AMD Ryzen 5000 Mobile 'Cezanne' SoC Deep Dive: Zen 3 Powers Into Notebooks

AMD whipped the covers off of its Ryzen 5000 'Cezanne' mobile chips at CES 2021, but today marks the official launch and AMD has unveiled the deep-dive details of the SoC that brings the Zen 3 architecture to the notebook market for the first time. The new chips will power a large range of notebooks, like the Asus ROG Flow X13 that we reviewed today, and finally bring Ryzen to the highest-end gaming notebooks to compete with Intel's Tiger Lake chips. 

On the surface, the Ryzen 5000 chips look like an iterative update to existing Ryzen 4000 Renoir processors, but there's a surprising amount of ingenuity under the hood. AMD’s Ryzen 5000 Mobile chips bring the 7nm process paired with Zen 3’s 19% IPC increase to notebooks for the first time, largely by replacing the Zen 2 CPU cores with faster Zen 3 cores while leveraging much of the Ryzen 4000 design (like the 7nm Vega graphics cores and existing I/O accommodations) to minimize time to market. 

And time is of the essence. For now, Intel will fend off the Ryzen 5000 chips with its Tiger Lake lineup that's currently capped at four cores and eight threads with its 35W H35 series chips. Intel says its long-awaited eight-core 45W H-series chips are still in development, with headline specs including a 5.0 GHz boost on multiple CPU cores, but the company's first eight-core 10nm processors won't come to market until later this quarter.

Meanwhile, AMD is shipping Ryzen 5000 and has wrung out extra performance through clever modifications to its existing Ryzen 4000 SoC design, which we'll cover below, culminating in a 23% overall improvement in single-threaded work and a 17% increase in multi-threaded performance for the 5000 series. 

AMD says the improvements open up new levels of performance and set a new standard for battery life in x86 notebooks, with up to a 20-hour increase in standby life and an additional two hours during general productivity work, all while remaining the only 8-core x86 chips for ultrathin laptops.

The Ryzen 5000 Mobile series looks promising. Provided that AMD can provide sufficient quantities, they could mark yet another escalation in the increasingly-competitive notebook market. Let's take a closer look. 

The 13 new processors span from low-power 15W chips up to two new overclockable 45W+ HX-series models designed to bring desktop PC-like gaming performance to notebooks. The Ryzen 5000 mobile processors all come with threading enabled, the 7nm Vega graphics engine with higher graphics boost clocks than the prior-gen models, support CPPC (Collaborative Power and Performance Control) technology, which we’ll dive into shortly, and have higher CPU boost clocks than the previous-gen. 

As before, the H-series models are designed for notebooks that will use discrete graphics. The two 45W+ eight-core HX models carve out a new high-performance niche by bringing CPU, memory, and fabric overclocking to AMD-powered notebooks for the first time, but overclocking headroom will largely be dictated by the thermal and power characteristics of each notebook. Naturally, bulkier notebooks with more robust cooling and power delivery will unlock better overclockability.

The two H models slot in with eight- and six-core variants and a 45W TDP rating, with the former having eight CUs that boost to 2.0 GHz, while the latter has seven CUs that stretch up to 1.8 GHz.  

AMD also expanded its HS series with four chips with boost clocks that reach up to 4.8 GHz within the 35W TDP envelope. AMD segments the HS stack with three eight-core models with varying base and boost clocks, but these models have lower base clocks than the H-series models to accommodate the 35W TDP envelope. AMD also has a lone six-core twelve-thread model to round out the bottom of the H-Series stack. AMD also segments the HS models with either seven or eight Vega CUs, with peak boost clocks weighing in at 2.1 GHz.

The 15W U-Series models slot in for thin and light devices and will often lean on the integrated graphics units. AMD recently chose to unify its Ryzen Mobile branding under the same Ryzen 5000 umbrella as its desktop chips to clear up the confusion with the Ryzen 4000 series processors that came with an older architecture than desktop Ryzen 3000 models. 

However, AMD also sprinkled in three Zen 2 'Lucienne' chips in the Ryzen 5000 Mobile stack, muddying the waters. AMD says this approach meets specific pricing criteria and customer (OEM) demand on the lower end of its product stack. These Zen 2-powered Ryzen 3, 5, and 7 models slot into the lowest-end 15W U-series category.

The Zen 2 variants come with the same design as their predecessors, but again, the targeted enhancements to the SoC (all of the same modifications listed below apply) and increased clock rates result in higher performance.

