- Joined
- Jul 29, 2023
- Messages
- 48 (0.08/day)
| System Name | Shirakami |
|---|---|
| Processor | 7800X3D / 2000 & 2000 IF & UCLK |
| Motherboard | Gigabyte B650i AORUS Ultra |
| Cooling | Corsair iCUE H100i ELITE CAPELLIX w/ ultra slim 2x120mm fans |
| Memory | 2x24GB Hynix M-Die @ 8000 - 38-48-48-40 - 40k tREFI & tuned subtimings |
| Video Card(s) | 6900 XT Reference / -120 mV @ 2.4 GHz |
| Storage | 1 & 2 TB NVMe - 1 TB SATA SSD |
| Display(s) | LG 34GN850 (3440x1440) 160 Hz overclock |
| Case | Lian Li Q58 |
| VR HMD | Reverb G2 V2 |
| Software | Fedora Linux 41 operating system (daily) |
Hi everyone! I recently heard that the efficiency cores have their power scale with the frequency they're set to. I couldn't find any data online for the power-to-frequency scaling, so I decided to test it myself.
I started at 800 MHz on the efficiency cores and moved up in increments of 800.
(800 > 1600 > 2400 > 3200 > 4000 > 4500 (maximum stable)).
CPU-Z was the used stress-test since it's simple, quick, and effective.
The CPU is a 13700k set to 1.40 v & receives 1.350 v during load in Cinebench.
I had forgotten there was a 200 W limit but it did not throttle for the first two tests, and was lifted for the others.
IA Cores Power is 180 W @ 800 MHz, going all the way up to 210 W @ 4500 MHz. My conclusion is that once you surpass 3200 MHz (or that frequency area, as I didn't do 100 MHz increments), the wattage begins to nearly double. Whereas the increments were 6 watts per 800 MHz step at 3200 MHz and prior, it begins to become 10 to 12 watts. There may be some type of built-in protection for the efficiency cores past a certain wattage, since after 4000 Mhz, the change in power is pretty much margin of error.
If you're tied between leaving them on for MT performance but incapable of handling the heat output (which is understandable, they soak the IHS really fast), my recommendation would be setting them to 1600 to 2400 MHz. I believe the 13900k would really benefit from this the most, or those of us with lower-end coolers.
As an additional note:
R23 pushes 265 W @ 4.2 GHz e-core
242 W @ 1.6 GHz e-core
Due to this, I can actually reduce my target load voltage by 0.020 mV, reducing power even further. Going from 1.35 load to 1.33.
I started at 800 MHz on the efficiency cores and moved up in increments of 800.
(800 > 1600 > 2400 > 3200 > 4000 > 4500 (maximum stable)).
CPU-Z was the used stress-test since it's simple, quick, and effective.
The CPU is a 13700k set to 1.40 v & receives 1.350 v during load in Cinebench.
I had forgotten there was a 200 W limit but it did not throttle for the first two tests, and was lifted for the others.
IA Cores Power is 180 W @ 800 MHz, going all the way up to 210 W @ 4500 MHz. My conclusion is that once you surpass 3200 MHz (or that frequency area, as I didn't do 100 MHz increments), the wattage begins to nearly double. Whereas the increments were 6 watts per 800 MHz step at 3200 MHz and prior, it begins to become 10 to 12 watts. There may be some type of built-in protection for the efficiency cores past a certain wattage, since after 4000 Mhz, the change in power is pretty much margin of error.
If you're tied between leaving them on for MT performance but incapable of handling the heat output (which is understandable, they soak the IHS really fast), my recommendation would be setting them to 1600 to 2400 MHz. I believe the 13900k would really benefit from this the most, or those of us with lower-end coolers.
As an additional note:
R23 pushes 265 W @ 4.2 GHz e-core
242 W @ 1.6 GHz e-core
Due to this, I can actually reduce my target load voltage by 0.020 mV, reducing power even further. Going from 1.35 load to 1.33.
Attachments
-
800 MHz.PNG -
1600 MHz.PNG -
2400 MHz.PNG -
3200 MHz.PNG -
4000 MHz.PNG -
4500 MHz.PNG
Last edited: