-running program spends the vast majority of its time running lots of iterations of loops (or levels of a recursion, which is really quite similar for the purpose of discussing code generation). I guess the question is how much code is touched within an iteration.
 
sYFg0tz.jpg
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
To be fair, every long-running program spends the vast majority of its time running lots of iterations of loops (or levels of a recursion, which is really quite similar for the purpose of discussing code generation). I guess the question is how much code is touched within an iteration.
Got to admit 128KB of L1 I$ is not the best use of ultra-expensive cache known to mankind.
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
This seems to be a confirmation of something I speculated a few months ago, that Nvidia's decision to combine Denver and A57 cores on Parker is a recognition that Denver has some fairly significant performance weak spots, and they're offloading threads which perform poorly on Denver to the A57 cores. I'd be interested to know what kind of heuristics they're using to allocate threads, although I very much doubt that's something they'd ever talk about.
The way I see it, there are two main possibilities here:
1) A57 is more power-efficient than Denver2 for a given workload. That does not put Denver2's power efficiency in particularly good light, as A57 is not the most power-efficient design in its class. Also, raises the question of why "big.SUPER" and not something more traditional like big.LITTLE with A53s.
2) A57 handles some workloads overall better (as in faster or faster + more efficiently) than Denver2, and is what Thraktor's been saying.

edit: apologies.
 

Peterc

Member
What do Nintendo need to do when revealing the nx? How should they need to do it?

What was their best announcement?


I still think their best one was the first twilight princess trailer at e3


Trailer other m was great too, even if the game was bad.

This of course is software based.

I think they did a good job with 3ds.
 

TunaLover

Member
What do Nintendo need to do when revealing the nx? How should they need to do it?

What was their best announcement?


I still think their best one was the first twilight princess trailer at e3


Trailer other m was great too, even if the game was bad.

This of course is software based.

I think they did a good job with 3ds.
TP was outstanding, but MGS:TTS was even mindblowing not only the trailer itself but the meaning of the franchise returning to a Nintendo home console.
 

Neoxon

Junior Member

BitStyle

Unconfirmed Member
Are there any other Nvidia related news to look out for in August that might be related to NX, or are our sights set on the future at this point?
 
Alright, so now it's confirmed that the FLOPS figures for Parker are FP16. https://blogs.nvidia.com/blog/2016/08/22/parker-for-self-driving-cars/?__prclt=loGXBCSr

That's a good find, thanks. So that basically tells us ~50% increased performance over the TX1. Doesn't tell us power consumption figures though.

ARM powered, a slightly better Wii U specs-wise and full of what appears to be gimmicks. Not interested at all.

Err.. and what exactly is wrong with ARM powered?
 
ARM powered, a slightly better Wii U specs-wise and full of what appears to be gimmicks. Not interested at all.

#1: the ARM in the current tegra are noticeably faster than the jaguar x86 in current gen

#2: even conservative clocks would put the portable form factor well above the wii u

#3: you can just be honest and say you weren't interested regardless, you don't have to put on a smug air about it beforehand.
 

MuchoMalo

Banned
You sure about that. Because if Eurogamer is right, we'll have to hear about the reveal soon. Even with a Digital Event, there's still press hands-on sessions.

I imagine that we'll get a digital event near the end of the month, a week after a press hands-on secret event, I think they'll announce the digital event no more than a week before. My money is on 9/22 or 9/27 for the announcement and 9/29 for the NX Direct Special Presentation.

Or maybe I'm being pessimistic for the sake of my sanity. I'm not sure at this point.
 
#1: the ARM in the current tegra are noticeably faster than the jaguar x86 in current gen

#2: even conservative clocks would put the portable form factor well above the wii u

#3: you can just be honest and say you weren't interested regardless, you don't have to put on a smug air about it beforehand.

True, I'm not interested in what Nintendo does by default since the Wii. But you'll never know, they could just have learned something from the Wii U and decided to develop a no-bullshit gaming console for a change. I'd be onboard, but I'll pass the gimmicks again.
 

ethomaz

Banned
Alright, so now it's confirmed that the FLOPS figures for Parker are FP16. https://blogs.nvidia.com/blog/2016/08/22/parker-for-self-driving-cars/?__prclt=loGXBCSr

I give this thread 5 more hours; then NX discussion is finally dead until the reveal date announcement sometime next month.
Nice to have confirmation.

So Tegra X2 has 625 GFlops of raw power... about 1220 mhz.

CPU was what us expected... 4x A57 + 2x Denver 2.0.

