Overclocking

The clock speeds and TDP of Ryzen 9000 series processors look rather restrained on the spec sheets, it's probably because the processors achieve the desired competitiveness for AMD, and the company wanted to leave some overclocking headroom to get the enthusiasts interested, probably also for power consumption reasons.

One such area is memory speed. Ryzen 7000 series processors came with native DDR5-5200 memory support, with DDR5-6000 being the "sweetspot" overclocking speed—the highest stable memory speed you can achieve with a 1:1 parity between the FCLK and MCLK clock domains. There are a couple of updates with Ryzen 9000. The native DDR5 speed has been increased to DDR5-5600, and support has been added for overclocked memory as fast as DDR5-8000. To achieve this, a 1:2 clock divider is engaged between FCLK and MCLK as soon as memory clock is set higher than DDR5-6000. You may manually force a 1:1 divider "and with some luck" you will be able to run DDR5-6400—so basically similar as before. The DDR5-8000 support is being released by AMD as an update to the AGESA microcode, for motherboard vendors to integrate with even their older AMD 600-series chipset products. There should be a new wave of enthusiast PC memory with AMD EXPO profiles for DDR5-8000 and everything under (such as DDR5-7200, DDR5-6800, etc.). In terms of performance it will be interesting to see whether DDR5-8000 1:2 can beat DDR5-6000 1:1 in a wide range of scenarios, which would make it a worthwhile investment.


A side effect of giving processor models such as the 9900X (120 W), 9700X (65 W), and 9600X lower TDP than their predecessors, the 7900X (170 W), and 7700X (105 W), is that Precision Boost Overdrive (PBO) now scoops out more performance, as it raises the package power for the processors. AMD claims performance gains of 6% for the 9900X, and up to 15% for the 9700X (which goes to show that 65 W was too conservative for this SKU).


AMD has also worked on the physical characteristics of the processor's dies' cooling performance, which see a 15% reduction in thermal resistance, resulting in an impressive 7°C reduction in temperature at the same TDP. We asked them to clarify this, and they confirm that no changes have been made to the IHS or the TIM. The improvements are due to better floor planning of the CCD to avoid generating so many hotspots. They also improved placement of the thermal sensor. On Zen 4, the sensor was located pretty far away from the hot spot, so they added quite a lot of thermal margin to account for that. Now the placement is more ideal, which results in a more realistic estimation of the actual temperature in the hottest part of the die. We did see mentions of "100°C" in some materials, but AMD wasn't willing to confirm or deny whether the 95°C thermal target has been adjusted and whether there is a way to now allow a higher thermal limit.


Granite Ridge is a chiplet based processor, just like Raphael. AMD has largely carried over the 6 nm client I/O die (cIOD) from the previous generation, although the substrate has many changes owing to the different dimensions of the 4 nm Zen 5 CCDs, compared to the 5 nm Zen 4 CCDs. We did ask whether the IO die is identical to that on Zen 4, which seems plausible, but AMD wasn't willing to comment on that.


The IHS looks identical to that of Ryzen 7000—no changes here.


AMD set up a manual overclocking session for us to see, where it achieved an impressive 6.90 GHz all-core OC on the 16-core 9950X, with 1.492 V core voltage, and liquid nitrogen cooling. The actual ceiling of the processors is much higher, they mentioned "well over 7 GHz" is in reach on a good sample. In this overclocking session, AMD demoed breaking several world records very trivially—just cool the CPU, give it voltage and clocks and run Cinebench—boom—world record. Given this and the memory OC improvements, we feel there's enough meat on the bone to draw competitive overclockers to these chips. The high core voltage was described as intentionally excessive, to ensure the demos could proceed in a reasonable time, without lots of reboots to find the sweet spot voltage.

Curve Shaper

Curve Shaper is a new feature being introduced with Ryzen 9000 series processors. It's part of the AMD CBS, which means it can be configured in the UEFI setup program, should the motherboard vendor choose to have a GUI there, or using Ryzen Master on Windows. Curve Optimizer enabled PBO and PMFW-aware under-volting, while Curve Shaper lets end-users reshape the underlying voltage curves to improve their undervolting.


AMD showed us an example of an ASRock motherboard's UEFI setup program exposing Curve Shaper as a series of manual values—not very graphical. I made an effort to capture the whole list of settings, ranging from "Min Frequency - Low Temperature," through "Low Frequency," "Med Frequency," "High Frequency" and "Max Frequency," each with three options for "Low Temperature," "Med Temperature" and "High Temperature." We hope that motherboard vendors will implement graphical versions of Curve Shaper that let users visualize the V/F curves based on their manual inputs.