Hardware Specifications of Nintendo Switch Reveal

All phones and tablets have severe throttling under sustained load. A console must stay at the same clock in any environment, up to the rated conditions of high altitude and 35C. This is completely impossible without a fan. If the limit of comfort for the case is 37C, you have effectively no cooling in a 35C ambient.

Fan + heatsink = headroom and active control.

Passive dissipation through casing = unpredictable shit leading to desparate measures like throttling.

Edit: aaah! beaten by lalaland
 
Last edited:
I don't think throttling is that unavoidable in a tablet sized device. I'm not sure if the Shield Tablet throttles. Tegra Note 7 plays games fine too. These have pretty awful battery life however. Nintendo needs to get more than 2 hours I think. :)
 
Last edited:
Phones have a workload that spikes from low teens to 80-90% while web rendering or some other task. Their CPUs are designed to 'hurry up and wait' I.e. they aggressively clock up to complete a task so they can aggressively clock down again. This results in lower overall power usage for a CPU with low base clocks and a high peak clock versus a CPU with no throttling. Mobile cores cannot sustain their peak or even nominal frequencies very long before having to throttle cores individually or as a whole.

Console CPUs are running at max frequency more or less constantly and the heat load is continuous. With a peaky workload a passive heatsink can absorb more energy than it can dissipate for a short time before having to throttle and if your workload can be completed before the heatsink saturates all well and good. A heatsink and fan ensures heat is evacuated at a rate equal to that at which it is absorbed so thermal throttling should never occur at all.

The fan may or may not be required in handheld mode (although my reading of the dock patent suggests it is on) but would certainly be required in dock mode.

TLDR phone SoCs are a designed for a different operating paradigm and so are a poor point if comparison for a handheld CPU. Equally they generate and dissipate heat in a similarly different way and thus require different cooling solutions.

Yep. Yep yep yep.

I've been trying to make a similar point - unsuccessfully - in the other forum.

So basically: like + yep.
 
Except that Tegra was designed as a tablet/phone chip. It failed because it was too hot and/or power hungry but as far as I know it was never designed as a console/handheld soc.

Also the argument that phones/tablets require thermal throttling with big performance drops is simply not true. It was for Snapdragon 810 devices but for example the Oneplus 3 which houses a Snapdragon 820 hardly has any thermal throttling while gaming.

https://www.xda-developers.com/onep...nd-thermals-analysis-redeeming-the-oneplus-2/

Unless X1 is ridiculously hot I don't see why Switch would need to be twice as thick as a phone/tablet AND require active cooling to stay cool at lowered clock speeds that would probably put performance below that of other high end SoCs. But than again X1 seems to do fine in the Google Pixel (no idea if it throttles).

Edit: From a quick google it appears Pixel C isn't really throttling.

On the Pixel C, the game runs smoothly, even after an extended period where GPU throttling would usually kick in.
http://www.techspot.com/review/1117-google-pixel-c-tablet/page3.html
 
Last edited:
Always pay attention to the bullshit reviews claiming no throttling without any mention of the nice 22C ambient temp of corporate offices.
 
Patent doesn't consider turning off the fan.
But even if we were to look at this from the opposite point, why would Nintendo or nvidia ever consider turning on the fan for a 2 SM GPU @ 300MHz + 4*A57@ 1GHz if the Pixel C with 2SM GPU @ 850MHz and 4*A57 @ 1.9GHz hardly ever throttles?

Pixel C is a 10" screen tablet, this is.. what, 6"? And how many stress tests have people ran on Pixel C, where they played some game that has a heavy CPU and GPU load for hours at a time? I doubt it actually maintains full clock for all four CPU cores and the GPU core at the same time, at least not indefinitely. Most tests out there are CPU intensive or GPU intensive but not really both. Even tests with actual games are going to generally fall short because they're hitting a relatively low CPU margin. Android games may be able to scale with GPU power to some extent but are most likely hitting pretty low CPU baseline requirements to actually be playable by a large audience. If it exists anywhere it certainly wouldn't be in a 2013 title that is itself a much older port like GTA: San Andreas.

Gaming consoles on the other hand are designed with single static CPU resource availability, at least as some max persistent target available for games. They're not designed for delivering big but short bursts in peak CPU load.

Looking back on the original Shield portable, that probably had similar or more room for passive cooling, ran at 1.4GHz max with all four (weaker Cortex-A15) cores active, had a vastly weaker GPU, wasn't on a much more power efficient process, and had a fan. That seems like the closer comparison point to the Switch.

