Asus Transformer Drops Quad-core In Favor of Dual-core

MrSeb writes with this news from Extreme Tech: "In a move that will shock and disgust bleeding-edge technophiles everywhere, Asus has announced at Mobile World Congress 2012 that its new Transformer Pads — the high-end Infinity Series — will use the recently-announced dual-core Qualcomm S4 SoC. The critically acclaimed Transformer Prime, the Infinity Series' predecessor which was released at the end of 2011, used the quad-core Nvidia Tegra 3. Why the sudden about-face? Well, the fact that quad-core processors don't really have a use case in mobile devices is one reason — but it doesn't hurt that the Krait cores in the S4 are significantly faster than the four Cortex-A9 cores in the Tegra 3, too. The S4 is also the first 28nm SoC, while Tegra 3 is still on 40nm, which means a smaller and cheaper package, and lower power consumption to boot. The S4 is also the first SoC with built-in LTE, which was probably a rather nice sweetener for Asus." The Snapdragon S4 "Krait" CPU is still a bit shrouded in mystery as far as hard specs (Qualcomm has never been one to release docs), but it appears to be similar to the Cortex-A15 in performance; how they stand up to Intel's new Medfield designs remains to be seen.

Core count obsession

[Cryacin]

Why is everyone obsessed at the number of cores? The more processors you ahve, the more complex scheduling your apps needs to perform to actually work faster. It's better to hav ea single core that is twice as fast, than two cores running in parallel.

Re:Core count obsession

[Psiren]

So seriously, most of the time, the number of cores doesn't even matter, because unless you're playing a high-end game, the cores won't even be woken up.

So, unless I was buying a tablet specifically to play high end games on, why would I want to spend money on CPU cores that are going to sit there doing nothing? Surely a dual core CPU is a better move?

Re:Core count obsession

[mikael_j]

More cores means better multitasking since threads can run in parallel. Also, even for handheld devices you are unlikely to find, for example, a single-core CPU that is four times faster than each core of a quad-core CPU.

Another major advantage of multi-core systems is if a poorly written piece of software locks up it is highly likely to also be single-threaded and your system will chug along nicely despite the misbehaving program, allowing you to kill the process (by comparison, on a single-core system you're likely to suffer through five minutes of waiting for the system to respond before you are able to kill the process). Sure, in an ideal world this wouldn't happen but when it does happen it's nice to not be locked out of your system because of a single process misbehaving.

Summary wrong again?

[Anonymous Coward]

From what I articles I saw yesterday I gathered that there would be two levels of the new Asus pads. One with the Tegra and the other with the new Krait. Here is one article that talks about it: http://www.anandtech.com/show/5586/the-asus-transformer-pad-infinity-1920-x-1200-display-krait-optional

Of course we won't know anything for sure until Asus releases the product details.

Only the LTE model

[jgfenix]

The wifi model will still use the Tegra 3. The LTE model will use the Snapdragon http://www.theverge.com/2012/2/27/2827588/asus-transformer-pad-infinity-series-and-transformer-pad-300-series [theverge.com]

Re:First HTC drops its quad core chip for a dual c

[Anonymous Coward]

The problem is Tegra does not support LTE - this is a problem for sales in the USA.
Europe will continue to get the full spec, full speed, Nvidia Tegra3 devices

Misinformed

[TheBogBrushZone]

They announced two Infinity models. Once of them has LTE/3G and the dual-core S4, the other is Wi-fi only and is still toting a 1.6GHz Tegra 3.

Re:Tegra is a flop.

[Fnkmaster]

Ummm Tegra 2 was the fastest platform for Android for quite some time. The G Tablets are still pretty blazingly fast. The issue is just that Tegra 2 was released for such a short time before Tegra 3 came out that it never got much saturation, and then Tegra 3 came out with a bunch of faster options close on its heels.

NVidia has great hardware engineers, but awful software driver people on their mobile platform. They have done a terrible job supporting their chipsets after release with Android, or getting good manufacturers to adopt them.

Re:Skype on standby

[bemymonkey]

Android has some sort of a built-in low power push-like mechanism that was implemented starting with Android 2.2 (Froyo), called C2DM. It's not quite real push, but the battery life is stupid good.

I'd assume Skype uses C2DM, as do most IM apps...

http://code.google.com/android/c2dm/ [google.com]

Re:In the market for a 4 core Android tablet

[errandum]

I'd say that the new Samsung phone with the projector has 6GB of RAM, so I'd go for it. It's not a tablet, but since you can use it as a 50" screen it should be enough for your needs?

And you have to understand, you're the 0.01%. Most people with your requirements will get a computer (any i7 notebook should fill your needs and give you 10 times the power of any tablet).

What's your specific problem? Maybe something else is more suited.

Re:Core count obsession

[Bill_the_Engineer]

More cores means better multitasking since threads can run in parallel. Also, even for handheld devices you are unlikely to find, for example, a single-core CPU that is four times faster than each core of a quad-core CPU.

According to Amdahl you're looking at 1 / [( 1 - P ) + (P/N)] where N = number of processors and P is the percentage of a program that could run parallel. So if 75% of a program can be made to run in parallel on a quad-core processor we are looking at 1/[(1-0.75)+(0.75/4)] = 2.29, so we are looking at a maximum speed increase of 2.29 times the speed of a single processor not 4 times.

Another major advantage of multi-core systems is if a poorly written piece of software locks up it is highly likely to also be single-threaded and your system will chug along nicely despite the misbehaving program, allowing you to kill the process (by comparison, on a single-core system you're likely to suffer through five minutes of waiting for the system to respond before you are able to kill the process).

I haven't seen this. In fairly modern operating systems you'll have multiple services in operation. This means you'll most likely have more threads in execution than there are cores. Context switches between threads within a core of a multi-core processor will still need to be made. I've had a misbehaving program slow down an embedded multi-core processor board because we were "unlucky" that the OS scheduler was running on one of the same cores and other resources on the processor board were being committed by the errant process (eg. Memory, I/O ports, etc.) so the system is not as foolproof as you'd like since memory takes time to to swap, and deadlocks across cores can happen when a computing resource is shared.

Sure, in an ideal world this wouldn't happen but when it does happen it's nice to not be locked out of your system because of a single process misbehaving.

It really is a speed versus power issue. In an embedded environment, where one would hope that the system was well tested prior to being released to the public, such a safety net is really not required.

Re:maybe the article is flat out wrong

[ChronoReverse]

That's not necessarily true. The Adreno 225 in the current Krait is slower than the 320 that'll be coming later in the year but it still trades benchmarks with Tegra3 in graphics tests.


Notably, the GLbenchmark offscreen test where the Tegra3 wins the most by, has a severe bug somewhere.