Nintendo Switch Tech Speculation discussion

Status
Not open for further replies.
I don't think not being a SoC would have been such a big problem, though. I don't doubt that AMD being able to provide a semi custom SoC capable of being >8x more powerful than the earlier generation got a substantial weight in the final decision, for both cost-effectiveness and power consumption. I don't think the first hardware teams from Microsoft and Sony started their design contests back in 2008/2009 saying "we only accept SoCs with 64bit CPUs", as that would make things way too narrow.

It's been a long time since discussion and articles focused on the consoles launched in 2013, but I think Microsoft at least was at least heavily leaning towards an SOC.
Various other factors like not regressing from the existing consoles, keeping tools in line with the rest of the market, the projected base for 64 bits, and the desire to rely on Windows for the platform probably contributed.

There may have been more leeway early on for Sony on the SOC angle, but there were obvious benefits.
Trying to keep in sync with the rest of the market and not regress would have produced similar pressures.
Possibly, just assuming that Microsoft wouldn't drop to 32 bits would probably encourage Sony to not aggravate developers or allow a marketing deficit.
 
? A57 was certainly available for that timeframe. Is A57 just not competitive with Jaguar?

While Cortex-A57 was announced about a year before PS4 and XB1 came out that doesn't mean it was realistic to incorporate it in time for the consoles. This is evidenced by the fact that the first A57 product didn't come out until about a year after PS4 and XB1 were released (Exynos 5433 in Galaxy Note 4).
 
the shield TV has a fan, the switch has none, can we expect superior performances from the switch without active cooling ?

Are you sure the Switch doesn't have one?

There are big holes in the middle of the top of the unit, clearly visible here:

http://static1.gamespot.com/uploads/original/1534/15343359/3145731-nintendoswitch6.jpg

The only other thing I'd expect holes for would be for speakers, but not this large or in this location.

Something looks visible under the holes.. just barely, but I think it could be heatsink fins, or a depressed finer plastic grill. Latter would look a lot like the venting on the Shield portable:

http://www.notebookcheck.net/fileadmin/Notebooks/NVIDIA/Shield/shield_2.jpg

If Switch is using 16nm it should at least be more efficient than X1, although maybe still not delivering the same performance as Shield TV while undocked.
 
Last edited:
I don't recall seeing 8 core A57 designs (full cache coherency + all cores active at same time) until last year. And it's not that fast either (see multithreaded scores): http://wccftech.com/amd-8-core-arm-cpu/

It's substantially ahead of Kabini in those charts and those are at a normalized clock speed, while 20nm A57's would have a substantial clock advantage over the 1.6 Ghz Jaguar's present in the consoles. Seems like more than just competitive to me, based on those numbers.
 
I have assumed Switch does indeed have a cooling fan because of the large vents.

Sent from my SM-G360V using Tapatalk
 
It's substantially ahead of Kabini in those charts and those are at a normalized clock speed, while 20nm A57's would have a substantial clock advantage over the 1.6 Ghz Jaguar's present in the consoles. Seems like more than just competitive to me, based on those numbers.
The A57 had massive amount of cache compared to the kabini, also those micro benchmarks don't really show real world performance. It shows to have better memory performance but the integer and floatpoint performance is almost identical to kabini. 2GH A57s with normal amount of cache probably isn't much faster than the 1.6GHz jaguar cores.
 
The A57 had massive amount of cache compared to the kabini, also those micro benchmarks don't really show real world performance. It shows to have better memory performance but the integer and floatpoint performance is almost identical to kabini. 2GH A57s with normal amount of cache probably isn't much faster than the 1.6GHz jaguar cores.

I don't know about that, here Tegra X1 with the same amount of cache and cores really outshines the Kabini 5150 and sometimes even 5350 (2 Ghz):

http://www.phoronix.com/scan.php?page=article&item=nvidia-tegra-x1&num=3

Almost every bench I've seen comparing Jaguar cores to ARM cores, Jaguar doesn't impress to say the least.

Here it doesn't even manage a clear win against 1.8 Ghz A15: http://www.notebookcheck.net/SoC-Shootout-x86-vs-ARM.99496.0.html

Anyway, the point here was not if A57 is faster, but rather if it was competitive enough to be an alternative and it surely seems to be, and then some. Until I see evidence showing the contrary I have to go with what I can find, and everything points to it being faster.
 
