Arm Cortex-X3 + Cortex-A715 Announced As Second-Gen Armv9 CPUs

[coder]

Wikichip is saying the performance (IPC) improvement is approximately 11% on real-world applications.
Their article is well worth reading: https://fuse.wikichip.org/news/6855/...ore-cortex-x3/

I finally got around to reading this + the companion article on the A715. Very interesting. I'll mention some noteworthy points, for those lacking time or motivation to read for themselves.

First, the A715:

• Finally drops AArch32 support.
• Actually drops the MOP cache, entirely! This enables a 5th decoder. Slightly better than 5-way throughput, since i-cache includes fusion.

The lack of a MOP cache actually reminds me of Intel's E-cores. Not sure how superficial that similarity really is.

Next, the X3:

• Seems to deviate more from its A-series counterpart than previous X cores? Front-end seems very different.
• Added a 6th decoder & kept the MOP cache, but halved its size (reducing latency). MOP cache bandwidth is 8-way (no change).
• Added 50% more integer ALUs (from 4-way to 6-way; one is ALU/MUL, one is ALU/MAC/DIV)
• ROB size increased from 288 -> 320

All told, the X3 still seems narrower than Intel's Golden Cove and what I think is expected of Zen 4. Therefore, I'm skeptical it can truly challenge them on single-threaded perf, but perf/W and therefore multithreaded performance is more likely in contention. It'll be interesting to see how many more changes ARM is putting into its server-oriented variant (V3?).

[Anux]

Interesting. How does it achieve that? It does look more or less the same adjusted by battery capacity.

Yeah, the M1 has to support a bigger display and probably has stronger CPU & GPU, so still more efficient.

The benchmark against which it should be measured is Nintendo Switch

The new OLED version is commonly credited with 4.5 h gaming. It has a much slower CPU und GPU (Tegra 1) but builds on an old 16nm+ process.

[drakonas777]

but perf/W and therefore multithreaded performance is more likely in contention

Talking about MT efficiency: https://www.youtube.com/watch?v=FWfJq0Y4Oos

I must say I'm quite disappointed. In some MT workloads (software only, no accelerators of course) M2 MT efficiency is actually worse than x86, despite the fact it's using a new node and "The ARM"

BTW almost all the tests they did used native Apple Silicon apps, though cross-platform.

[coder]

Talking about MT efficiency: https://www.youtube.com/watch?v=FWfJq0Y4Oos

Nah, don't watch youtube.

Doesn't anybody read, any more?

[t.s.]

Of course, Qualcomm could shoot themselves in the foot by demanding unrealistic pricing. People aren't going to abandon x86 for something more expensive that's not really a lot better. But, if it's comparably fast, has better battery life, and is even a little bit cheaper, then they probably have all the ingredients needed to gain some real market share.

It's not that Qualcomm shoot themselves in the foot, but their greedy friends (OEM) that shoot Qualcomm's foot. They (Qualcomm) used to complain that the ARM PC prices is too high that it hamper the adoption. But as usual, the OEM's not listening. Too greedy.

[t.s.]

On the M1 playing games reduces the runtime to 5 h, M2 doubling that sounds a little bit to good. You have a link?

Doubt it:

• https://www.extremetech.com/extreme/...der-heavy-load
• https://www.extremetech.com/computin...nder-full-load

[Anux]

Doubt it:

• https://www.extremetech.com/extreme/...der-heavy-load
• https://www.extremetech.com/computin...nder-full-load

This just says it throttles and has a slower SSD, what about gaming runtime, did I miss something?

[t.s.]

This just says it throttles and has a slower SSD, what about gaming runtime, did I miss something?

I've seen the M2 slides that the M2 power + freq higher than M1. Forgot where I see it. But from Apple's site, looks like I'm in the wrong.

[qarium]

Talking about MT efficiency: https://www.youtube.com/watch?v=FWfJq0Y4Oos
I must say I'm quite disappointed. In some MT workloads (software only, no accelerators of course) M2 MT efficiency is actually worse than x86, despite the fact it's using a new node and "The ARM"
BTW almost all the tests they did used native Apple Silicon apps, though cross-platform.

i told this to others in this forum thread and i will tell you this:

the main difference between x86_64 and apple ARM you on the same node is not the higher performance or power efficiency the main different is the "tranistor count". or lets say the "Death tranistors" who do nothing not even consuming power.

on the same chip node apple can build smaller in mm² chips means less fabrication node costs means higher profit.

its maybe only 5% on the cpu side compared to x86 but if you sell million of chips this is a lot.

but i think it is much more than 5% just compare the intel ARK gpu vs AMD GPUs on 100mm² vs 150mm² amd is faster...

apple has a zero copy design means nearly zero buffers all this saves a lot of tranistors.

if they only do macOS and Ios they can not compete agaist amd or intel but if they enter the linux market to you could see intel and amd losing marketshare because apple could always sell the chips cheaper because less tranistors....