State of mobile chip (Qualcomm Kryo, ARM Cortex A72, Intel Goldmont)

[icrf]

Why isn't Mongoose included in the thread?

My understanding is it's a very slightly modified A72[1], kind of like how Apple's A4 was really just a tweaked ARM Coretex A8.[2] That's not to say future versions won't be fully custom, but for the short term, we can probably just lump it in with A72.

 

[Lepton87]

My understanding is it's a very slightly modified A72[1], kind of like how Apple's A4 was really just a tweaked ARM Coretex A8.[2] That's not to say future versions won't be fully custom, but for the short term, we can probably just lump it in with A72.

Well, I have seen other data, from Anandtech I read

While the power numbers are interesting we also have to put them into context of the achieved work. ARM has made several optimizations to the architecture to improve performance when compared to the A57. We'll get into more detail in just a bit - but what we are looking at is a general 16-30% increase on IPC depending on the kind of workload. Together with the power reduction, we now see how ARM is able to advertise such large efficiency gains for the same fixed workload.

From Samsung slides Mongoose is claimed to be 45% faster than A57 although that wasn't IPC comparison but absolute performance comparison

 

[witeken]

Although I don't buy the VISC hype/BS, I do think the A9X/Skylake comparison here is interesting:

It's even more interesting in the other graph where just one of the ridiculous normalizations was removed (the cache). And never mind the 1.65x hit that SKL got from downgrading to 16nm. And that was just a midrange i5 chip from random non-Apple laptop.

Also note the IPC advantage.

Skylake already blew the A9X out of the water after the cache handicap was removed, so never mind if this was against real life 14nm Skylake. Intel is so much superior. Apple still has a long road to go.

-----

BTW, from that huge image from Samsung: it's ridiculous that they say "shorter distance from 20nm -> 14nm".

 

[Exophase]

Skylake already blew the A9X out of the water after the cache handicap was removed, so never mind if this was against real life 14nm Skylake. Intel is so much superior. Apple still has a long road to go.

You do know they did the same normalizing games with A9's score too, right? I say A9 and not A9X because that's what they actually tested on, and that means they probably corrected for its L3 cache and large L2 cache too.

Their numbers aren't really worth considering at all, IMO.

 

[witeken]

You do know they did the same normalizing games with A9's score too, right? I say A9 and not A9X because that's what they actually tested on, and that means they probably corrected for its L3 cache and large L2 cache too.

Their numbers aren't really worth considering at all, IMO.

Fair enough. Their fancy PowerPoint slides are too heavily marketing influenced, not scientific.

 

[Andrei.]

My understanding is it's a very slightly modified A72[1], kind of like how Apple's A4 was really just a tweaked ARM Coretex A8.[2] That's not to say future versions won't be fully custom, but for the short term, we can probably just lump it in with A72.

M1 looked similar to A72 at first glance and from a high level but it's not a "slightly modified A72". I wish people would stop expanding my comments on it into something that I didn't say.

 

[icrf]

M1 looked similar to A72 at first glance and from a high level but it's not a "slightly modified A72". I wish people would stop expanding my comments on it into something that I didn't say.

My apologies. I was remembering the old Apple A4 and read more into "might be a derivative" than I should have.

 

[Systems analyst]

ARM do offer licences now for slight modifications of some of their cores, but when the firstlings will appear is not clear.