The ARM Cortex-A9 Can Beat Out The Intel Atom

[chuckula]

Anyways back to the tests, anyone else find it weird that Atom under 64 bit overall did worse? Any particular reason for that?

I could see issues with x86_64 dues to things like the pressure on the instruction cache and possibly the data cache if there are a lot of memory pointers. There are architectural advantages to x86_64 in higher-end Intel parts (think Sandy Bridge & Ivy Bridge) but in an Atom the simple in-order instructions and less sophisticated front-end may negate many of these advantages.

[chuckula]

Just noticed that the Atom is being run with the new x32 mixed instruction set. This takes advantage of the 64 bit improvements in the architecture and enables 64 bit registers for doing math (faster in some cases) but limits the memory addressing to 32 bits (less cache pressure). Especially in a system like Atom I could see that giving a performance bump in many cases.

[gururise]

Clock speed is useless, too.
Compare something relevant, like power consumption or price.

I disagree. Comparing the ARM A8, A9 and Atom N450 all at 1Ghz is very relevant and gives us a very good idea of how well the architecture competes as well as what its IPC is. Many other hardware sites purposely under-clock newer processors to compare the latest generation architecture to the last generation.

Having said that, the Atom N450 really is not considered a dual-core processor. Rather it is Single Core + Hyper-threading (two-threads). Would be very interesting to see a newer dual-core (ie Atom N550 - 4 threads) at 1Ghz go up against a newer ARM A9 chip such as the Tegra 3 or even an A15 chip. I'll bet it would be a really close battle, with the power consumption win probably going to the ARM architecture.

The x32 results are very interesting, in that it scored a few wins against X86_64. Would be nice to see x86 results for comparison... =)

[rvalles]

Given Cortex-A9 is already old and SoCs based on the Cortex-A15 are starting to appear, I'm looking forward to seeing how those perform.

[gururise]

Running multi-threaded benchmarks on a top of the line dual-core ARM setup then declaring that it is amazingly superior to an UNDERCLOCKED SINGLE CORE ATOM FROM ALMOST 3 YEARS AGO is not exactly showing that Intel is about to go out of business.

I don't remember reading about Intel going out of business in the article; furthermore, the A9 architecture is also almost 3 years old too, so I don't see what your point is.

Larabel *repeatedly* misrepresents the N450 calling it a "dual core" processor. It is *not* dual core but is a single core + HT chip. He could have spent 5 seconds using this new thing called "Google" to go here: http://ark.intel.com/products/42503 and get a full product page describing the N450 that *clearly* shows it is a single core chip. I'd love to see the A9's "beating" the N450 with only one core....

Yes, technically not a dual-core. Its a single core + hyperthreading making it capable of scheduling and partially processing two simultaneous threads. Nonetheless, it is a very interesting comparison, and I hope Michael can get us some results for an A9 Quad Core (ie Tegra 3) vs. A15 vs. latest gen Atom dual core (ie N2800)... that would be a very interesting showdown, though it has been widely reports that the Cedarview ATOM's are no faster than their Pineview predecessors.

[blacknova]

Aren't most benefits of HT are getting visible as CPU frequency is going up? There is little point in downclocking HT Atom and than getting it measured against real dual-core CPU.

[sykobee]

This ARM Cortex-A9 would be great for a nettop, netbook, and chromebook or chrometop.

It was always in my mind that it was the compilers that were holding back ARM when compared to the Intel Atom, rather than the architecture, and this article does a lot to confirm it - although I'm sure that someone will tweak the Atom compiler flags to get better figures soon.

Did the OpenSSL benchmark use any hardware acceleration on the ARM platform?

This also goes a long way to confirming that ARM based tablets aren't lower performing than an Intel Atom based tablet. In addition the Intel Atom based phones aren't actually going to be faster despite the advertising claims that Intel is paying for for the various new Intel based phones - especially as the ARM devices are getting quad-cores at around 1.5GHz in comparison to the single-core Atom at 1.6GHz.

[sykobee]

Running multi-threaded benchmarks on a top of the line dual-core ARM setup then declaring that it is amazingly superior to an UNDERCLOCKED SINGLE CORE ATOM FROM ALMOST 3 YEARS AGO is not exactly showing that Intel is about to go out of business.

Larabel *repeatedly* misrepresents the N450 calling it a "dual core" processor. It is *not* dual core but is a single core + HT chip. He could have spent 5 seconds using this new thing called "Google" to go here: http://ark.intel.com/products/42503 and get a full product page describing the N450 that *clearly* shows it is a single core chip. I'd love to see the A9's "beating" the N450 with only one core....

The idea was to compare the two architectures clock-for-clock.

You can also get 1.7GHz Quad-core ARM Cortex A9 chips.

So a dual-thread A9 compares very favourably to a dual-thread Atom at the same clock. That's a worthy finding.

Of course the quad-core ARMs will be in tablets and phones, but Intel's tablet and phone Atoms are the dual-thread single-core varieties.

[ldesnogu]

Running multi-threaded benchmarks on a top of the line dual-core ARM setup then declaring that it is amazingly superior to an UNDERCLOCKED SINGLE CORE ATOM FROM ALMOST 3 YEARS AGO is not exactly showing that Intel is about to go out of business.

And Cortex-A9 is more than 4 years old. It just takes some time going from what ARM Ltd delivers to final production chips, which have been available for 2 years.

The Atom here is being given two advantages:
1. it's targetting x32, I'd like to see pure 32-bit results
2. its memory is probably not underclocked, and probably is some desktop memory, at lest nothing similar to what you find on ORIGEN board.

Nice results anyway, and unsurprising

[ldesnogu]

This also goes a long way to confirming that ARM based tablets aren't lower performing than an Intel Atom based tablet. In addition the Intel Atom based phones aren't actually going to be faster despite the advertising claims that Intel is paying for for the various new Intel based phones - especially as the ARM devices are getting quad-cores at around 1.5GHz in comparison to the single-core Atom at 1.6GHz.

Especially when the claimed frequency of 1.6 GHz is turbo frequency. The "standard" frequency of Medfield is 1.3 GHz.