Apple iPad: It might be worse than you think

Apple introduced its greatly anticipated tablet computer, the iPad, yesterday.  It was impossible for the product to meet the unbelievable hype surrounding the computer’s debut, but even taking this into account, the reaction to the device, which will not ship for months, has been mostly negative.  Critics point out the iPad’s lack of multitasking, the low resolution XGA 4:3 screen, the clunky aesthetics (very unusual for Jobs-ian Apple), the lack of standard ports, no support for Adobe Flash, the high price compared with more capable netbooks and the awkward “iPad” moniker.

But few if any observers have mentioned that the iPad is probably very underpowered compared with much cheaper netbooks.

The Apple iPad uses a custom, 1GHz Apple A4 system-on-chip (SoC).  While details are lacking, the microprocessor is probably either the ARM Cortex-A8 (a superscalar design a little like the Intel Atom, but sans HyperThreading) or Cortex-A9 (an out-of-order superscalar design superficially analogous to the VIA Nano).  If that is the case, this explains why Apple chose to forgo multitasking because ARM chips are very slow.

ARM has been very successful at avoiding direct, objective performance comparisons of their chips with x86 counterparts.  However, I’ve been able to test an 800MHz ARM Cortex-A8 running Ubuntu Linux and compare its results to an 800MHz AMD Mobile Athlon and a VIA Nano L3050 downclocked to 800MHz.  The x86 systems ran the same version of Ubuntu Linux as the ARM box.

While the Nano and the Athlon are close to performance parity with each other, the ARM Cortex-A8 is less than one-half as fast as either x86 chip.  Moreover, the ARM CPU is much weaker on floating point calculations, providing lower that 25% of the performance of either x86 chip.

Worse, the ARM system frequently becomes unresponsive for several seconds at a time.  And even though I only ran the ARM system at 16-bit XGA resolution while the x86 systems ran at 24-bit 1080p resolutions, both x86 systems trounced the video performance of the ARM box.

After using the ARM Cortex-A8 Ubuntu system, it is safe to say that people migrating from x86 netbooks (which typically use 1.3-1.6GHz x86 processors) — or, heaven forbid, thin-and-light notebooks — will be very disappointed with the performance compromises they will encounter from stepping down to ARM.

So if Apple is deploying a 1GHz ARM Cortex-A8 in the iPad, overall performance will be worse than a 500MHz x86 chip — sometimes much worse.  And that level of performance won’t make many people happy.  For Apple’s sake, if the company is indeed using an ARM design in its iPad, let’s hope that they at least chose a multi-core Cortex-A9.  BSN* confidently claims that this is indeed the case.

I’m trying to get my hands on an ARM Cortex-A9 system.  The Cortex-A9 will boost performance over the A8 because it adds an Out-of-Order engine which reduces pipeline stalls.  However, I expect a 1GHz Cortex-A9 to be no faster than a 600-650MHz x86 counterpart at best.

These results don’t surprise me because several years ago when I was working at Centaur I compared the performance of an Intel XScale chip against a VIA C7.  The C7 creamed the XScale even though the ARM chip ran at a slightly higher clock speed.

I’m convinced Intel sold off XScale to Marvell because the chipmaker recognized that this performance deficit would be untenable once ARM chips inevitably began competing with x86 designs.  Of course, Intel’s response was to create the highly successful Atom.

I’m trying to get permission to publish the ARM Cortex-A8 results.  Check back in a few days if you are interested.

ARM CPUs certainly have their strengths, but raw performance is not one of them.  They will face very stiff competition as they go head-to-head with much more powerful x86 designs.