Nintendo Switch Dev Kit Stats Leaked? Cortex A57, 4GB RAM, 32GB Storage, Multi-Touch.
- Thread starter Robert at Zeboyd Games
- Start date
- Status
Graphics Horse
Member
Indeed. Nothing that panned out hadn't already been leaked by Emily Rogers or NateDrake.
'The controllers also attach vertically' lol. Only if there was some weird shell that adds a second pair of rails.
Skittzo0413
Member
His original leak seemed to have a lot of either mistranslations or misunderstandings of the difference between FP32 and FP16. We also don't know if his report was a "leak" or his own speculation, though I'd guess the latter.
He did say 4GB of RAM long before anyone else, though as you guys said it's likely just throwing shit at the wall to see what sticks.
ecosse_011172
Junior Member
I just explained my reasoning for bringing to attention a lack of consideration of power consumption in this thread. Earlier, you wrote that "There has been barely any discussion in this thread about CPU clock speeds" which supports the argument I made in the post you just responded to. (You also linked to posts of yours outside this thread. The reason for that still escapes me, since I am responding to the discussion in this thread.) My initial post was a direct response to a specific post citing the performance figures that have been undeniably widely accepted as reasonable in this thread (at least over the last 10-15 pages or so in which I participated), which was the entire basis of my argument, as I just explained.
Instead of engaging that argument I just made, you are instead distracting with a meta discussion about how I am supposedly not engaging other people's arguments. This is strange. Maybe you are reading "everyone" so hyperliterally that you take something personally that was never addressed to you, which would explained why you—quite irrelevantly—quoted posts of yours from another thread. Or maybe you perceive my posts as party pooping, which I am not intending to do. In any case, I don't think you are producing fair criticism here.
Eolz
Member
Ohhh thanks.
Isn't that from the japanese article that was mostly speculation some days back then?
Edit: nvm, just seen it's the basic Parker specs. So not a leak, and not the Switch.
No, I apologise because I was harsh on you.
It's just that after the fact that so many frauds got exposed when the Switch was revealed, I thought that at least people would do more to question things when they are reading claims from a rumour. Especially when that rumour comes from somewhere that isn't a reputable media outlet.
That isn't really the case though since a lot of people still follow those "leakers" on Twitter.
Akhe
Member
Yeah. All this is pure speculation but the dude was spot on about the ps4 pro. Once again, all this is pure speculation.
You're blowing that a bit out of proportion. It's not hard work for a lot of devs because they would of had tons of experience with FP16 on PS3 and mobile. All that knowledge just doesn't go away. And devs are encouraged to use FP16 on PS4 Pro now. Its going to be very relevant (again). It's not hard for devs to test FP16 to see which shaders and algos work or not, they literally only need to change 1 line of code to test. And in my research a good amount of shaders can use FP16 without any work or little work at all and still produce the same result as FP32.
Any game on Switch which is purely FP32 only is a poorly optimised one.
The work to get as many shaders and algos on FP16 is well worth it for those to perform at double the performance.
ecosse_011172
Junior Member
No problem.
I pretty much skimmed, didn't think too much and posted. I should know better having seen how people like poor old 10K got the piss ripped out of them by internet pretend insider twats.
Let's look over the facts. We have a known Gaffer/insider with a track record dating back more than a decade, who claimed the Switch CPU was weak... only marginally more powerful than the PS4 and XBO. Yes, MORE powerful than both the CPU's inside those consoles. Considering CPU can't as easily be downclocked in portable mode, this is rather relevant.
Secondly, we have a very detailed leak by Eurogamer that so far seems 100% accurate. They said the GPU inside the Switch (NX) is the GPU from the Tegra X1 which was running hot in the devkit, which was being actively cooled. If this chip is running hot while being actively cooled, we can assume the chip was close to running at full speed, full speed or maybe even overclocked. Let's assume it was running at full speed, which would mean 512GF at fp32.
Then we have rumors and speculation that the chip in the final retail hardware will be Pascal based. Pascal is basically identical to Maxwell, with some tweaks and on a smaller fab node. This shrink brings us to 40% more performant at the same power draw, or 60% more power efficient at the same performance. This comes from a document on the website from the factory that makes the chips. SInce there is little to no reason to downclock the chip when docked, and the performance in docked mode would ideally be better than when portable, it makes sense to make use of those 40% performance increase when docked, and the 60% more power efficient when portable. In this case you would get +/-750 GF at fp32 when docked.
So, this is where it all comes from. Maybe overly optimistic? Sure, maybe, but not just taken out of thin air.
You're right, my response to you certainly wasn't adding anything of value to the thread, so I'll leave it at that and try to return to more constructive comments.
