F5F Stay Refreshed Hardware Desktop Discover Curve Shaper and dive into the 79503XD experience.

Discover Curve Shaper and dive into the 79503XD experience.

Discover Curve Shaper and dive into the 79503XD experience.

L
LeooGamer_YT
Junior Member
35
01-14-2023, 12:25 PM
#1
I'm curious if anyone has experimented with the Curve Shaper on this forum. With the AMD 9000 they introduced this new option, and it also applies to my 7950X3D running the latest BIOS. I'm using an Asus ROG Strix X670E-E with BIOS 2403, AGESA 1.2.0.2. When I used Curve Optimizer, my 5900X achieved close to -30 across most cores, but the 7950X3D performed poorly, ranging from -8 to -28. After recently applying all updates, I found that a more aggressive setting helped under stress, though it caused issues when idle. I realized adjusting the Curve Optimizer further might be better. The idea is that Curve Shaper lets you fine-tune the V/F curve to closely match your chip's characteristics, with 15 points on the curve. Once you identify the best shape, adding Curve Optimizer can account for per-core differences. BIOS lists options but they're confusing. In practice, my machine stays around 35°C and 3.5GHz during idle (low + low/med temp). Using OCCT with an SSE workload, CCD0 (Vcache) reaches about 65°C at 5.0GHz. For medium/high frequencies, it's around 70°C and 5.4GHz. If I run an AVX2 test, the frequency drops roughly by 200MHz. Ignoring Curve Shaper for idle, the focus should be on narrowing down to the four main frequency-medium temperature ranges. It might help to explore the Max frequency section too, though my chip doesn't reach those extremes. I'm aware that Scatterbencher has done a video on this topic, but it seems like a thorough search is needed due to the complexity. From that video, we learn the temperature effect on the V/F curve is minimal—perhaps just a few steps if any. I'm wondering if others have insights on optimizing this efficiently. Usually, I restart with two extra points and check for stability during daily tasks. In my experience, Curve Optimizer tends to struggle more in idle compared to load testing. Any thoughts or experiences with improving this process would be appreciated!
L
LeooGamer_YT
01-14-2023, 12:25 PM #1

I'm curious if anyone has experimented with the Curve Shaper on this forum. With the AMD 9000 they introduced this new option, and it also applies to my 7950X3D running the latest BIOS. I'm using an Asus ROG Strix X670E-E with BIOS 2403, AGESA 1.2.0.2. When I used Curve Optimizer, my 5900X achieved close to -30 across most cores, but the 7950X3D performed poorly, ranging from -8 to -28. After recently applying all updates, I found that a more aggressive setting helped under stress, though it caused issues when idle. I realized adjusting the Curve Optimizer further might be better. The idea is that Curve Shaper lets you fine-tune the V/F curve to closely match your chip's characteristics, with 15 points on the curve. Once you identify the best shape, adding Curve Optimizer can account for per-core differences. BIOS lists options but they're confusing. In practice, my machine stays around 35°C and 3.5GHz during idle (low + low/med temp). Using OCCT with an SSE workload, CCD0 (Vcache) reaches about 65°C at 5.0GHz. For medium/high frequencies, it's around 70°C and 5.4GHz. If I run an AVX2 test, the frequency drops roughly by 200MHz. Ignoring Curve Shaper for idle, the focus should be on narrowing down to the four main frequency-medium temperature ranges. It might help to explore the Max frequency section too, though my chip doesn't reach those extremes. I'm aware that Scatterbencher has done a video on this topic, but it seems like a thorough search is needed due to the complexity. From that video, we learn the temperature effect on the V/F curve is minimal—perhaps just a few steps if any. I'm wondering if others have insights on optimizing this efficiently. Usually, I restart with two extra points and check for stability during daily tasks. In my experience, Curve Optimizer tends to struggle more in idle compared to load testing. Any thoughts or experiences with improving this process would be appreciated!