brochiller
Member
You must be fun at parties.
Nintendo Switch Dev Kit Stats Leaked? Cortex A57, 4GB RAM, 32GB Storage, Multi-Touch.
- Thread starter Robert at Zeboyd Games
- Start date
- Status
You must be fun at parties.
well, 20nm yes, but we are talking about a 16nm version which would be better it seems
ShadowFox08
Member
That is on fp16 only, which can't be can't be uses most of the time l. Fp32 is 512 GFLOPs. It's going to be closer than on the middle, if the custom chip ends up being x1.
Of course, there are Maxwell chips that are quite a bit more capable than the PS4 and X1, anyway, so even that it has Maxwell shouldn't be doom and gloom. The details do actually matter, as it turns out.
Skittzo0413
Member
16nm Maxwell seems to actually make every rumor we've heard from reputable sources fit together nicely.
I'm probably missing some rumors but this seems to be pretty consistent at satisfying all of those rumors.
- Emily Rogers initially said that the Switch may outperform XB1, though "that might be a stretch" which, at the very least, should indicate that the Switch will be relatively close to XB1 performance. Perhaps she saw FP16 numbers, which would still require higher than 1TFlop to be at XB1 numbers.
- The Eurogamer rumor implied that the TX1s in the devkits were overclocked. This would make no sense if base TX1 performance is their final target (which would be the case with 20nm Maxwell w/ 2SMs). If the final chip is to be on a 16nm process, then it can be clocked higher than a base TX1 regardless of CUDA cores.
- Nate has said Pascal, which, depending on who you ask, can mean Maxwell on a 16nm process. Again, this could be an issue of semantics.
- Matt and OsirisBlack have said that ports from PS4/XB1 should not be much of a problem, tech wise. This suggests that performance should be around XB1 levels at a minimum.
- We have confirmation that there is a fan in the Switch which can run in portable mode, albeit it a low RPM. The presence of a fan suggests that the device will run at a fairly high clock speed.
- Nate has also suggested that the final target for battery life is 5-8 hours. This would be impossible with a 20nm process, unless downclocked close to Wii U levels, and not make sense if a fan is required.
I'm probably missing some rumors but this seems to be pretty consistent at satisfying all of those rumors.
blu
Wants the largest console games publisher to avoid Nintendo's platforms.
Exactly. The one thing we should care about is the fab node, as that will be the single most important tech factor for what nintendo make or don't make out of the Switch, at least in its first iteration.
sir_bumble_bee
Member
Those last two bullet points are kinda confusing, unless I'm misunderstanding something. If it's 16nm Maxwell, wouldn't the fan in portable mode be unnecessary? Unless heat is still a concern even at 16nm?
yes but, correct me if I am wrong: there is no rumor about a 16nm maxwell.
A gaming console can't throttle it's clockspeed when the SoC reaches a certain temperature, like most mobile devices. The fan should allow the Switch to keep a constant power target without sacrificing performance when in heavy load scenarios.
Papacheeks
Banned
If the updated rumor about better battery life on final devkits.hardware are to believed it points to custom 16nm maxwell actually being true.
It makes sense since NVIDIA and couple of leakers said it's a custom tegra chip. If it's custom one of the things Nintendo would specify since this is a mobile type device is smaller Dye shrink to be of the biggest priority's for better power efficiency.
And seeing how there were leaks about devkits having 3 hour battery life, then months later a leak about final hardware/kits having somewhere 5 hours of battery life it makes sense that they had early devkits on probably 20nm as they awaited the newer revision by nvidia.
Which would probably go along with the rumors we heard around E3 about chip issues, the chips with 16nm process were not ready probably at the time.
This all is starting to make a lot more sense. I really do believe that there is some truth to this. I believe the issues around E3 were maybe issues in getting the 16nm ready for production and yielding the power efficiency Nintendo wants.
My understanding of this is that it may depend on clockspeeds even at 16nm. It's a custom chip, so we don't know what they've done (and haven't done), which is why I think we really don't know much at this point.
Buggy Loop
Member
A custom tegra -> a custom maxwell =/ tegra X1
20nm process of X1 is not necessarily imposed on custom designs.
But we simply don't know.
What would be odd is with all the money TSMC has invested in 16nmFF, that they would be willing to lose QTY per wafer for the hell of it by going 20nm in 2017 for a mass produced item.
Thats... pretty important actually, more so than performance.
sir_bumble_bee
Member
Skittzo0413
Member
I'm saying, for both of those features to coexist (a fan and 5-8 hours of battery life) it really can't be a 20nm device. A 20nm device run at clock speeds which require a fan would have a very, very poor battery life in the Switch's form factor, and a 16nm device could possibly run at a high enough clock speed to warrant a fan, while still not using quite enough power to decimate the battery life.