Fourth Storm
Member
You know what? I'm sticking with Zlatan over on Anandtech on the bandwidth. He says it's about 1/3 Xbox One (with Tegra X1 being 1/8 Xbox One). If we add Xbone's minimum 109 GB/s eSRAM + 68 GB/s DDR3 to get 177 GB/s (simplifying, I know, but he seems to support this in his comments), then the Switch bandwidth could be about 60 GB/s . That's the throughput of 128-bit lpDDR4 3733. This spec makes more sense to me than Nintendo skimping on RAM bandwidth, of all things.
I agree that my post was a little exaggerated lol. I could have expanded on my point more. My post was aimed mostly at those who expect a 512/768 gflop device to suddenly act like 1/1.4 tflop device. Fp16 has been a thing on Nvidia hardware since dx9 (many of the benifits were present) but for some reason a few ppl are treating it like the second coming all of the sudden. I also agree with your last statement and would like to add that any game which only uses fp32 is a poorly optimized one lol.
My UE4 example was mostly referencing fp32 stuff that would not traslate to fp16 well(obvious artifacts) w/o some major changes. Stuff like that is why I personally believe we wont see NX suddenly have double performance.
Yes, his comments make sense from the point of view of someone who might have access to info about the Switch hardware, but simply misunderstands the difference in bandwidth efficiency between Pascal and GCN 1.0/1.1 era AMD. I wouldn't quite expect 3733 MT/s LPDDR4, though, just because I haven't actually seen it in any shipping products (despite it being available for a while now, afaik), which would indicate it might be a bit too expensive or power hungry to be viable. A standard 3200 MT/s rate would seem more likely to me.
It should outperform Wii U without too much difficulty in handheld mode, perhaps by a factor of two in terms of raw GPU computational performance, plus whatever advantages come from using a substantially more modern GPU arch and more capable CPU.
As far as I understand it there's absolutely zero chance it'll be worse than Wii U in portable mode
Lord Ghirahim
Member
It's ZI, a site that should have been blacklisted here years ago. I guess people don't spam it enough for the mods to notice.
It could be slightly more power efficient depending on the density of the transistors.
https://www.computerbase.de/2016-08/nvidia-tegra-parker-denver-2-arm-pascal-16-nm/
Since it was confirmed that Parker which is Pascal-based could perform at 1.5 TFLOPS (FP16)
That means they got more performance out of designing a new chip compared to just doing a die shrink of a TX1.
Hence, it's possible that the Parker's GPU is slightly more power efficient at 1Ghz compared to a die shrunk (16nm) Tegra X1 performing at 1GHz.
All speculation of course until the wattage can be tested between TX1 and Parker GPUs.
From Image and Form games Twitter " Can't comment on specs, I'm afraid. But Nintendo are definitely not skimping on power! "
https://twitter.com/ImageForm/status/793482754767945728
https://twitter.com/ImageForm/status/793482754767945728
Hah of course, next to impossible to go full FP16 with everything too. Its why on PS3 you had to use the SPE instead of the GPU for FP32 shaders, or suffer from poorer performance trying to do them on the GPU (like you said, Nvidia dx9 era
could only do FP16 well ). But the ratio of FP16 to FP32 for a game aiming for post processing effects and such galore should aim at least for a 50/50 balance for me to consider it optimised 
this would mean 25% extra performance over 500GF FP32 if half the shaders were FP16.Well that's nice to hear. Most likely it means that it performs very well as a portable device and fairly well as a home console, but I am sure some will soon be assuming it means 1.5 TFLOPS in fp32 when docked.
ecosse_011172
Junior Member
Not exactly a developer who uses any power though, are they?
Yeah I am waiting to hear more. The Toki Tori devs said that the WiiU was very powerful too. I guess it deoends how much power your game actually needs.
They would still be familiar with the capabilities of the hardware potentially lol. It's not like devs who target more modest and manageable goals simply do so because they are ignorant of any other options.
Well I'm sure they use some power but yeah maybe not the best to judge the power but it sounds like this time Nintendo actually gave official specs out so they may know them afterall. In fact supposedly they are gonna give the specs to the public come jan.
ecosse_011172
Junior Member
"Not skimping on power" is such a subjective term though, could mean relative to the Wii U or for a portable device etc.
Sheesh. Does that even matter? They have dev kits and are making a game for the system. They would know more than any of us.
Certainly not but that doesn't mean they don't have documents and the dev kits itself to know how powerful it is.
edit: as you say, tho.. "Not skimping on power" doesn't tell us anything.
I imagine for them to say anything like that in the first place means it's not a repeat of the Wii and Wii U staying a gen behind in terms of power? As that is the only times they skimped on power.
