Friday, November 15th 2024

Intel Removes DLVR Bypass for "Arrow Lake" in Latest 0x112 Microcode Update

Intel has significantly changed its latest 0x112 microcode update, removing users' ability to bypass the Digital Linear Voltage Regulator (DLVR) through standard BIOS settings on "Arrow Lake" processors. DLVR, a technology designed to provide precise voltage control for individual performance cores and efficiency core clusters, offers great benefits during gaming sessions and light workloads. According to overclocker der8auer's analysis, DLVR can effectively manage power consumption during gaming, with power losses of around 20 W at typical gaming loads. However, these losses can shoot up to approximately 88 W under full CPU utilization. Previously, users could disable DLVR through a BIOS setting called "Power Gate" mode, which is particularly useful for intensive workloads where power losses might impact performance. With the new microcode update, this option has been removed from standard BIOS settings. It is also worth pointing out that DLVR is in its second iteration inside Arrow Lake CPUs, after the initial debut in "Raptor Lake," which had DLVR fused off.

Intel explained to Hardwareluxx that this change was implemented to "prevent accidental misuse of DLVR bypass," restricting its use to extreme overclocking scenarios involving sub-ambient cooling methods like liquid nitrogen. The update has already been rolled out through BIOS updates on some Z890 chipset motherboards, with ASRock and MSI being among the first manufacturers to implement the new microcode. While DLVR bypass may still be accessible through specialized LN2 profiles on high-end motherboards, the average enthusiast user loses direct control over this feature. This development mainly affects early Arrow Lake adopters, as not all motherboards include extreme overclocking profiles. While Intel's move appears to prevent potential issues, we must remember that power settings are something that users should only change with plenty of consideration. Removing this power gate mod is Intel prevention for Raptor Lake-like situations where these chips had an issue with Vmin shift.
Sources: HardwareLuxx.de, via Tom's Hardware
Add your own comment

10 Comments on Intel Removes DLVR Bypass for "Arrow Lake" in Latest 0x112 Microcode Update

#1
Hyderz
Tpu should retest after this update
Posted on Reply
#2
Ed_1
HyderzTpu should retest after this update
Huh, What to retest?
Default is on and I am sure TPU tested with it on, you can still OC but I think it would be better to just have a warning message and let users have the option.
Performance improvements will come later (end on month or early next month). Then be good time to retest.
Posted on Reply
#3
Hyderz
Ed_1Huh, What to retest?
Default is on and I am sure TPU tested with it on, you can still OC but I think it would be better to just have a warning message and let users have the option.
Performance improvements will come later (end on month or early next month). Then be good time to retest.
Cool! Didn’t realize that by default it’s on
Posted on Reply
#4
Daven
To all you retesters, there are enough processor updates, processor releases and processor what if scenarios throughout the year to have an occasional new review. This year alone there have been 12 different review days for processors. That's an average of one per month. And since W1zzard retests all recent processor generations every time under the new conditions, this is effectively a retest using the latest STABLE bioses, windows updates, microcode updates, drivers, etc.

Just wait until the next planned review.
Posted on Reply
#5
N/A
If at 250 watts load there is a power loss of 88 watts to step down or convert the input 1,5 volts to 1,2 volts that is extremely insanely wasteful. We need real measurements at the 8 pin and 24 pin connector to see if that's really happening.
Posted on Reply
#6
DavidC1
N/AIf at 250 watts load there is a power loss of 88 watts to step down or convert the input 1,5 volts to 1,2 volts that is extremely insanely wasteful. We need real measurements at the 8 pin and 24 pin connector to see if that's really happening.
Of course. DLVR stands for "Digital Linear Voltage Regulator". Even though it's a Low Dropout regular it's still linear so it loses power.

And the idea for DLVR is to save power during bursty scenarios, which is perfect for laptops, but not so good for load, because DLVR adds multiple parallel regulators to reduce voltage droop.
Posted on Reply
#7
:D:D
N/AWe need real measurements at the 8 pin
Agree, that would be nice even though efficiency loss of the mainboard VRM.

Maybe I missed something but don't know where 1.1V came from and calculating 220A presumably from 240W with it! Sensor tables only showed around 160A IIRC. So if 220A at 1.1V then package power at 1.5V is at least 330W?

Also I would think bypassing the DLVR would mean all DLVR gates turned on and still in circuit.
Posted on Reply
#8
N/A
:D:DAgree, that would be nice even though efficiency loss of the mainboard VRM.

Maybe I missed something but don't know where 1.1V came from and calculating 220A presumably from 240W with it! Sensor tables only showed around 160A IIRC. So if 220A at 1.1V then package power at 1.5V is at least 330W?
the 220A value is likely based on a lighter load where the CPU can boost higher. therefore more amps are needed. whatever the case the CPU will always be 1.5V externally and lowering the voltage does nothing to save power, only core temperature while power loss is generated elsewhere in the die.
Posted on Reply
#9
chrcoluk
HyderzTpu should retest after this update
Only if they bypassed it, it wont affect default configuration.
Posted on Reply
#10
:D:D
N/Athe 220A value is likely based on a lighter load where the CPU can boost higher. therefore more amps are needed. whatever the case the CPU will always be 1.5V externally and lowering the voltage does nothing to save power, only core temperature while power loss is generated elsewhere in the die.
Having a second look shows the 1.5V was probably based on max SVID but this only seems to happen when switching to light loads. R23 shows around 1.4V for the most part while running R23, probably loadline adjusted.
10:34
the problem with dlvr is the high load scenario for example running cinam R23 as we just did and that's also the example that I took so you can see in this example the CPU is taken with a 240 watt that is exactly what we saw and we are assuming an input voltage from the vrm to the CPU that could be 1.5 volt and internally the CPU would break down this from 1.5 to 1.1
He is using the CPU Package power but DVLR's do not supply the whole package, SVID (VRM) power is around 220W so current is around 160A not 220A


If SVID was always at 1.5V then as a crude example demonstrating dynamic power

1.5V in
1.4V Cores, dynamic power 280W = 200A so DLVR at 0.1V = 20W
Total SVID power 300W

Drop core voltage to 1V same clock
1.5V in
1.0V Cores, dynamic power ~140W = 140A so DLVR at 0.5V = 70W
Total SVID power ~210W

However such a huge change in core voltage might be done to reduce clocks also which would reduce power even further.

Note that he later changed SVID 1.5V
21:07
if we go all the way down to the voltages the only thing that we changed is now getting regulation mode under the dlvr the performance core voltage is set to 1.28 and the vrm core input voltage lowered to 1.38
21:45
scroll again down to Asus EC we see that Pcore voltage is slightly higher than stock stock was 1.18 now we are at 1.22 so that's slightly increased we decreased the input voltage and this allows us to now only have 190 to about 200 watt under load whereas with stock we saw about yeah 220 to 230 so I would say that's like a 10 to 15 watt decrease by simply increasing the Vcore and decreasing the input voltage
I would call that a 30W decrease but whatever.

Seems like a tweakers paradise :D
Posted on Reply
Dec 11th, 2024 20:31 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts