I see what they are doing...
Zen Lite is Jaguar with Steroids and L3 cache
So it will have greater ST performance than BD, but MT wise it will need more power.
Zen-Lite should be Bulldozer-ish with the focus of not being a "Premium" High Performance core. To save design costs; the Zen front-end stays; 64 KB L1i / NNBP / etc. Depending on if they plan to completely restart it. 4-wide decode to two 2-wide clusters like Bulldozer, or two 2-wide decode to two 2-wide clusters like Excavator. Actual core-wise it should be more Jaguar than Excavator. Excavator has the FPU not in the core cluster, while Jaguar and Zen has the FPU in the core cluster. The main change between Zen/Jaguar to Zen-Lite is the fusion of the Integer Rename and Floating Point Rename; Mapper and Non-scheduling Queue becomes one unit. This in turn reduces the wiring from two units to one unit. Primary benefit is the retire queue gets to remain huge with this. It may also resolve dependencies across the LDC faster than Zen/Jaguar.
The CCX stays as the change from SMT to CMT is not visual beyond the L2 cache. It looks like a standard SMT2 core to the L3 caches control and control/data fabrics, and other things.
MCMT and Cluster-based Multithreading has a EPI bonus in every early design. EPI focus means that the architecture can maybe utilize the extended voltage range that FDSOI supports. 22FDX is capable of a larger Vdd range than with 22PDSOI. Extended voltage range means extended frequency range... which means >5 GHz single module boosts can be done. So, add FDSOI self heating and AMD's Ti-states. Ti-states are temperature inversions, it is where the more hot a transistor is the better the device performs. A FDSOI transistor being a hot ah heck has a larger Ti-state range than FinFET transistors do.
They uses that in order to reduce consumption and to make it way smaller.... But what is the intention?
Taking down Atom?
It is a more cost-effective alternative than Banded Kestrel/Raven Ridge-L. Since, it is replacing those, each APU die including the CPU versions have a single 64-bit DDR4/LPDDR4(X) multi-model controller/PHY. What that means is AMD can produce a "Value Threadripper" that competes with Denverton(Goldmont)/Jacobsville(Tremont). These chips would utilize the same standard AM4 socket.
Dual Module CPU + 8 CU GPU => <$83 APU -> 1x 64-bit DDR4
Quad Module CPU => <$75 CPU -> 1x 64-bit DDR4
2x Quad Module CPU => <$170 CPU -> 2x 64-bit DDR4
... for example.
If we use the historic Athlon AM1 for reference though...
Dual Module CPU + 8 CU GPU => <$54 APU -> 1x 64-bit DDR4
Quad Module CPU => <$51 CPU -> 1x 64-bit DDR4
2x Quad Module CPU => <$97 CPU -> 2x 64-bit DDR4
So, instead of a small 2 Zen core/ 3 GCN die that costs $xx. AMD sells a medium sized 4 Zen-lite/ 8 GCN die that costs $xx. $xx being the same between the two, etc. The larger SKU has more utility for the mainstream than the smaller SKU.