BronzeWolf
Banned
So this is my current bet for the Nintendo Switch:
Pascal 16 nm SoC
CPU Complex
4x A57's + 2x gimped Denvers @ 900 MHz
L2 2+2 MB
256 Cuda cores
RAM 3.2 games + 0.8 OS GB
Average RAM Bandwidth 60 GB/s
32 GB eMMC Storage
6200 mAh
But I am no computer engineer. Maybe I am betting something idiotic.
Pascal 16 nm SoC
CPU Complex
4x A57's + 2x gimped Denvers @ 900 MHz
L2 2+2 MB
256 Cuda cores
RAM 3.2 games + 0.8 OS GB
Average RAM Bandwidth 60 GB/s
32 GB eMMC Storage
6200 mAh
But I am no computer engineer. Maybe I am betting something idiotic.
Cheers everyone. That's reassuring.
I was a bit worried by the low framerate of the Zelda footage in the trailer. I know the footage was all added in post but...
If the machine 2x Wii U in handheld mode then I'll be happy. I'm hoping docked mode is merely the difference between 720p and 1080p with no difference in framerate. I'd hate for going mobile to mean a lower framerate.
I was a bit worried by the low framerate of the Zelda footage in the trailer. I know the footage was all added in post but...
If the machine 2x Wii U in handheld mode then I'll be happy. I'm hoping docked mode is merely the difference between 720p and 1080p with no difference in framerate. I'd hate for going mobile to mean a lower framerate.
LordOfChaos
Member
In which dozens or optimistically a few hundred megabytes a second of NAND cartridge storage replaces RAM which operates in the dozens of gigabytes per second range.
He guessed the number right, but I'm not sure what the timeline of other details like the custom Nvidia chip was compared to his thread. But since he seems to not really know much about the innards I wouldn't give it too much thought. Even the 3GB/s reading SSD in the new MacBook Pro isn't appeasing some creative professionals who wanted 32GB RAM, let alone what a SD-card like storage cart would be.
WhiteWaterM00se
Member
Ok, I know you all have crunched numbers based on theory. First, question is can this system run Xbox 360 ports better than the 360 could? Nvidia Shield doesn't run their games as well as a 360.
Second, Can this Switch REALLY run downgraded ports of Xbone games reasonably well?
Second, Can this Switch REALLY run downgraded ports of Xbone games reasonably well?
I think someone analyzed the BOTW footage and found that the one clip with the giant Moblin was a constant 15 frames per second, which just implies that the video was slowed down 50% in post, as opposed to frame drops which fluctuate. Dumb move on the marketing's part, but I would take the performance in the trailer with a tremendous grain of salt.
The thing is, I don't think we've seen any games actually running on the system yet. I'd say it's hard to gauge how well things run until we know more and actually see some games running on the system. In any case, I wouldn't be too worried about these sort of things, because I think the system will still be able to play games quite well.
Keep in mind that Image & Form have used the 3DS as their target platform for their last few games, and ported from there, so even on the lowest end of the possible spectrum for Switch it would be literally orders of magnitude more powerful than what they're used to.
Oh I agree with that very much. It means nothing in and of itself. Just nice to hear a dev being positive about it, even if just in a subjective and vague sense.
BronzeWolf
Banned
Nobody will want to talk shit just yet. They don't know if this is the next Wii and won't want to get caught fucking the chicken before it hatches. No way this is 1 TF even docked
Certainly, but regardless, it's the first Nintendo platform in more than a decade that's powerful and modern enough to run pretty much every single important piece of middleware out there, And that's not subjective, even if it doesn't really tell us all that much about raw performance. That said, I'd argue it's also more important than raw performance.
It's using a SoC, a simple teardown won't tell us all that much. It might even use custom memory chips (like 3DS) or a stacked design and we won't be able to figure out the memory configuration on our own. Last time, we needed Chipworks to figure out what makes the thing tick, and a former GAF member had to ask them directly to give us access to their die shots.
Grampa Simpson
Banned
I'd love this personally, but I don't think we'll see it.
I concur. I don't believe there will be Denver cores either. I think we'll have 4 (or maybe 6) Cortex-A57(or 72 or 73) cores.
Yeah, I remember that.
At least there could be some things that could be identified from a teardown, like the RAM was identified on the Wii U with no trouble wasn't it? Because the RAM was labeled with the brand and model.
Edit: Although, as you said. Anything could happen since it's an SoC.
I think they are going to give us the actual specs this time.
And question... Why do you guys believe there will be no denver cores? Those are proprietary Nvidia ARM cores. It seems like they would want to use those as well.
why would they release the system's specs to the general public? What benefit do they get out of that?
- Status
Similar threads
- Featured