[Hexus] ARM compares A72 vs Intel Broadwell-Y

[Mondozei]

They make a bunch of interesting claims, such as that A72 is 75% more energy efficient and in some applications offer 50% more performance.

If Chrome OS becomes more than a glorified webclient, this could seriously undermine Intel's plans. Intel basically has no foothold in mobile. All they have gained has come at a cost of so-called "contra revenue" which is unsustainable.

Or if Microsoft continues it's pro-Android charm offensive. Having Android apps on Win 10 isn't a bad move and with proper underlying software support for ARM, could crush Intel from that side, too.

 

[ShintaiDK]

When was the last time ARMs claims held water? Because I cant remember it.

Even their own numbers against A57 doesnt hold. The equalized performance delta is 47.3% arcording to one slide. Yet its much less in the next between 16 and 50%.

 

[kimmel]

Comparing an estimated performance of a currently non-existent 2016 product ingredient to a real shipping product that released in 2014.

Also they are comparing it to the product that limited the performance the most as was documented in Anandtech's report.
http://anandtech.com/show/9117/analyzing-intel-core-m-performance/13

:awe:

 

[jj109]

Okay.

I'm sure there's some crazy RTL model at Intel that blows ARM out of the water, too.

Is ARM scared?

 

[Arachnotronic]

Comparing an estimated performance of a currently non-existent 2016 product ingredient to a real shipping product that released in 2014.

Also they are comparing it to the product that limited the performance the most as was documented in Anandtech's report.
http://anandtech.com/show/9117/analyzing-intel-core-m-performance/13

:awe:

There are just so many things wrong with this ARM comparison that I don't even know where to begin...

 

[monstercameron]

Interesting responses, how exactly is it inaccurate?

 

[Fjodor2001]

So perf/watt dominance for ARM A72 in the server segment, where perf/watt is all that matters according to the Intel crowd. Interesting. x86 death ahead?

 

[erunion]

If Chrome OS becomes more than a glorified webclient, this could seriously undermine Intel's plans.
.

How? Intel is the majority supplier for chrome OS.

 

[ShintaiDK]

So perf/watt dominance for ARM A72 in the server segment, where perf/watt is all that matters according to the Intel crowd. Interesting. x86 death ahead?

If you believe ARMs claims, it happend long before A72. But when products gets tested, ARM doesnt live up to its claims. What a shame. But again, marketing is marketing. Reality is reality.

 

[Fjodor2001]

If you believe ARMs claims, it happend long before A72. But when products gets tested, ARM doesnt live up to its claims. What a shame. But again, marketing is marketing. Reality is reality.

Yeah, that's why they dominate mobile totally.

 

[ShintaiDK]

Yeah, that's why they dominate mobile totally.

Its more about cost structure and integration.

But since you claim its so good. Why isnt ARM everywhere in the serverspace? How was that X-Gene review again?...

 

[Tuna-Fish]

Interesting responses, how exactly is it inaccurate?

For one, only one of the cpus in the comparison actually exists as of now. RTL models are good, but trying to divine performance/watt from one is not exactly something you can do with any certainty. Especially when talking about an implementation on a process that just hasn't made enough chips yet for any real numbers. This kind of RTL model would have and did show Prescott clocking >5GHz with no heat problems. The physical effect that did it in was simply not sufficiently characterized when the simulations were made so it was not part of them.

 

[Fjodor2001]

Its more about cost structure and integration.

But since you claim its so good. Why isnt ARM everywhere in the serverspace? How was that X-Gene review again?...

Just give it some time and it'll happen. Main problem is likely some SW is x86 only.

 

[ShintaiDK]

Just give it some time and it'll happen. Main problem is likely some SW is x86 only.

No its not. The X-Gene showed exactly what was wrong. It was severely lacking hardware performance and performance/watt.

 

[Fjodor2001]

No its not. The X-Gene showed exactly what was wrong. It was severely lacking hardware performance and performance/watt.

And that's the only test that matters?

 

[CHADBOGA]

And that's the only test that matters?

Real life > simulations

 

[Guest1]

Real life > simulations

Unless you live in a simulated world :-D

 

[Fjodor2001]

Real life > simulations

So the answer to that question is what...? Got any solid evidence that A72 has worse perf/watt in all use cases compared to Intel?

 

[Tuna-Fish]

So the answer to that question is what...? Got any solid evidence that A72 has worse perf/watt in all use cases compared to Intel?

No. But we got no real evidence to the other direction either.

The sensible criticism to this piece is not "ARM will surely suck", but "There is insufficient evidence backing the claims stated in the article, shame on them for making it."

Power consumption numbers from RTL models? Yeah, right.

 

[monstercameron]

No. But we got no real evidence to the other direction either.

The sensible criticism to this piece is not "ARM will surely suck", but "There is insufficient evidence backing the claims stated in the article, shame on them for making it."

Power consumption numbers from RTL models? Yeah, right.

benefit of the doubt? do we have any issues trusting ARM's claims?

 

[ShintaiDK]

benefit of the doubt? do we have any issues trusting ARM's claims?

A long record of being notoriously wrong/misleading in their marketing?

 

[TurtleCrusher]

Its more about cost structure and integration.

But since you claim its so good. Why isnt ARM everywhere in the serverspace? How was that X-Gene review again?...

Just because Honda and Toyota makes some of the best consumer vehicles, you don't see them getting trash talked because their products aren't in semi-trucks and tanks. The reason is because it's a non-sequiter.

Every adult has a phone, and most now have an ultramobile device in a tablet of some kind. Who walks around carrying servers in their pockets? That's not their primary focus and you know it.

 

[CHADBOGA]

So the answer to that question is what...? Got any solid evidence that A72 has worse perf/watt in all use cases compared to Intel?

As much evidence as there is to suggest that A72 will be 75% more energy efficient with 50% performance edge.

 

[Hulk]

Even if this isn't really accurate, and I actually don't believe it is, I like it. Anything that pushes Intel is good by me.

By the same reasoning while I personally don't like Apple products I love the fact that other people like them and they are doing enormously well in the marketplace.

Go Apple!!! Go ARM!!!

 

[Tuna-Fish]

benefit of the doubt? do we have any issues trusting ARM's claims?

The thing is, it's not just that they are possibly misleading us, which they certainly have in the past (go look up old cortex-a8 marketing...), but that given the current state of the 14nm processes on which this would be manufactured, there's simply no way for them to know how power-sippy the final, mass-produced products will be.

IPC is represented well by a RTL simulation. Those claims are probably correct on the workloads they tested. Clock speeds are a little shaky. So long as there is nothing major and unexpected, they should have some idea how their chips will clock, to a few tens of percent. Power consumption estimates could be totally off.

Making strong claims about performance/watt at this point in development cannot possibly be honest. They can't know yet.