The Ryzen 7, 5, and 3 families also include one Zen 3 model apiece with either eight cores and 16 threads, or four cores and eight threads. Unlike the previous-gen Ryzen 4000 chips, all of the 15W models come threading enabled. 

AMD Ryzen 5000 Mobile Architecture Deep Dive - Design Goals

Image 1 of 2

AMD’s performance claims for the Ryzen 5000 chips are quite impressive, with a 23% generational improvement in single-threaded performance being the highlight feature. The increased single-threaded performance over the Zen 2-powered Ryzen 4000 chips stems from the 19% IPC improvement borne of the Zen 3 microarchitecture and more fine-grained SoC-level optimizations that optimize power delivery, among other factors. AMD also claims two additional hours of battery life and a 108% advantage in multi-threaded work over Intel’s 11th-gen mobile processors.

AMD’s design goals focused on three key areas: Performance (particularly in IPC and per-core performance), latency improvements that come via the unified L3 cache and eight-core CCX, and improved power efficiency.

Zen 3 Unification

Image 1 of 2

The Ryzen 5000 Mobile chips come with all the hallmarks of the Zen 3 microarchitecture, which you can read more about here. Like the desktop chips, AMD increased the L3 cache to 16MB over the prior gen’s 8MB and unified the eight cores and cache into one contiguous cluster inside the CCX (Core Complex). In contrast, Zen 2 had two four-core clusters, each with 4MB of cache. 

This new arrangement improves both core-to-cache and core-to-core latency. For highly-threaded applications, this new design imparts a 2X cache increase, and lightly-threaded workloads now have access to a full 16MB of cache, which is equivalent to a 4X increase in directly-accessible cache.

Ryzen 5000’s monolithic die, meaning it is one large die instead of the multi-die arrangements on desktop chips, allows for tighter control of power efficiency. The company also improved performance-per-watt (power efficiency) by targeting the highest-end of the frequency/voltage curve it could maintain while controlling thermal density. That's a key consideration for thin-and-light devices with limited cooling capability. By sustaining the highest clocks possible while the chip is still in the thermal and power efficiency sweet spot, the company increased single-threaded performance beyond Zen 3’s 19% IPC uplift to a 23% improvement in per-core performance.

The Ryzen 5000 Mobile Die

Image 1 of 2

In many respects, AMD simply replaced the Zen 2 cores with Zen 3 cores, leaving the rest of the design untouched. AMD’s re-use scheme allowed the company to accelerate time to market; AMD introduced the mobile variants a mere 90 days after the desktop models, which is the fastest transition of a new architecture to mobile in the company’s history (it typically takes a year). 

The company’s forward-thinking modular design eased the process: AMD planned for this eventuality when it designed the Ryzen 4000 SoC. AMD also preserved the same pinout (BGA mounting scheme) for the Ryzen 5000 processors, meaning they are backward compatible with existing designs. This also allows OEMs to use the same motherboards and componentry for both existing and future products, speeds the design of new laptops, and allows OEMs to leverage existing supply chains. However, while these design elements largely remain the same, AMD fine-tuned the various elements to extract more performance, which we'll dive into below.

Although it doesn’t equate directly to the physical layout on the die, the topology in the second slide shows us the design's basic outlay. Here we can see the eight threaded cores clustered around the central 16MB slice of L3 cache, two DDR4-3200 / LPDDR4x-4266 controllers, and the 7nm Vega GPU with 1MB of L2 cache and eight CUs. 

The SoC also has a host of other fixed-function acceleration blocks for multimedia, the System Management Unit (SMU) for power delivery control, Fusion Controller Hub (FCH) for external sensors, audio engine, Platform Security Processor (which comes into play in the Pro models), and the usual external interfaces, like NVMe and three flavors of USB (Type-C, 3.1, 2.0).

As before, the chips have 16 lanes of PCIe 3.0 connectivity, meaning Cezanne will not have PCIe 4.0 support. AMD allocates eight lanes to the graphics, while the remaining eight lanes are split between two NVMe interfaces for storage. 

The Ryzen 5000 mobile die spreads 10.7 billion 7nm TSMC transistors across 180mm2 of silicon, with the core complexes, L3 cache, and GPU in the center of the die, while the I/O componentry lines the periphery. This is slightly larger (~15%) than Renoir's 156mm2 die and comes with 900 million more transistors (the larger caches contribute to the higher transistor count). 