Is there any point in going with the X1 over something like a snapdragon 820 if they are going to gimp it like that? I'm sure Nintendo likes Nvidia's software/tools but it seems they are making a mess of the hardware again.

nVidia's contribution outside of hardware really can't be understated. They're rounding up third party PC-focused developers in ways that are way beyond a company like Qualcomm and they have a more feature rich GPU with better drivers. This is a pretty big deal.
 
Without making GFLOPS speculation, the Antutu benchmark score for Tegra X1 is 75000, iPhone 6 is at 80000 and 5s is at 40000. So in the portable mode Switch is comparable to iPhone 5s which is reasonable for a $199/$249 tablet. Engines like Unity should have a profile for mobile devices where lights are baked, if the game design allows such optimizations porting from PS360 games will be possible.
 
Without making GFLOPS speculation, the Antutu benchmark score for Tegra X1 is 75000, iPhone 6 is at 80000 and 5s is at 40000. So in the portable mode Switch is comparable to iPhone 5s which is reasonable for a $199/$249 tablet. Engines like Unity should have a profile for mobile devices where lights are baked, if the game design allows such optimizations porting from PS360 games will be possible.

AnTuTu is honestly garbage, I wouldn't read anything whatsoever into this.
 
Always pay attention to the bullshit reviews claiming no throttling without any mention of the nice 22C ambient temp of corporate offices.

Maybe that is because every review since like... uhhh I don't know... forever.... always are under normal temperatures. When was the last time you've seen a review where hardware was tested under a 35 degrees ambient temperature?

So that review isn't bullshit at all as that same review shows snapdragon 810 devices, which we know are running hot, throttling a lot. Even within snapdragon 820 devices the level of throttling is different so the internal design definitely matters as well. With the Switch being so much thicker and a lot bigger than a 5.5 inch phone I don't see how such a case couldn't keep something like a snapdragon 820 cool even when ambient temperatures are high.

nVidia's contribution outside of hardware really can't be understated. They're rounding up third party PC-focused developers in ways that are way beyond a company like Qualcomm and they have a more feature rich GPU with better drivers. This is a pretty big deal.

I'm sure it is but what's the point if you end up with vastly underpowered hardware? PC is either performance focused so those devs won't care about Switch or indies but why would indies bother with this? If the X1 inside Switch really gets gimped that much, will it even be able to compete with Intel IGP's?
 
If I had to guess I'd say it's because X1 is mature which makes it easier to customise, plan for production and the toolchain is fully developed for the platform. I going to go out on a limb and say X1 is cheaper for two principle reasons firstly X2 is on 16nm FF and it seems TSMC is charging top dollar for fab capacity and secondly as an older design with very few wins to it's name Nvidia are likely to have cut Nintendo a very good deal on X1.

Nvidia has no actual design wins for their X2 design yet so I'm not sure why everyone is so damn convinced it works just peachy. X1 has 8 cpus (4 x A57, 4 x A53) cores but no shipping product has ever shipped with the A53 cores turned on, we don't have an official explanation but it seems that the design may lack cache coherence between the two clusters which makes seamlessly swapping from one set of cores to the other all but impossible.

Nvidia makes GPUs very well, their CPU efforts have been doubtful at best thus far so trusting them with an entirely bespoke design (as the Denver CPU cores are supposed to be) would be pure madness IMO. Recall their claims they could make an x86 chip at one point?
 
Also rememberthat Nintendo Shield was delayed by at least 6 months, perhaps more, so keep that in mind when looking at hardware that is yet to come out.
 
Maybe that is because every review since like... uhhh I don't know... forever.... always are under normal temperatures. When was the last time you've seen a review where hardware was tested under a 35 degrees ambient temperature?

So that review isn't bullshit at all as that same review shows snapdragon 810 devices, which we know are running hot, throttling a lot. Even within snapdragon 820 devices the level of throttling is different so the internal design definitely matters as well.
The "not throttling" claim is bogus.

It makes a big difference with fanless devices which must keep the case temperature comfortable to hold (like a game console, for example). It becomes extremely sensitive to ambient temperature. Even with a rather uncomfortable 40C case limit, a large tablet with enough area for a 20W consumption at 20C, would have only 5W allowed at 35C.
With the Switch being so much thicker and a lot bigger than a 5.5 inch phone I don't see how such a case couldn't keep something like a snapdragon 820 cool even when ambient temperatures are high.
It absolutely can, because it has a fan and a heatsink. This significantly reduces the impact of ambient temperature.
 
Throttling in tablets is an interesting topic. Awhile ago I had a cheapo Onda V919 9.7" which ran an Atom Z3735. It throttled in games rather extremely. It would even power off eventually. Throttling is often obvious with Intel tablets because these on-screen thermal event notifications pop-up. I took the tablet apart and found the CPU had no cooling beyond a small RF cage. I put a thermal pad in there, which interfaced the CPU to the large rear metal casing and that was a huge improvement. It could actually be used for games. It certainly got toasty though.

