Hacker News Reader

Comment by ip26 | Hacker News Reader

Something has got to be screwy. The iPhone has a TDP of 6W and a clock speed of 2.5GHz. The MBP has boost clocks of 5GHz and a 45W TDP.

View on HN

Each A13 Lightning core has around 2.5x the transistors of a Skylake core. The Skylake design is also 5 years old and on a node at least a generation out of date. Geekbench is also a short enough test that thermals aren’t a huge concern.

Where are you getting the per-core transistor counts from?

It’s a very rough estimation. Basically ((core + L2 area)/(total die area))(Total transistors on die).

For A13 with unified L2, I used L2 per core.

Of course, transistor density isn’t consistent across the chip, but this should be good enough for the relative comparison I’ve done.

Yes but AMD is on a much newer node and still only achieves a very slight improvement in IPC nowhere near >2x these benchmarks imply.

Zen 2 is much more economical in its core area. Lighting cores likely have around 2x the transistors.

Also, and I’m not sure how big of a deal this is, but Zen 2 is the 2nd gen of a practically new arch. Lighting has a lineage of at least 6 generations.

Something to consider Intel's TDP is given at base clock - boost is irrelevant (i.e. they skew the facts the CPUs are way more toasty than TDP). Increasing the frequency requires more voltage, power is proportional to frequency and square of the voltage.

Other than that - x86 cores (skylake) are generally more powerful per clock than any Arm.

I'd not trust geekbench a bit between different platforms. for example:

The LZMA workload compresses and decompresses a 2399KB HTML ebook using the LZMA compression algorithm with a dictionary size of 2048KB. The workload uses the LZMA SDK for the implementation of the core LZMA algorithm. The test effectively is about L2/3 cache sizes, if it fits there - great result, if not - a horrid one.

Most of the testing lacks any specific data, just what library they use.

If those numbers were in any way real world relevant, the server market would be flooded with ARM servers. 80% power savings for the same computing power would revolutionize data centers in terms of power and cooling needed.

There's much more to server hardware than CPU power and efficiency, and it's difficult segment to enter because of the risk involved making large investments in unproven hardware. But despite that, it still seems like things are actually moving in that direction and ARM is starting to get a serious foothold in the server market now. For example consider Amazon's new Graviton2 CPU's [1], which according to AnandTech 'puts all x86 alternatives to shame' [sic] when it comes to the price per unit of compute time it provides. I think it's entirely likely that the server market will be flooded with ARM servers in the not too distant future.

[1] https://www.anandtech.com/show/15578/cloud-clash-amazon-grav...

Synthetic benchmarks are extremely unreliable.

I imagine even more so than usual when the two things being tested have different operating systems and CPU architectures. Although these benchmark results would still have to be wrong by a lot to stop being impressive.

I think, Geekbench favors bursty performance because it gives very close scores to new MacBook Air with the Y-series Intel chip (7 watt TDP) and newest Dell XPS 13 with U-series chip with TDP of 25 watts.

Reply on news.ycombinator.com