Ryzen 5000 Mobile Memory Controller Enhancements

AMD’s engineers turned their eye to optimizing the memory PHY to reduce the overall SoC power draw. With Ryzen 4000, the PHY remained fully ‘on’ if the SoC wasn’t in a standby power state, consuming a significant amount of power draw. Ryzen 5000 introduces a deep power state that the SMU engages when the PHY is in a lower-activity state (but not idle). This shifts the digital portion of the PHY into a lower power state, bypassing it to reduce supply voltage. This technique engages a new low-dropout (LDO) power regulator and turns off the primary on-chip voltage regulator for the memory PHY during low usage. This approach reduces power consumption from the PHY and the power delivery network as well.

Of course, minimizing latency when the controller shifts back to its full-performance state is paramount. AMD leverages the data fabric’s hardware-assisted fast entry/exit mechanisms to shift into different power states, thus lowering latency.

Each memory controller supports a single 64b memory channel or dual 32b virtual channels. Each controller supports DDR4-3200 in single-channel (1DPC SR/DR) or two channels of LPDDR4x-4266. 

Quad-channel LPDDR4x provides 33% more theoretical bandwidth than DDR4-3200 and has twice the density of standard DDR4, thus improving maximum capacity to 32GB of memory with a two-channel die.  

Vega Rides Again

Image 1 of 2

As before, the Vega graphics unit comes with a maximum of 8 compute units (CU) and 1MB of L2 cache, but AMD increased the peak frequency by 20% (up to 2.1 GHz).

AMD fine-tuned the power management mechanisms, like the on-chip voltage regulator, to improve sustained performance. AMD also reduced Vmin (the minimum voltage required for operation) to improve power efficiency (performance-per-watt). These improvements enable higher-frequency operation in tandem with improved frequency-per-watt throughout the entire voltage/frequency range. That frees up additional power that's used to sustain higher frequencies in either the CPU or GPU. 

AMD’s new intelligent graphics Dynamic Energy Management (DEM) functionality steps in to detect which workloads are impacted heavily by graphics frequency, allowing the GPU to operate at lower frequencies when increasing the frequency wouldn’t result in tangible performance gains (for example, when the workload is primarily memory bound).

Ryzen 5000 uses the latest revision of the 7nm process, which has less leakage. It also marks the debut of per-core CPU voltage regulation to the Ryzen Mobile lineup, which reduces overall power consumption during gaming workloads. The saved power savings from these two factors can be directed to the GPU to boost gaming performance.

AMD also optimized its graphics drivers for enhanced memory management and tweaked anti-aliasing and other functions, with peak performance weighing in at 2.15 TFLOPS (FP32) and 4.30 TFLOPS (FP16). AMD used benchmark results from Timespy, a CPU intensive workload, to quantify the performance improvement, noting a 171% increase over the first-gen Ryzen processors. Notably, the Ryzen 5000 Vega iGPU is 15% faster than Ryzen 4000.  

AMD carries all the security features present in the 4000 series to the 5000 series, with the addition of Control-flow Enforcement Technology (CET) that eliminates control flow attacks that hijack legitimate applications to run malicious code. As before, the chips also have other security features that serve as the bedrock of the Pro Security suite for Ryzen Pro processors. 

Ryzen 5000 Mobile, Boosting Boosts

Image 1 of 3

AMD’s addition of its Collaborative Power Performance and Control (CPPC) interface was truly a watershed breakthrough for its Ryzen 3000 desktop chips. This tech tells the operating system which cores are the fastest, and the Windows scheduler then targets lightly-threaded workloads at those cores. This allows AMD to bin its chips into higher speed grades based on the fastest available cores on the chip, instead of the older technique of binning the chip based on the lowest common denominator (slowest core). 

This technique discards the old paradigm of only having three power states, which provides more granular power control to improve both power efficiency (battery life) and performance. The CPPC interface also allows the operating system to dictate power state transitions on a per-core basis via a broader range of voltage and frequency settings. Now the processor can assign any voltage or frequency within its operating range on the fly. Additionally, this technique reduces transition latency from ~30ms when it is controlled by firmware down to 1-2 ms, which improves performance in bursty workloads and saves power. 

AMD also added per-core power and frequency management to provide another layer of granularity. In the past, the CPU cores were tied to the same voltage plane as the GPU, resulting in a 1:1 relationship between CPU vCore and GPU Vgfx, regardless of load on either unit. 

