Zen 5 Ryzen 9000 Power Efficiency and Thermals
AMD says Ryzen 9000 series runs at much lower temperatures than its predecessors, helping keep frequency residency high (better effective frequency and longer boost durations). This is due to a 15% improvement in thermal resistance, which allows the chip to operate at a 7C lower temperature than a Zen 4 chip running at the same TDP.
As you can see in the second slide above, Ryzen 9000’s improved thermal performance allowed dialing back the TDP for three of the four new desktop chips, all while delivering substantial performance improvements in the types of heavy multi-threaded workloads that push the chip to its power limits. AMD used the Blender benchmark to highlight the improved performance in heavy workloads.
Some of the power improvements undoubtedly result from moving from the 5nm node to N4P, but AMD also made other targeted optimizations to wring out more performance. AMD’s engineers said hotspots were a challenge with the Zen 4 design, so they reduced the number of hotspots and better dispersed the remaining hot spots on the compute die. The engineers also fixed Zen 4’s ‘suboptimal’ placement of the temperature sensors spread across the die. Paired with a more intelligent firmware control algorithm, the better placement of the thermal sensors enables more accurate decisions for tighter control of the available temperature margin.
AMD says the result is a pronounced reduction in power usage that allows the system to run cooler and, ultimately, quieter than the previous-gen models, not to mention Intel's competing Raptor Lake Refresh.
New 800 Series Chipsets and Overclocking
AMD’s original AM4 platform is legendary for its longevity, supporting all versions of Ryzen since the lineup’s debut in 2017 (and even the Bristol Ridge series before). It now encompasses 145 CPU and APU models — and that list continues to grow.
AMD says we can expect AM5 to last through 2027+, as it's also engineered for the long run. AMD already unveiled its top-tier X870 and X870E chipsets at Computex 2024, with the most notable additions being the mandatory addition of USB4 and PCIe 5.0 for both graphics and storage.
AMD is now announcing the value-oriented B-series chipset with a new tier in addition to the standard B850 we’d expect. The new B840 motherboards land between the bargain-basement A-series motherboards and the standard fully overclockable B850. There are several differences between the two tiers. B840 fully supports memory overclocking but does not support CPU overclocking like the higher-tier models. The B840 motherboards will also only have a PCIe 3.0 x16 connection — a big reduction from the PCIe 5.0 and 4.0 connections available on the B850 boards.
Both the B850 and B840 support USB 3.2 instead of the mandatory USB 4 present on the X-series motherboards. Even though AMD’s nagging AM5 pricing issues have largely receded, the B840 alterations create a less expensive B-series board that will provide more pricing flexibility, particularly for OEMs.
Unfortunately, AMD’s 800-series motherboards will not be available on the market when the Ryzen 9000 processors launch at the end of the month. AMD says that the various motherboard vendors will release on their own schedules, and our early indications are that the first motherboards will arrive at the end of September, well after the Ryzen 9000 launch date. The Ryzen 9000 chips are fully compatible with existing AM5 motherboards (after the requisite BIOS update), so there are options for early adopters. However, the staggered launch is obviously not ideal.
AMD has also made several enhancements on the overclocking front. The new 800-series AM5 boards will support higher memory overclocking ceilings than existing motherboards. AMD says the 800-series can support memory overclocking up to DDR5-8000. Ryzen 9000 also adds real-time memory overclocking, allowing you to change memory speeds and timings in real time from within the Ryzen Master application.
AMD also has a new Memory Optimized Performance Profile that allows either manual or automatic switching between the JEDEC and EXPO overclocking profiles in real time to prioritize either bandwidth or latency (timings) for the type of application that’s running. You can toggle between the different profiles in the Ryzen Master utility or enable automatic switching. The details behind this mechanism aren’t entirely clear yet, but we’ll learn more when the chips arrive for testing.
AMD’s existing Curve Optimizer feature allows you to set one negative or positive offset for the entire voltage/frequency curve, which the processor then applies across the full curve. The new Curve Shaper feature provides much more granular control by allowing adjustments for three temperature and five frequency points, resulting in 15 points of control for the user. That’s a major increase over the single point of control allowed with the existing mechanism.
AMD’s Ryzen Master already has an option for an automated test that can assign the Curve Optimizer offset. This will continue to work, but it will not offer the full 15 points of optimization — the Curve Shaper requires manual adjustment. However, the two features can be used in tandem, with the user assigning additional Curve Shaper points in addition to the single Curve Optimizer value (Curve Shaper acts as a sort of global modifier that will adjust all the assigned ranges).
The auto-overclocking Precision Boost Overdrive (PBO) also returns, and AMD says Ryzen 9000’s lower TDP range enables more headroom for PBO gains. The company points to a 6% to 15% improvement in multi-threaded Cinebench performance with PBO engaged for the Ryzen 5, 7, and 9 processors but didn’t include the Ryzen 9 9950X in its PBO examples (likely a low single-digit percentage of gain). We’ll put all these new overclocking knobs to the test in our reviews.
On the following page, we will move on to the Ryzen AI 300 benchmarks and then on to the CPU, GPU, and NPU architectures.
