What I learned when exchanging my old Kaveri for a Zen+
Here is the overdue write-up for my CPU upgrade from Kaveri A10-7850K (2M/4T) to Zen+ R7 2700 (8C/16T) . Original plan was to get a Zen2, but what can you do if AMD doesn't feel like releasing B550 boards?

The R7 2700 is rated for 3.2GHz base and 4.1GHz turbo.
I poked around and found that 3.4GHz base -25mV is achievable.
If I skip undervolting, 3.6GHz tested successfully and I wouldn't be surprised if the CPU also pulls off 3.8-4.0GHz...
Problem 1: When am I gonna need 16 Threads?
Problem 2: Once you play around with the multiplicator, you lose the Turbo.

So I decided to stick with Stock settings (hue hue) and slapped a -100mV undervolt on it.
Bring on ze benches. Brace for _slight_ differences

Games [7850K -> R7 2700]
Metro Last Light Redux: 65 -> 75
Rise of the Tomb Raider: 44.2 -> 55.5 @DX11 | 55.7 @DX12
Shadow of the Tomb Raider 39 -> 68 @DX11 | 92 @DX12
Far Cry 5: 43 -> 41
Company of Heroes 2: 40.65 -> 54.46
DÖÖM: 130-140 -> 200
GTA V: 61.6 -> 115.8
Strange Brigade: 104 -> 132 (Vulkan just works)
Natural Selection 2: 96 -> 117

Annotations for Games
RotTR: Eidos completely ♥♥♥♥♥♥ up on their DX12 implementation. You get graphical artifacts. No cigar.
FarCry5: I detected an anomaly here. The 2nd run is somehow much faster at 66 fps. Got no clue how well the 7850K would do on its 2nd run...
Dööm: 200 seem to be a hard limit, it doesn't go any faster. Vulkan FTW!
GTA5: As the Pr0B3nch0r that I am, I tested GTA5 with vsync on for years
NS2: I learned that there was a major engine update in September, in order to actually compare results, you need to set Atmospherics to Low, as it previously could only be turned on or off)

Artificial Benchmarks [7850K -> R7 2700]
Unigine Heaven: 76 -> 80.3
Unigine Valley: 64.7 -> 77.9
CPU-Z single/multi: 255 / 848 -> 400 / 4167
Winrar: 4049 kB/s -> 13783 kB/s
7zip (32MB dictionary): 17654 MiPS -> 61398 MiPS
IntelBurnTest (standard): 40.4 sec / 22.08 GFlops -> 11.8 sec / 74.96 GFlops

Power Draw
Idle: 71 W -> 77 W
VLC@1080p HEVC: 82 W -> 91 W
Shadow of ze TR: 293 W -> 321 W(-25mV) | 331 W@DX12(-25mV) | 315 W@DX12(-100mV)
Frames per Second: 39 fps -> 68 fps(-25mV) | 92 fps@DX12(-25mV) | 93 fps@DX12(-100mV)

Full Power Hog
For fun I went full power hog by removing all UV from CPU and GPU. This is worst case:
Shadow of ze TR: 341 W, 86 fps
Yep, it's hard to believe.
Without UV, the system draws 26W more while producing 7 frames less .
Morale: A 450W PSU is plenty enough to power 2700 + Vega56.

Here is the normalized Efficiency rating Polaris (RX580UV+OC) vs Vega (Vega56):
Polaris = 100%
Vega56@Bios1 (180W Board Power) = 125.4%
Vega56@Bios2 (165W Board Power) = 131.9%
Vega56@Bios2+UV+HBMOC = 143,7%

Of course, you can raise your Power Target and let your Vega56 go full power hog.
The card can easily gobble up more than 100W extra, maxing out somewhere north of the abyssmal Vega64. In that case, efficiency tanks and drops way below Polaris.