I also had a Lenovo Tab Pro 3 for awhile, which is Atom X5. It throttled similarly but I didn't feel like trying to open its strange casing and just sold it off instead. I tried KOTOR on it and performance eventually dropped off a cliff with that tablet.

Now the Shield Tablet, which I use in an enclosed TPU folio case, doesn't seem to throttle. It's not noticeable. I've played some games long term on there, with even it plugged into a charger. I bought their silly gamepad and did some TV HDMI gaming with it. It draws over 10W, more than the usual cheapo charger can feed, and so people frequently post about it (and are told they need a 2A charger). I've been impressed with NV's work on sustained performance with that tablet. Tegra Note 7 is solid too. Is there a good program to detect throttling?

The Surface Pro 4 is another story too. I don't think the world is aware of how inadequate its cooling is if you try to play games on it. Even with its fan. Having a small fan blowing at the back all but eliminates its problems, but then it consumes more power than the standard charger can output. :)
 
Last edited:
I'm sure it is but what's the point if you end up with vastly underpowered hardware? PC is either performance focused so those devs won't care about Switch or indies but why would indies bother with this? If the X1 inside Switch really gets gimped that much, will it even be able to compete with Intel IGP's?

Vastly underpowered compared to what? Snapdragon 820? Because I don't think that was ever realistically in the running. Knowing Nintendo they probably froze the SoC choice in 2015, hence why we appear to be looking at something based on Tegra X1 and not a newer 16nm+/FinFet SoC from nVidia. Nintendo tends to be more conservative about this than they need to be, but all things considered there's really only so aggressive any gaming hardware maker can be while still delivering a decent launch lineup with properly optimized titles.

The realistic comparison would more likely be with something like Snapdragon 810, and no, I don't think that would be a clear winner hardware-wise. Not when you're stuck with the same baseline which is the ability to run all of the CPU cores at the same speed all the time without any regard to potential burst performance. You'd likely end up with something pretty similar CPU-wise with somewhat worse perf/W, lower peak GPU power and much worse perf/W, and none of nVidia's software or institutional benefits.

You guys are looking at these CPU clocks and are thinking wow, they could have had something so much faster with X, Y, or Z. But you have to consider that this isn't about the fastest they could have pushed CPU performance but about a tradeoff that has been made with just about every gaming platform (and especially handheld) that has existed: less CPU power in favor for more graphics power.

Look at XB1 and PS4 - did they clock the Jaguar cores as high as they could have? Based on other products probably not, but they backed them down in favor of freeing up die area and TDP for the GPU, while letting a large core count run at peak clocks consistently and reliably. Look at PSVita and 3DS and you'll see even starker examples where they're using stock CPU cores that are clocked far below what other mobile devices clocked them at. This is all for a good reason and it's not just because they could have and should have chosen Qualcomm instead.

Now ~300MHz (in handhehld mode) for the Maxwell GPU in Tegra X1 might not sound like an awful lot. And it isn't really, but given its form factor, probably a less than substantial battery capacity and a desire to actually be able to play games reliably for at least a few hours it's actually pretty aggressive. And given the CPU/GPU balance we see on other gaming platforms 1GHz CPU is pretty reasonable alongside it.

Is there any better place to get ballpark performance figures for mobile devices?

For CPU Geekbench 4 is one of the best options currently available. And even there I'd advise to look at ST int, MT int, ST FP, MT FP, memory and total scores separately and put some consideration into the particular implications of each for the platforms at hand.

There's a lot of 3D GPU benches that are at least not totally useless, like GFXBench and 3DMark.

AnTuTu's individual tests are mostly all awful, but more than that the total number aggregates a lot of things that are going to involve system performance details that may not be that relevant to gaming handheld performance (like SD write speed)
 
Do we even know for sure that the Shield Console doesn't or cant throttle? Has there been stress test on the little console while gaming for many hours straight with a game like Doom BFG to really see just how hot that little console can get?
 
I don't think throttling is that unavoidable in a tablet sized device. I'm not sure if the Shield Tablet throttles. Tegra Note 7 plays games fine too. These have pretty awful battery life however. Nintendo needs to get more than 2 hours I think. :)

Yeah, according to Toms the Shield Tablet doesn't throttle (at least in 30 minutes of GFXBench 3.0) but you pay the price in terms of battery life (2.3 hours).
http://www.tomshardware.com/reviews/nvidia-shield-tablet-controller,3949-16.html
 
Back
Top