I don't know about that, here Tegra X1 with the same amount of cache and cores really outshines the Kabini 5150 and sometimes even 5350 (2 Ghz):

http://www.phoronix.com/scan.php?page=article&item=nvidia-tegra-x1&num=3

Almost every bench I've seen comparing Jaguar cores to ARM cores, Jaguar doesn't impress to say the least.

Here it doesn't even manage a clear win against 1.8 Ghz A15: http://www.notebookcheck.net/SoC-Shootout-x86-vs-ARM.99496.0.html

Anyway, the point here was not if A57 is faster, but rather if it was competitive enough to be an alternative and it surely seems to be, and then some. Until I see evidence showing the contrary I have to go with what I can find, and everything points to it being faster.
In the link you posted, the X1 wins some and loses some. It doesn't seem consistent at all. And FFTE is exactly the kind of micro benchmark that is terrible at showing real world performance because it does literally 1 small function written in fortran recursively over and over, meaning compiler and specific instruction optimization will easily give a cpu magnitudes of improvement.

I don't think anyone is going to argue that A57 is a lot worse than Jaguar but it wasn't an alternative to the consoles because there wasn't A57s when the xbox one and ps4 launched, especially not ones that contain console level GPUs on the SoC.

Jaguar isn't exactly an up to date arch, being on 28nm and pretty much end of life. If you want to compare, then compare it to excavator at least because even though it is on 28nm still, it actually has been updated.
 
Could be like the old Mac Cube, passively cooled, would be nice in fact, who knows maybe it's the case when handheld and it turns the fan on and overclocks once docked...
 
Anyway, the point here was not if A57 is faster, but rather if it was competitive enough to be an alternative and it surely seems to be, and then some. Until I see evidence showing the contrary I have to go with what I can find, and everything points to it being faster.
A57 is definitely competitive. I don't doubt that at all. The problem is that the first A57 based SOCs were available in Q4 2014 (Qualcomm + Samsung). And these were quad cores. First eight core SOC was released year later. So one has to wonder whether a similar 8 core A57 based SOC could have been released 2.5 years earlier. Consoles launched 2 years earlier, and you'd need final devkits roughly half a year before the launch (at the latest). I would guess that competitive 8 core ARM based SOC was slightly too late (~1 year).
 
Since Switch is doubling as a handheld I don't see the point of including a hdd connector in the base as whatever you put on there won't be available on the go. A SD slot would make a lot more sense. With SD cards as cheap as they are I don't really see that being an issue.
 
you are right, so we could expect shield TV performance without android overlay at the very least.

It could be that low but even then looking at Nintendo's last four consoles (handhelds included) is it really low ?. A 500gflop GPU is a ridiculous amount of power for Nintendo's first party developers to work with (almost 3x WiiU in pure numbers, probably closer to 4 or 5x when you take the fact it's a much more modern Nvidia chip into account vs the 2008 AMD GPU in WiiU). On top of that the console will likely have at least 4x the RAM and a CPU which is a quantum leap over the WiiU CPU which is based on a 15 year old design. Nintendo will produce astounding visuals and performance on the Switch when you look at what they achieved with a 15 year old CPU architecture, 1GB of terribly slow RAM and a 176glfop GPU on WiiU.

Super Mario 3D World, Mario Kart 8, Captain Toad, Smash Bros 4 from Nintendo, Xenoblade Chronicles X from Monolith Soft and Fast Racing Neo from Shin'en are all fantastic looking games on WiiU (and all 60fps outside Xeno X) esp for it's pathetic amount of power on paper.

Personally I think NS is downclocked when in handheld mode and when it's docked will upclock meaning games run at a higher resolution and / or framerate in home console mode.
 
It could be that low but even then looking at Nintendo's last four consoles (handhelds included) is it really low ?
Yes. ;) Measured against the competition in the same price bracket, it's low. Within those limits, devs will make great looking games, no doubt. But with more power they could do more. Taking your Wii U examples, if Wii U had twice the power it could play the same games with far better IQ. Most importantly, it'll be able to play games from other developers that don't limit themselves to simpler art styles, adding considerable value to the platform.
 
Status
Not open for further replies.
Back
Top