Discussion Disaggregated tile latency penalty and Arrow Lake/Lunar Lake prediction for core to core latency

[igor_kavinski]

The Intel Core Ultra 7 155H Review: Meteor Lake Marks A Fresh Start To Mobile CPUs

www.anandtech.com

https://images.anandtech.com/doci/21282/Intel%20Core%20Ultra%207%20155H%20Core%20to%20Core%20Latency.png

https://images.anandtech.com/doci/21282/AMD%20Ryzen%209%207940HS%20Core%20to%20Core%20Latency%202.png

Looking at the awful core to core latencies on MTL, it seems it was supposed to be a multicore monster that has been let down by Intel's disaggregated tile inexperience. In contrast, AMD is the MASTER of chiplet design.

So, three questions:

What kind of performance improvement did MTL miss out on if it had the same core to core latency as Ryzen?

How worse do you think desktop ARL's core to core latency will be compared to, let's say, 13900K (14th gen is currently under investigation for degradation so let's not bother about that)?

How many of you seriously believe that LNL will fix MTL's bad core to core latencies and bring them in line with Ryzen's?

 

[mikk]

MTL lost like 5% in ST IPC over Raptor Lake I think, no clue how big this penalty grows for MC. Lunar Lake should fix it given that the IMC moves back into the compute tile.

 

[SteinFG]

MTL lost like 5% in ST IPC over Raptor Lake I think, no clue how big this penalty grows for MC. Lunar Lake should fix it given that the IMC moves back into the compute tile.

Did Meteor lose IPC? First time I'm hearing that. Can someone link a resouse? I'm out of the loop on intel stuff

 

[igor_kavinski]

Intel's Meteor Lake CPUs are slower at single-core work than previous-gen models — new benchmarks show IPC regressions vs Raptor Lake

Probably not what Intel wanted.

www.tomshardware.com

Sad 🙁