AMD

There seems to be a lot of criticism and worry about AMD using Sam 4nm LPP+. But if you look into it, things have improved with Sam 4 nm LPP+ compared to 4nm LPP from 2 years ago when Qualcomm moved their SD8G1 to TSMC 4 nm. But to understand AMD's use of Sam 4 nm LPP+, we have to see if it passes the 'back of the envelope' calculation or the 'smell test.' There still may be a reason to worry.

Let's start with the greatest common criticism: Sam 4 nm process 'is trash'. Although the original 4 nm process was not as efficient as TSMC's 4 nm (about 25% less efficient when comparing SD8G1 to SD8+G1), recent evaluations comparing the Exynos 2200 in the Galaxy s23FE (Sam 4 nm LPP+) to the original Exynos 2200 show a 8-20% gain for the new processor even when warm. Since the new Exynos is operating at the same frequencies as the old, these results suggest the new LPP+ revision is up to 20% more power efficient than the old. So far, so good.

Couple that information with a comparison done by Golden Reviewer which spans across leading phone flagships processors across nodes and architectures, we have the ability to further compare and estimate Samsung's new 4 nm LPP+ to TSMC's 4 nm while estimating if Samsung's claims of Exynos 2400 being 70% faster is truthful or not as well as see how this relates to AMD and Zen 5c on Sam 4 nm LPP+.

Golden Reviewer's analysis shows:

a) For same TSMC 4 nm process, Qualcomm/ARM is in trouble with SD8G3 with power efficiency decreasing by 30% to drive the number of transistors required to go from X2->X4 (SD8G1 -> SD8G3) and provide an architecture performance uplift of +27.8% (Perf/GHz). Raw total power increases by 85.5% (6.27 W/3.38W).

b) X2 to X4 architectures are able to sustain a frequency of 3.05 to 3.35 GHz on TSMC 4 nm

Looking at how this affects AMD and it using 4 nm LPP+ for Zen 5c cores, is what frequency can these dense cores obtain on these processes. If Zen 5c cores scale the same to Zen 5 as Zen 4c to Zen 4, then the Zen 5c cores only need to get to about 70% of the large Zen 5 cores. So these dense Zen 5c cores need to get to about 3.5-3.7 GHz.

Starting with what AMD may be giving up on TSMC 4 nm, both the X3 and X4 cores are sustaining 3.3-3.4 GHz at 5-6 W. These sustained frequencies are pretty close to what AMD needs. But the report is for AMD to use Sam 4 nm LPP+. The original X2 Exynos 2200 on Sam 4 nm LPP was 2.8 GHz. The new Exynos 2200 scores are 10-20% higher suggesting frequency gains of up to 20% for an estimated sustained frequency of up to 3.36 GHz. The Dimensity X2 based 9000 does run at almost 3.1 GHz on TSMC's 5 nm so this does suggest there is not a lot of sustained frequency difference between the X2 and X4 architectures on these nodes. This further suggests Sam 4 nm LPP+ should give comparable frequency-power consumption to TSMC's 4/5 nm as was observed with the new vs old Exynos 2200.

These results suggest AMD may very well get sustained frequencies of 3.5-3.6 GHz out of Zen 5c cores on Sam 4 nm LPP+. As a result, there should not be any significant performance-power differences between TSMC 4 nm and Sam 4 nm LPP+ and this will allow AMD to have more Zen 5c based processors on the market - which is what many have been complaining about: there aren't enough AMD based laptops available. If Sam 4 nm LPP+ can take care of the Ryzen 5 & Ryzen 3 'Prometheus' Zen 5c APUs, this will allow TSMC 4 nm to be utilized for Kraken, Strix Point, and Strix Halo and AMD to put more APUs out in 2024 and into 2025. If these Zen 5c cores are all chiplet based and to be utilized in servers, it still frees up more capacity on TSMC 4 nm for AMD to make more, higher performing processors and APUs. Sounds like a win either way

​

​

​