Schnozberry
Member
With semi-custom chips, it really boils down to timing and money, as with everything in life. The most likely scenario in my mind for why Nintendo would keep 20nm is that Nvidia had unfulfilled fab commitments that they would be penalized heavily for, and gave Nintendo a fire sale type deal to use them up.
Nearly all of these points are based on misunderstanding the information given, or have pure misinformation.
would you kindly argument why?
Skittzo0413
Member
Do you care to elaborate?
If you go back to the GAF thread from May 13th, she said it would be "stretching it" for Switch to reach XB1 power. She didn't believe NX would hit XB1 specs.
Skittzo0413
Member
Which is exactly what I said... "stretching it a tiny bit" indicates that it's pretty close, no? I can't really read that as 50% XB1 power...
Hmm maybe I misread your post.
I do agree with you that Maxwell makes sense with all of the rumors out there.
PowerThroughLov
Member
I remember a semiaccurate article around a year back (I cannot find it at the moment - sorry!) that essentially said Nvidia had a TON of 20nm wafers they needed to use, and were willing to dump them off SUPER cheap to anyone.
I always wondered if that's what the switch was. Just a giant dump of a lot of Silicon Nvidia was stuck with on 20nm Maxwell.
As disappointing as that is (not being Pascal), the actual abilities of the chips are the same. Just hotter / more power hungry (which isnt great since its a handheld, but if they invest the savings in screen and/or battery its probably worth it).
I always wondered if that's what the switch was. Just a giant dump of a lot of Silicon Nvidia was stuck with on 20nm Maxwell.
As disappointing as that is (not being Pascal), the actual abilities of the chips are the same. Just hotter / more power hungry (which isnt great since its a handheld, but if they invest the savings in screen and/or battery its probably worth it).
I think the 20nm wafers was the speculation of Thraktor, not from semiaccurate.
Schnozberry
Member
This has always seemed odd to me. Why would they buy so many 20nm wafers? It was known as early as 2012 that they were unhappy with 20nm, and that it was a complete waste for dedicated GPUs, and barely tolerable for SOCs. The Tegra X1 was supposed to be the product that discharged their 20nm commitments.
To be clear, I'm not disputing that they have commitments they are trying to get rid of, just unsure why they purchased so many in the first place, and why it wouldn't be something their shareholders regularly pressured them about.
Do you mean the final dev kits are off the shelf parts or are simply 20nm GPU's?
Emily never said that it's faster than Xbone; she only said that it was closer to Xbone than PS4 and everyone assumed that she was saying that it's between the two.
Eurogamer only assumed an overclock based on finding out that there was a noisy fan. It means nothing, since Jetson dev kits have fans anyway and need them running to run at full speed. Overclocking was nothing more than a possibility, and an unlikely one at that.
Nate has already backed down on Pascal for the most part.
Being able to get ports from Xbone doesn't imply that it needs to be very close to Xbone. There can (and likely will) be a significant difference in performance between Xbone and Switch, with Switch ports simply running at drastically reduced detail settings and resolutions.
Nate said it would be around 3DS battery life, then said 5-8 hours. One of those statements is wrong, obviously. Plus his information is uncorroborated.
You're clearly reading what you want to read here. Just accept Switch for what it is and think about what it can do, not what you wish it could do. It'll be under half as powerful as Xbone on-paper, but also more modern than Xbone.
can you lend me your switch please?
That would mean when it is docked it is stronger than Freiza final form, the PS4.
Skittzo0413
Member
... did you read anything I said? I said that her statement of "even that's a tiny bit of a stretch" implies that it's close to XB1 in performance, not above. Essentially the most conservative reading of her statement is that it is fairly close to XB1 performance. I don't think a maximum of 512GFlops qualifies.
They said that TX1s at full clocks typically run silent, which was corroborated by many people here. An audible fan implies- doesn't 100% confirm- that it is potentially overclocked. I'm pretty sure that's how the article framed it too- that overclocking was a possibility.
Excuse me? I haven't seen him do anything of the sort. He's saying that the final dev kits are Maxwell based but he's currently trying to get clarification from his sources regarding the Pascal rumor from the summer.
While this is true, Matt and OsirisBlack implied that making said ports wouldn't be that big of a technical issue. Drastically reducing settings requires a lot of development time and money, so ideally Nintendo wants to lower all of those barriers as much as possible. You're right that this isn't a clear indication of power level though, so I'll give you that one.
