The Snapdragon 8 Gen 1 Performance Preview: Sizing Up Cortex-X2

anandtech.com

4 points·neogodless··4 comments

It really isn't as bad as I thought.

The Cortex X2 CPU in 8 Gen 1, on a Samsung 4nm which is more like TSMC 6nm+ ( Yes Samsung mess up node number again ) needs to be around 40% faster in order to catch up to Apple's TSMC 5nm+ A15. That is not too bad considering the X2 has 1MB of Cache per Core, compared 12 MB of Cache shared between two A15 Core. Yes. 1MB and 6MB difference. And a 4MB of System Level Cache compared to 32MB on A15. Six and Eight times the difference.

I would imagine a Cortex X2 with the same cache, and on a TSMC 5nm+ node would easily be 20% faster. The Power difference should also be level with same cache. Still not a fast as A15. But the gap isn't as big as many have thought.

The GPU is actually on par with Apple. That probably requires more testing in terms of energy usage.

This becomes interesting to see when Mediatek release their own X2 as well.

In the best case Scenario, Apple A15 Single Core has access to 8.8 times the cache size compared to Cortex X-2. With a single node generation advantage. The cache size was likely a requirement for Apple's 8-wide decoder design. But ARM's design could have squeezed more performance with larger cache as well.

> There is no AV1 decode engine in this chip, with Qualcomm’s VPs stating that the timing for their IP block did not synchronize with this chip.

I wonder is they did much effort to support AV1 since they have many patents for MPEG/ITU codecs. Dimensity 1000 (announced in 2019) supports AV1.

The new media engine will obviously comes with VVC as well. It makes much more sense to do both at the same time since their IP are in house. I believe Mediatek's Video Encoding and Decoding Engine are licensed IP ( May be Chips & Media).

Reply on news.ycombinator.com