128bits LPDDR4 @ 50 GB/s.

Pretty good for NX mobile but I expect more for the docked mode.
 

Ryoku

Member
You all really need to cool down, but this thread isn't the place for that. Y'all already got two damn threads locked. If you even want an NX speculation thread, stay on topic or don't post at all.

No one cares about how excruciatingly painful and difficult the wait for the announcement of the reveal is to you. We know that already. This is a tech thread.
 
I think if they can get peak X1 performance but portable and passively cooled I'd be pretty happy.

That's be a sizable jump up from Wii U and using mixed precision would allow for some damn nice looking games. Especially depending on what resolution they target.

So a unicorn.
 

Knobiwan

Neo Member
I'll repeat what I said before, that the new latest timetable for announcement would be after the release of Pokemon Sun/Moon, launching November 18, 2016. Now it would be disadvantageous to announce a new console during your holiday console sale surge, especially if the new console is a hybrid handheld, where the 3DS will likely see a spike in sales this holiday due to Pokemon. This puts the expected reveal after Christmas and New Years, so a console reveal may not happen until Mid to Late January.

Alternatively, at this point, a console delay is probably unavoidable and may end up pushing this whole thing to E3 of next year, where the console won't launch until holiday of 2017.

But absolute worst case scenarios notwithstanding, I would guess February as the latest reveal date at this point in time, with an announcement of the reveal in mid to late January.

I think everyone is going to have to chill until a new leak happens, because we will theoretically be waiting a while.
 

Mpl90

Two copies sold? That's not a bomb guys, stop trolling!!!
So folks are looking to get a third thread locked?

If you're referring to me, I wanted to carry on a possibly reasonable discussion about what we seemingly got from Hot Chips on Tegra Parker and its possible implications for NX, given the most relevant rumours right now, given this is a thread about the tech behind it and, well, the last one left. I hope the conversation stays civil and doesn't go crazy in such a short amount of time as happened elsewhere.
 

NateDrake

Member
If you're referring to me, I wanted to carry on a possibly reasonable discussion about what we seemingly got from Hot Chips on Tegra Parker and its possible implications for NX, given the most relevant rumours right now, given this is a thread about the tech behind it and, well, the last one left. I hope the conversation stays civil and doesn't go crazy in such a short amount of time as happened elsewhere.
Not directly referring to anyone.
 

G.ZZZ

Member
Nice to have confirmation.

So Tegra X2 has 625 GFlops of raw power... about 1220 mhz.

CPU was what us expected... 4x A57 + 2x Denver 2.0.

128bits LPDDR4 @ 50 GB/s.

Pretty good for NX mobile but I expect more for the docked mode.

I guess it's about 6W then? I'd like to have confirmation on the power draws though.
 

10k

Banned
Since this is the tech thread I wonder if these specs can work in the NX and around 5w.

4 A72's at 1.8Ghz (All for gaming)
4 A53's at 1.2Ghz (two reserved for OS)
6GB LPDDR4 at 4000Mhz (2000Mhz handheld mode)
Tegra GPU 3 SM at 1.2Ghz (550Mhz handheld mode)

Docked mode = 922 GFLOPS
Handheld mode = 461 GFLOPS (around 5-6w)

Should be enough for 720p30 in handheld mode and 1080p30 in the dock. 720p and 1080p use the same assets so it would be cheaper and save more space on carts for Nintendo to go with a 720p screen.

If they go 540p, they'll need two sets of assets for each game.

Obviously the docked mode is the normal clock speeds and the handheld bios downclocks to half speed when out of the dock.

Should be around Xbox One in power but Unreal Engine 4 games with its efficiency with FP16 might actually run better on NX then Xbox One and maybe on par with PS4.
 

Hoo-doo

Banned
For real. What is wrong with some of you. Why do you keep asking for topics to get locked. It's annoying.

Nobody is asking for threads to be closed..

It just so happens that these NX threads get endlessly bumped with zero news and zero on-topic discussion just because it becomes a stronghold of a handful of diehard NX enthusiasts. It usually devolves in page after page of people tearing their hairs out about when the NX is going to be revealed or when some new tidbit of news is going to drop. With nothing but offtopic nonsense in between.

So why not just make a general NX anticipation thread or something where that kind of discussion might actually have a place. Bumping this thread to just continue general NX conversation will just get this one locked too.
 
Since this is the tech thread I wonder if these specs can work in the NX and around 5w.

