• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Cortex A15 v.s. Nehalem/Sandy Bridge?

Arachnotronic

Arachnotronic

Lifer
So, anybody here able to give me an idea of how the Cortex A15 compares to Nehalem/SNB? The hype has me believing that the A15 will smoke the current i7 chips, but I don't know too much about the ARM stuff...
 
While we're at it, who will be the 99th president of the USA?



















...impossible to answer both questions until the time comes and we have solid evidence
 
As has been said, you either have to be trolling or don't know much.

ARM has a LOOOONG way to go to match Nehalem in IPC at integer workloads, and a LOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOONG way to go to match it in floating point workloads.
 
Last edited:
CHADBOGA said:
The A15 core might have 20% of the IPC of a Nehalem core.
Click to expand...

It's nowhere near that bad. Honestly, the IPC for both chips is probably fairly similar, but x86 chips have a lot more sophisticated hardware that allows them to make use of that potential. Also, the Nehalem core is 64-bit whereas the A15 is only 40-bit for the purposes of the amount of memory it can address. The Nehalem architecture is also designed to run at significantly greater clock speeds, but neither really has anything to do with IPC.

Edit: Wikipedia lists the two architectures as having similar instructions per second / per clock / per core, but as I said before, that's just theoretical numbers. A15 based SoCs are going to make for some incredible devices.
 
Last edited:
Intel17 said:
So, anybody here able to give me an idea of how the Cortex A15 compares to Nehalem/SNB? The hype has me believing that the A15 will smoke the current i7 chips, but I don't know too much about the ARM stuff...
Click to expand...
No idea, until someone benchmarks them. But then, have to use different binaries since Cortex A15 is based on RISC (different instruction set). :hmm:

Mopetar said:
It's nowhere near that bad. Honestly, the IPC for both chips is probably fairly similar, but x86 chips have a lot more sophisticated hardware that allows them to make use of that potential.
Click to expand...
Also x86 has more developed and mature compilers which include optimizations (for highly efficient code execution). 😉
 
Already have a Cortex A10, will this work in my current motherboard? I don't want to have to buy a new board so soon, I just got this one.
 
What devices actually have a Cortex A15 right now? Also, does a high performance ARM still retain the low-power consumption advantage over x86 processors?
 
gevorg said:
OP, you might find the link below interesting. ARM won't be anywhere near as fast Intel x86 anytime soon, but A15 might be fast enough for general/casual laptops.

http://semiaccurate.com/2011/05/05/apple-dumps-intel-from-laptop-lines/
Click to expand...

That link is full of shit really. x86 is a moving target, and competing with an ARM-powered notebook against 22nm and beyond Intel x86 chips on the Windows side would be suicide. The most powerful ARM solutions on earth would probably have trouble keeping pace with a 7-year-old Opteron DC.

Now expansion of the iPad type device to have an integrated foldup keyboard? I could totally see that, sort of a half-step between a tablet and a notebook, particularly if it had a 12" screen or so, would be a very interesting product indeed, and would mesh well with the increasing (but still waaaaay behind) power of ARM-based solutions.
 
Intel17 said:
So, anybody here able to give me an idea of how the Cortex A15 compares to Nehalem/SNB? The hype has me believing that the A15 will smoke the current i7 chips, but I don't know too much about the ARM stuff...
Click to expand...

Not sure about the A15, but the A9 (Dual 1ghz used in tegra2) is slow, even if compaired with a C-50 APU
(the bobcat dual core 1ghz) (these go down to 4.5w TPD on 40nm).

Also like others have said, they need differnt compilers so its very limited what benchmarking excists.

Do I think "Cortex A15 compares to Nehalem/SNB?"

954-not-sure-if-serious.jpg
 
Arkaign said:
That link is full of shit really. x86 is a moving target, and competing with an ARM-powered notebook against 22nm and beyond Intel x86 chips on the Windows side would be suicide. The most powerful ARM solutions on earth would probably have trouble keeping pace with a 7-year-old Opteron DC.

Now expansion of the iPad type device to have an integrated foldup keyboard? I could totally see that, sort of a half-step between a tablet and a notebook, particularly if it had a 12" screen or so, would be a very interesting product indeed, and would mesh well with the increasing (but still waaaaay behind) power of ARM-based solutions.
Click to expand...

Time will tell, ARM doesn't need to beat Intel in speed, it just need to be fast enough for general use and cheap enough to produce. And of course extremely power efficient. Even A5 in iPad 2 already has enough power to be used as a primary/main computing device for some people. The x86 has way too much outdated and inefficient baggage, and sooner or later it will have to fade away.
 
The x86 has way too much outdated and inefficient baggage, and sooner or later it will have to fade away.
Click to expand...

Or

Intel/AMD sit down and agree on what to remove of the legacy support, and a new Modern up-to-date x86 is born.

That suddenly has better power consumption, and is smaller because of those things removed.
 
