ARM CPUs vs X86 CPUs?

[Pawnless Endgame]

I'm really curious on how the current mobile processors fare in real computing terms when compared with a desktop processor. Granted, ARM processors have a smaller instruction set and run on much lower power, but they have come a long way!

e.g. The original Raspberry Pi had a 700MHz ARM processor, which had a desktop comparison to a Pentium II 300MHz from the late 1990s.

So now, take a Galaxy S4 or a Note 4, and how does that compare in the desktop world? Have we surpassed Pentium 4 yet or are we at more like Pentium D / Core 2 Duo level now?

Then as a more general question, what are you guys using nowadays to benchmark CPUs? In the 1990s, it was all about the megahertz, then mid-2000s introduced hyperthreading, extra cache, more levels of cache, extra cores and 64-bit capability so it all got a bit complicated to "quantify" a CPU. For the last few years, I have used Passmark's web site which displays a table of most of the processors since approx. Pentium 4 for score comparisons.

 

[Orcvader]

I think I remember one site testing this, and the A15 core in the Nexus 10 (or was it a Chromebook?) performed pretty competitively against the Intel Atom of that time. And I think I remember reading something a couple of years back that the Atom was finally at P4 levels. Remember a PD is just two P4 dies slapped together, so I guess ARM has already passed that level.

I have no idea how the A57/A72 cores compare to the current x86 CPUs though. Nvidia's Denver cores are pretty high end too. Qualcomm's Kryo core is expected to at least be a minimum of A72 level.

The fact the A57 core is used in some servers though shows that ARM does have the performance.

These days it seems to be about heterogeneous computing, software is finally offloading tasks that are better suited to the GPU now. That just makes it a little more complicated on the best way to benchmark a system. I just check review sites to see the average single threaded performance and multithreaded performance over various different benchmarks to get a good idea on general performance.

EDIT: Found the A15 vs Atom (Pineview): http://www.anandtech.com/show/6422/samsung-chromebook-xe303-review-testing-arms-cortex-a15/6

They only did a couple of tests to check the CPU but the A15 seems to beat the Atom.

EDIT 2: And here's the Atom (Diamondville, predecessor to Pineview) vs the P4: http://www.tomshardware.co.uk/atom-d510-pentium-4-nettop,review-31948-10.html

I remembered it wrong, the P4 still beats it (apart from power usage ). But I'm pretty sure there was an Atom core with significant CPU improvements... I forgot which one it was.

 

[Gazb]

ARM cpu's in servers can not be realistically compared with x86 cpu's as they tend to be used for very different tasks. You don't need a full featured x86 cpu when all you are doing is running some java virtual machines/runtime for a web applications.

 

[Dave2150]

Compared to Skylake/Haswell-E, they are complete garbage. Obviously so, since they are designed for tiny devices and therefore cannot support a high TDP.

 

[Rroff]

Its hard to directly compare the performance. Generally though they've surpassed P4 performance for awhile in a loose sense.

I used to use RISC OS for a long time back in the day and with properly written software for the instruction set the little 12MHz CPUs would give up performance comparable to 100-200MHz Intels for things like video rendering and general desktop use, etc.

A lot of those mobile lower end cores though are fast catching up with ~2007 era mid-range CPUs which is quite impressive when many of them have like 2 watt TDPs.

 

[JonSmith]

It depends for what kind of application. Many server workloads are very cache intensive and you don't need a particularly powerful core, but you could tune the CPU cores to minimize cache latency (or increase cache size), stick ethernet, SATA and memory controllers on chip and you can pack more RAM and disks into the same footprint without blowing the power/heat budget in the rack.

Being able to purchase the chip design and modify it to suit your application is a massive advantage.

 

[Pawnless Endgame]

Thanks guys - I was really curious

I totally forgot that Pentium D was 2 P4 cores on separate dies, as opposed to both cores on the same die.

What's TDP by the way? Is that relative to FLOPs / floating point calculations?

 

[pmc25]

Compared to Skylake/Haswell-E, they are complete garbage. Obviously so, since they are designed for tiny devices and therefore cannot support a high TDP.

An A72 core's ~2.5x faster than the equivalent marchitecture to Skylake for mobile, and less than half the wattage.

I'd say it's the opposite. x86 is 'garbage'. it's slow, hot, and power hungry. There just aren't any really big ARM cores ... yet. It's not just the instruction set as some people seem to think, it's Intel's patents and the terms of the x86 license (which AMD and VIA have, but NVIDIA were refused -- Denver was meant to be a GPCPU with x86 & ARM compatibility). Intel have been restricting development and advances for the past 30 years. Now they're reaching a point where they've finally been caught (and arguably overtaken) in process tech, ARM is gutting large sections of the server market, it'll soon move into thin client desktops for the corporate and commercial markets, and a lot of PCs and terminals are being replaced with tablets and smartphones. Ultimately they're going to have to do an about face and make ARM chips soon (they won't relax x86 rules / enforcement - they'll cling on to that to the bitter end). If Zen is actually really good, this may expedite it, as they can't magic a new desktop / laptop / high end server x86 roadmap out of thin air, after purposely dragging their feet for so long.