4 A72's at 1.8Ghz (All for gaming)
4 A53's at 1.2Ghz (two reserved for OS)
6GB LPDDR4 at 4000Mhz (2000Mhz handheld mode)
Tegra GPU 3 SM at 1.2Ghz (550Mhz handheld mode)

Docked mode = 922 GFLOPS
Handheld mode = 461 GFLOPS (around 5-6w)

Should be enough for 720p30 in handheld mode and 1080p30 in the dock. 720p and 1080p use the same assets so it would be cheaper and save more space on carts for Nintendo to go with a 720p screen.

If they go 540p, they'll need two sets of assets for each game.

Obviously the docked mode is the normal clock speeds and the handheld bios downclocks to half speed when out of the dock.

Should be around Xbox One in power but Unreal Engine 4 games with its efficiency with FP16 might actually run better on NX then Xbox One and maybe on par with PS4.

Are you guessing 5-6W for the GPU, the SoC or for the whole device? I would think 3-4W for the GPU might be the upper limit we can expect if we want a decent battery life.
 

Cuburt

Member
Nobody is asking for threads to be closed..

It just so happens that these NX threads get endlessly bumped with zero news and zero on-topic discussion just because it becomes a stronghold of a handful of diehard NX enthusiasts. It usually devolves in page after page of people tearing their hairs out about when the NX is going to be revealed or when some new tidbit of news is going to drop. With nothing but offtopic nonsense in between.

So why not just make a general NX anticipation thread or something where that kind of discussion might actually have a place. Bumping this thread to just continue general NX conversation will just get this one locked too.
Didn't they decide against a catch-all thread like that because the WUST became exactly what you are describing these threads are, but even more off-topic and even more arguing?
 

MuchoMalo

Banned
Nobody is asking for threads to be closed..

It just so happens that these NX threads get endlessly bumped with zero news and zero on-topic discussion just because it becomes a stronghold of a handful of diehard NX enthusiasts. It usually devolves in page after page of people tearing their hairs out about when the NX is going to be revealed or when some new tidbit of news is going to drop. With nothing but offtopic nonsense in between.

So why not just make a general NX anticipation thread or something where that kind of discussion might actually have a place. Bumping this thread to just continue general NX conversation will just get this one locked too.

I really think that a community thread is the best option. Keeps it off of the main page, and keeps it contained.
 

ethomaz

Banned
Sorry guys I made some mistakes in my maths... nVidia claims up to 1.5TFs per Tegra Parker in FP16.

Tegra Parker in DRIVE PX2:

+ GPU: 4x Cortex A57 + 2x Denver 2.0
+ GPU: 256 CUs up to 1460Mhz (Up to 750 GFlops FP32 / 1.5 TFlops FP16)
+ RAM: Up to 8GB LPDDR4 (50+ GB/s)

DRIVE PX2 can (and will) use two Tegra Parker and additional Pascal dGPU if needed.

Of course for NX we can expect only one Tegra Parker.
 

10k

Banned
Are you guessing 5-6W for the GPU, the SoC or for the whole device? I would think 3-4W for the GPU might be the upper limit we can expect if we want a decent battery life.
Whole device being 6w. GPU topping out around 4.

The cpu clocks stay the same but the ram and gpu would drop by half when unplugged to conserve power.
 

Ck1

Banned
Nobody is asking for threads to be closed..

It just so happens that these NX threads get endlessly bumped with zero news and zero on-topic discussion just because it becomes a stronghold of a handful of diehard NX enthusiasts. It usually devolves in page after page of people tearing their hairs out about when the NX is going to be revealed or when some new tidbit of news is going to drop. With nothing but offtopic nonsense in between.

So why not just make a general NX anticipation thread or something where that kind of discussion might actually have a place. Bumping this thread to just continue general NX conversation will just get this one locked too.

Actually what you are stating isn't happening is actually what really happened with the last general NX thread we had. Everyone was pretty much on topic for the most part and we were having some really great discussions about the SCD patent, when someone came in and stated the thread should be locked and shortly thereafter a mod came in and locked the thread...

So having general discussions are apparently very offensive to some around here!
 
Actually what you are stating isn't happening is actually what really happened with the last general NX thread we had. Everyone was pretty much on topic for the most part and we were having some really great discussions about the SCD patent, when someone came in and stated the thread should be locked and shortly thereafter a mod came in and locked the thread...

So having general discussions are apparently very offensive to some around here!

Yep! They were discussing on topic stuff and someone just said the thread should be locked or would be locked. Same with this topic-they're discussing the Tegra stuff and someone is posting about locking the thread...
 
Status
Not open for further replies.
Top Bottom