Arkadrel said:
Or

Intel/AMD sit down and agree on what to remove of the legacy support, and a new Modern up-to-date x86 is born.

That suddenly has better power consumption, and is smaller because of those things removed.
Click to expand...

That would be awesome and it won't be x86 anymore. 😉
 
gevorg said:
The x86 has way too much outdated and inefficient baggage, and sooner or later it will have to fade away.
Click to expand...

The first time I heard that, an engineer was eyeballing new Sparc5's I was unpacking.

That was a while ago.

:whiste:
 
Mopetar said:
It's nowhere near that bad. Honestly, the IPC for both chips is probably fairly similar, but x86 chips have a lot more sophisticated hardware that allows them to make use of that potential. Also, the Nehalem core is 64-bit whereas the A15 is only 40-bit for the purposes of the amount of memory it can address. The Nehalem architecture is also designed to run at significantly greater clock speeds, but neither really has anything to do with IPC.

Edit: Wikipedia lists the two architectures as having similar instructions per second / per clock / per core, but as I said before, that's just theoretical numbers. A15 based SoCs are going to make for some incredible devices.
Click to expand...

neither Intel nor AMD cpus can currently address 64 bits worth of memory. AMD can address something like 2^44 and Intel 2^40 or 2^42.

IE, plenty for anything we need within the next 10 years. This can be expanded its no big deal, but to act like it's bad that ARM can only touch 2^40bits of memory is to be misguided
 
I've been shaking my head in bewilderment since some ignorant (and yes, I mean that in the literal sense) media person got confused and thought arm was a competitor to x86 in any way other than ultra-low power applications. It keeps getting parroted by people and I really, really can't figure out why. It's kind of like asking if a fiat 500 is a performance competitor to lotus exige because their curb weights are within 400 lbs of one another....

Cannot tell if trolling!
 
I really am not trolling. What prompted me to ask is actually the NVIDIA claim that their Kal-El @ 1.5GHz = Core 2 Duo @ 2.0GHz. If the Cortex A15 = 2*Cortex A9, then at 2.5GHz, wouldn't this be a very viable competitor to the modern Intel chips?
 
gevorg said:
Time will tell, ARM doesn't need to beat Intel in speed, it just need to be fast enough for general use and cheap enough to produce. And of course extremely power efficient. Even A5 in iPad 2 already has enough power to be used as a primary/main computing device for some people. The x86 has way too much outdated and inefficient baggage, and sooner or later it will have to fade away.
Click to expand...

x86 won't die because the processors are relatively inexpensive due to economies of scale and performance is exceptional for the price as a result. Companies have been trying to kill it for the past 30 years with absolutely no success. Whatever baggage that x86 brings with it has been easily mitigated by process and design improvements. Just look at the architectures that have fallen: Alpha, PA-RISC. And then there are the ones that have been greatly diminished on the desktop: PowerPC, 68k, Sparc, etc.
 
Intel17 said:
I really am not trolling. What prompted me to ask is actually the NVIDIA claim that their Kal-El @ 1.5GHz = Core 2 Duo @ 2.0GHz. If the Cortex A15 = 2*Cortex A9, then at 2.5GHz, wouldn't this be a very viable competitor to the modern Intel chips?
Click to expand...

That's because NVIDIA is full of crap. CoreMark, like has been mentioned earlier, is a SYNTHETIC benchmark that inflates the results for ARM by a huge amount, not to mention NVIDIA used dirty tactics to change the results. They used a recent version of CoreMark for Kal-El, and an old one for Core 2 Duo. Some people present saw the different versions for each, and later retested it on a laptop with the same C2D NVIDIA used. The result? Kal-El had gone from being "10% faster" to "40% slower" than the Core 2 Duo T7200. It's not even possible to completely compare the two directly, but even using a heavily biased benchmark Core 2 still won.

No, it's not gonna catch up. They used Quad-Core Kal-El for the comparison, and for all performance improvements they assume perfect scaling, which we all know is not possible on most programs.

Then there's also the huge issue ARM has with floating point performance. Just for a quick comparison, a 600MHz Cortex A8 scores 23MFLOPs on Linpack, while a 1.6GHz Intel Atom N270 scores 933MFLOPs on the same test. Again, they using different instruction sets could change somewhat the results (in Intel's favor, that is), but you can see ARM won't be able to catch X86 in this metric ever as the gap is too huge.

TL;DR:


  • In NVIDIA's own biased benchmark, Quad-Core Kal-El actually loses by 40% to the Core 2 Duo T7200.

  • NVIDIA tried to heavily skew their already biased results, so we shouldn't trust them too much in the future.

  • CoreMark assumes perfect scaling, which is not possible on most programs.

  • ARM has a huge deficit in floating point performance, to the tune of being more than 200x slower than Atom in Linpack.
 
Memory performance on the arm architecture is 5-8% of what it is with typical x86.

No, that's not a typo. 5 to 8 percent.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top