News Asus Confirms Intel DLVR Was Fused Off in Raptor Lake

DLVR had the potential of reducing Raptor Lake CPU power consumption by up to 25%, if Intel had enabled the feature.

p78BEBqZAeF6QoPWT4edSN.png


A 20% power reduction would put it at 226.4w, the same as a Ryzen 7950X...
 
  • Like
Reactions: artk2219

cyrusfox

Distinguished
A 20% power reduction would put it at 226.4w, the same as a Ryzen 7950X...
From hothardware.com article it states it is only that effective at Sub 40Amp(Sub 40 W essentially). Gains disappear at higher amperage.

Quote:
However, those power savings were primarily during lower-load situations under 40A. If you're familiar with modern CPUs, 40A is usually around 40 watts. Intel's recent desktop CPUs spend most of their time well above 40 watts, and the savings from this design (by Intel's own reckoning and admission) fell off sharply past that point.
 

JamesJones44

Reputable
Jan 22, 2021
852
782
5,760
From hothardware.com article it states it is only that effective at Sub 40Amp(Sub 40 W essentially). Gains disappear at higher amperage.

Quote:
However, those power savings were primarily during lower-load situations under 40A. If you're familiar with modern CPUs, 40A is usually around 40 watts. Intel's recent desktop CPUs spend most of their time well above 40 watts, and the savings from this design (by Intel's own reckoning and admission) fell off sharply past that point.

This article seems to indicate that their estaminets include going above 40A in normal cycles.
Once the CPU current consumption exceeds 40A, some of the D-LVR power gates activate to supply the additional current to the Vin. For example, at 45A, the motherboard VR sends 40A, while the D-LVR power gates administers the extra 5A. The D-LVR power gates consume very little energy, assuming that the duration and duty cycle above 40A is short. With Intel's example, we can expect a 21% decrease in CPU power, which translates to roughly to a 7% performance uplift.

Buy yeah, in your Blender, Cinebench, Geekbench, etc. it probably doesn't make as big of a difference. However, I don't like those tests as they do not replicate normal workloads, just worst case scenarios which doesn't really give a clear picture (good or bad) of how well the CPU will work for a build, but I degrees.