The illustration on the right of the second slide illustrates the impact – Ryzen 4000's cores all ran at the same voltage regardless of voltage, matching that of the graphics unit, while Ryzen 5000 adjusts on a per-core basis regardless of GPU voltage.

During multi-threaded heterogeneous workloads, the chip can match frequency and voltage on all the cores running that specific workload, thus ensuring consistent operation. AMD’s on-chip power regulation also supports software quality of service (QoS) hooks that allow the operating system to more effectively identify higher- and lower-priority workloads, which then is used to dictate voltage and frequency of the affected cores to maximize power efficiency.

As you can see in the last slide, AMD says that the culmination of these technologies delivers significant reductions in SoC power draw, with a head-to-head comparison between the Ryzen 7 5800U and 4800U showing impressive reductions in a few different scenarios. That includes a 26% reduction in standby mode, an impressive 47% reduction at idle, a 33% reduction in video playback, and an 18% reduction when the chip is under load in the MobileMark 2018 benchmark.

Ryzen 5000 Mobile Battery Life

Image 1 of 2

AMD also made a host of other refinements at the system level to help reduce power draw, including working with audio codec vendors to increase the occurrence and duration of low-power/idle residency states for the audio codec.  Likewise, AMD worked with hardware vendors for certain components, like DC-to-DC regulators and embedded controllers, to enable support for new ‘hints’ that trigger lower-power states.

AMD says it has increased standby state time by 20 hours, idle battery life by 3.9 hours, video playback by 1.1 hours, and MobileMark by 2 hours. The company also claims that the Ryzen 9 5900HX is 19% faster in single-threaded workloads than Intel's flagship 10980HK. 

AMD Ryzen 5000 Mobile Performance Benchmarks

AMD provided a host of performance benchmarks in three key areas: Mobile gaming, content creation, and ultrathin performance. As with all vendor-provided benchmarks, you should take these results with a grain of salt. We're sharing the full slide deck for each category, but without commentary – take the results as you will.  We also include the endnotes that contain the test configurations. 

Today marks the launch of the chips, so there should be plenty of third-party reviews for another take on performance, like our Asus ROG Flow X13 review. 

AMD Ryzen 5000 Mobile Cezanne Gaming Benchmarks

Image 1 of 11

AMD Ryzen 5000 Mobile Content Creation Benchmarks

Image 1 of 19

AMD Ryzen 5000 Mobile Ultrathin Benchmarks

Image 1 of 21

Thoughts 

Ryzen 5000 Mobile brings a new level of competition to the laptop market through a more refined 7nm process, the Zen 3 architecture, and good old fashioned engineering. Some will see AMD's decision to re-use key parts of the design, like the Vega graphics unit and PCIe 3.0 controllers, as a liability. Still, AMD managed to wring out enough extra performance and longer battery life to offset most of the downsides. 

AMD's re-use tactic allowed it to speed the chips to market and also simplify the design process for its OEM partners. That's important in the face of a much larger and entrenched competitor like Intel. AMD's deeper levels of engagement with ecosystem partners are emerging as it gains more market share and thus has access to more resources. And this is a virtuous cycle: AMD has made the types of adjustments and collaborations that OEMs appreciate, which should further deepen its partnerships.

AMD's Ryzen 4000 series mobile processors have already taken the notebook market by storm, granting the company its biggest slice of the mobile market in history, and the Ryzen 5000 Mobile chips look to continue that momentum. AMD pulled off this feat in record time — the Raven Ridge chips didn't come to the mobile market for a year after debuting on the desktop, while the Zen 3 transition should take roughly four months.

Image 1 of 4

Speaking of Raven Ridge, those chips came to market in 75 designs, while the next-gen Ryzen 4000 chips landed in 100 devices. AMD expects Ryzen 5000 to power over 150 notebooks by the end of the year, many with the highest-end graphics cards, a distinction that has long eluded the company. As you can see above, the chips will debut in a wide range of devices from leading OEMs. 

For now, supply could be AMD's biggest obstacle. Record demand, coupled with the trade war and conditions associated with the pandemic, has strangled the supply of chips for almost every segment, spanning from automobiles to the CPUs and GPUs craved by enthusiasts, which has lead to price gouging. Hopefully AMD can ensure enough supply to keep OEM pricing within reasonable bounds.

AMD's partners do have Ryzen 5000 Mobile laptops on the market, but we'll have to wait and see how long term supply pans out. Given what we've seen from the first reviews, it looks like AMD will have plenty of demand for its latest chips.