He has stuck by "targeting 5-8 hours" many times, even after being told that the 3DS doesn't get there.
You're clearly ignoring a good chunk of what I've been saying in each of those points. I never said Emily is claiming this will be more powerful than the XB1 (Edit: at one point I believe she implied it was between XB1 and PS4 levels, but she then clarified with the "stretch" quote), just that she implied a PS4-XB1 then saying closer to XB1 "but that's stretching it a tiny bit". I don't think a 512GFlop device qualifies there, regardless of whether or not it's more modern in a lot of ways.
Now, I'm not saying this is the only way to read these rumors. Emily could've seen FP16 numbers for instance. What I'm saying is that this is one way to read each of the rumors such that they all make sense together.
We don't know enough yet about the specs for you to come to the conclusion in your last paragraph. It makes no sense for you dismiss the previous rumors you mentioned based on conjecture. The reveal is less than a month away and we should be able to tell what the relative power levels are of the device then, or get more conclusive information about the specifics of the GPU, CPU, etc.
th3sicknness
Member
I think people need to realize that the best case scenario is usually never what happens. To me having custom Pascal SOC with fast memory bandwidth and a range of 500 gflops - 700 gflops is the best possible situation we could hope for.
I would scale those expectation back big time and that is what we will get.
I would scale those expectation back big time and that is what we will get.
i mean, even in the worst case scenario, maxwell 20nm, the only difference will be heat and battery life, graphically the games gonna be roughlty the same
LordOfChaos
Member
Or the other half of the equation could be tinkered with.
Performance = efficiency x power use
More efficient doesn't only have implications for battery life, it could be used towards performance, battery life, or both.
It's like saying battery life is the only thing to be gained going from a G4 to a Core Solo. Within a fixed power use, efficiency has implications for every aspect.
Dude that's a tag if ever I've seen one.
Malo isn't a "glass half-empty" kind of guy, he's a "the glass fell in a blackhole" kind of guy.
we will have to wait and see.
i mean, i really dont care, i just want more nintendo games, hahaha, but yeah third party support may be hindered by this
Here's the thing, this revelation doesn't change anything compared to previous rumors other than potentially battery life. None if this is really a huge deal. I'm also annoyed by people pretending to know more than they actually do.
did nate kill your dog or something?
No not really. Power draw and form factor are going to be the main limitations of this system, at maxwell 20nm you get less performance for equal power draw / size.
Skittzo0413
Member
This revelation isn't really even a revelation because architecture alone doesn't tell us anything when it's been confirmed that this is a customized chip. The only thing this potentially changes, if we choose to give a little trust to the Maxwell rumor, is Nate's rumor that the final unit will use Pascal architecture. And that could easily just be a misunderstanding of 20nm vs 16nm processes.
There are devices out now with similar internal volume to Switch running Tegra Maxwell on 20nm at 850Mhz (435Gflops) passively cooled, meaning no fan at all. We now know for a fact that a fan is running in the handheld part of Switch. Nintendo have never included a fan in a handheld so they don't do it unless its absolutely necessary. That should tell you something about the performance in handheld mode.
Even with more consistent high loads for a gaming device I think its a certainty we aren't looking at under 800Mhz if its Maxwell 20nm or a fan would simply be pointless under any circumstances.
We also know for certain now that in docked mode resolution is increased. Which will need a significant increase in performance. IMO something around this is the worst case:
Maxwell 20nm:
Handheld - around 400+ Gflops
Docked - Around 600+Gflops
Best case:
Maxwell 16nm:
Handheld - 550+Gflops
Docked - 850+Gflops
Obviously I'm not trying to be exact, just something around there IMO.
Honestly Im expecting something like worst case. I kind of given up hope on 16nm.
Schnozberry
Member
This is presuming they didn't add additional CUDA Cores in comparison to the original Tegra X1.
Especially when both chips (Tegra Maxwell/Tegra Pascal) use the same architecture anyway.
Of course, would likely be better for power consumption that way and with it being custom they could absolutely do that. Its easier to look at a straight comparison using Mhz since you can kind of guess where the limits are easier but I had considered extra cores at 16nm.
Nate wasn't even close to my mind when I typed that.
Okay, so tell me: what do you lose from going with the most likely scenario, that it's 20nm Maxwell? Also, no, there's no chance that the final dev kits and the retail units are using different architectures. That's pure desperation. If the dev kit isn't using final hardware, it's not a final dev kit.
That best case isn't happening, that "worst case" is closer to the actual best case.1.2GHz on 20nm docked is optimistic due to yields; 1.7GHz on 16nmFFis just trying to make sure that you're disappointed.
- Status