 

[dbmzk1]

Thanks guys - I was really curious

I totally forgot that Pentium D was 2 P4 cores on separate dies, as opposed to both cores on the same die.

What's TDP by the way? Is that relative to FLOPs / floating point calculations?

TDP = Thermal Design Power. Basically it's the maximum amount of heat that it can generate and therefore dictates what kind of cooling it needs. It's used rather than the actual power usage of the chip as this can vary massively and not necessarily have as direct impact on the thermal output as you might expect, with differing architectures etc. TDP is massively important - you can't put a huge desktop rated chip into laptop as you just can't cool it efficiently, etc.

The Pentium D (well, most of them) was rated at 130w TDP. Most higher end CPUs vary between 80-130w. Where as, for example, the Exynos 7420 SoC used in the Samsung Galaxy S6 is rated at around 3w TDP and that includes the graphics etc. As you can see there is a massive difference there, which explains why we're basically comparing apples and oranges here.

 

[Blackjack Davy]

I'd say it's the opposite. x86 is 'garbage'. it's slow, hot, and power hungry.

Heh, yes they're behemoth's. Monsters, really. They exist for one reason only: Microsoft Windows and backwards software compatibility. Inside every x86 processor there is a 1987 8088 processor like a singularity sitting in the core of a Black Hole struggling to get out. Its totally and utterly irrelevent to modern computing and completely necessary to allow Windows to function. I don't think consumers have any inkling to the lengths to which hardware designers have gone to replicate the original 64bit memory address space duplicated countless times to allow each process to run inside its own little version of the original IBM PC.

ARM chips don't carry any of that baggage and their instruction set is much more efficient.

 

[Rroff]

An A72 core's ~2.5x faster than the equivalent marchitecture to Skylake for mobile, and less than half the wattage.

I'd say it's the opposite. x86 is 'garbage'. it's slow, hot, and power hungry. There just aren't any really big ARM cores ... yet. It's not just the instruction set as some people seem to think, it's Intel's patents and the terms of the x86 license (which AMD and VIA have, but NVIDIA were refused -- Denver was meant to be a GPCPU with x86 & ARM compatibility). Intel have been restricting development and advances for the past 30 years. Now they're reaching a point where they've finally been caught (and arguably overtaken) in process tech, ARM is gutting large sections of the server market, it'll soon move into thin client desktops for the corporate and commercial markets, and a lot of PCs and terminals are being replaced with tablets and smartphones. Ultimately they're going to have to do an about face and make ARM chips soon (they won't relax x86 rules / enforcement - they'll cling on to that to the bitter end). If Zen is actually really good, this may expedite it, as they can't magic a new desktop / laptop / high end server x86 roadmap out of thin air, after purposely dragging their feet for so long.

The newer Core M, Z8700, etc. easily hold there own against the equivalent ARMs i.e. A72 (some caveats in regards to thermal/power scenarios aside) and some nice stuff in the pipe - certainly won't be going back to android in the tablet space personally.

 

[Orangey]

Heh, yes they're behemoth's. Monsters, really. They exist for one reason only: Microsoft Windows and backwards software compatibility. Inside every x86 processor there is a 1987 8088 processor like a singularity sitting in the core of a Black Hole struggling to get out. Its totally and utterly irrelevent to modern computing and completely necessary to allow Windows to function. I don't think consumers have any inkling to the lengths to which hardware designers have gone to replicate the original 64bit memory address space duplicated countless times to allow each process to run inside its own little version of the original IBM PC.

ARM chips don't carry any of that baggage and their instruction set is much more efficient.

Bloat is unavoidable for an ISA that tries to be all things to all people. I have even seen people call the new ARMs bloated.

 

[almoststew1990]

In terms of gaming performance, my nexus 7 first gen (tegra 3 I think that is) runs GTA 3, vc and San Andreas just fine (and is graphics limited). Obviously Tegra3 is now very old but even that was equivalent of... I dunno, an early 1.8GHz P4?

Although I assume Android 4.4 is a lot lighter than Win XP so the Tegra chip doesn't have as much "other stuff" to think about compared to the Pentium chip running GTA3.

Maybe there is a pi calculation app that runs on both windows and adnroid that I could try out with some old CPUs I've got hanging around...

 

[Rroff]

My Galaxy Note 1 (Dual-core 1.4 GHz ARM Cortex-A9/Mali400) runs GTA 3 and Max Payne fine at 1280x800 full quality settings which is quite surprising.

 

[TheMorningStar]

I often wonder who far the current gen APUs, IGPUs are behind the high end dedicated cards, are we at HD 4800 series performance yet, for instance.