You know that processors can have more cores. You know that processors can have faster cores. But what about smarter cores? That’s a new horizon we’re exploring with AMD SenseMI technology in the all-new AMD Ryzen™ processor!
AMD SenseMI wraps up five features that work in concert to enhance the performance of the AMD Ryzen CPU1. Some of the features optimize power and clockspeeds, while others bring important data into the processor or optimize processor pathways for new work. Altogether they make a rational intelligent machine that’s constantly obsessing over how to optimize performance and power efficiency for you and your applications. Let’s take a look!
It All Starts With the Infinity Fabric
Today’s processors can often be called systems-on-chip (“SoC”), which means support for USB, PCI Express® and SATA are integrated directly into the CPU core. Getting these technologies to communicate with the CPU cores has historically been a time-consuming, expensive, or inefficient task. But the Ryzen SoC is a different beast thanks to the Infinity Fabric.
The Infinity Fabric is a common interface that allows us to quickly mesh these pieces together and get them “speaking the same language,” almost like snapping toy building blocks together. We also use the Infinity Fabric to establish fast communication between groups of CPU cores, as we do in the 8-core Ryzen processors that contain two groups of four cores.
Most importantly for AMD SenseMI, the Infinity Fabric gives us command and control powers to nearly all areas of the CPU. That’s crucial because the Ryzen processor has a networked “smart grid” of several hundred sensors, each accurate to 1 milliwatt, 1 milliamp, 1 millivolt and 1°C. These sophisticated sensors are what allow the Ryzen processor to dial in voltages, clockspeeds, and optimal datastreams. Having extensive insight into the readings of these sensors via the Infinity Fabric allows the processor to orchestrate them for best results.
Extended Frequency Range
Temperature is king when it comes to determining a processor’s maximum clockspeed, as cooler temps improve the efficiency and reliability of the tiny transistors that make up a processor. Other factors in the clockspeed include power draw from the CPU socket, what percentage of the CPU’s circuits are in use, and the distance to maximum thermal output (TDP). But temperature is the one factor that you can easily control with better CPU or chassis cooling.
Thanks to AMD SenseMI technology, the Extended Frequency Range (XFR) feature, available on select Ryzen processors, can measure the difference between the current CPU temperature and the operating temperatures we’ve designed the Ryzen processor to handle. If the current temp is sufficiently low, that extra thermal headroom can be converted into extra top-end frequency.
For example, the AMD Ryzen™ 7 1700X processor has a maximum clockspeed of 3.8GHz at 60°C, but XFR can automatically get the maximum frequency to 3.9GHz if the current temperature is lower than that. As you can see, select Ryzen processors are capable of giving a little more to users that build premium systems with robust system and CPU cooling. Pretty neat!
AMD SenseMI also comes in handy for boosting the CPU clockspeed with a feature we call “Precision Boost.” The same temperature/current/TDP analysis that governs XFR is once again in play to establish the boundaries of safe operation for a Ryzen processor. Like any other processor, we want to make sure that the Ryzen processor consumes only so much power, operates within an expected temperature range, and emits only so much heat (TDP). Overclocking can naturally expand or override these boundaries, but we’re talking out-of-the-box functionality in this case.
As long as a Ryzen processor isn’t bumping up against any of those boundaries, Precision Boost can raise the clockspeed in exacting 25MHz increments. Relative to past processors, these small
increments allow the Ryzen CPU to get that much closer to a optimal frequency taking all thermal and electrical boundaries into account. The 25MHz increments can also enable higher sustained frequencies by minimizing clockspeed reductions that occur when a reliability threshold is encountered.
Example A: A Ryzen processor is running a lightly-threaded workload using just a few CPU cores. Because the other CPU cores are dormant, or working on background tasks, there is significant thermal or electrical headroom for the processor to just go faster. The Ryzen processor can use Precision Boost to convert that headroom into additional clockspeed (e.g. 3.0GHz → 3.7GHz on the AMD Ryzen™ 7 1700X processor).
Example B: A Ryzen processor running at 3.8GHz could encounter a heavy workload that’s on a trajectory to use more power than the CPU socket is designed to provide. This is an ordinary and manageable event for processors, and perhaps a short dip to 3.775GHz would be sufficient to correct the trajectory back into expected levels. Precision Boost can make that possible, and the clockspeed could quickly be pushed back to 3.8GHz when the workload lightens. Other processors might have to drop to 3.7GHz, taking off another 75MHz of frequency that a Ryzen processor might not.
The exemplary power efficiency of the Ryzen processor comes from two key areas: 14nm FinFET manufacturing and low-power design methodologies. Pure Power orchestrates those methodologies, imbuing every Ryzen processor with the power to inspect and adjust its own electrical characteristics.
Pure Power is especially vital during manufacturing. When a Ryzen processor rolls off the assembly line, each chip is capable of looking into itself and analyzing the quality of its own silicon. The results of that analysis allows the processor to zero in on an idealized voltage vs. frequency curve for itself. That fine tuning allows the processor to get pretty close to the perfect voltage for a given frequency. A magic wand wouldn’t do much better!
During the design phase, this self-tuning opens the door for AMD to reduce or eliminate guardbands, which is “slack” built into the voltage or frequency targets that can compensate for moments when the processor’s automated routines can’t quite nail a specific value. This can happen for any number of reasons, including transient fluctuations in a power supply’s output, or sudden large jumps in CPU utilization. But Ryzen processors came off the line with precise knowledge of themselves, so reducing or eliminating these guardbands allows for higher overall clockspeeds and lower operating voltages for you.
And in day-to-day use, Pure Power is aggressively managing dynamic or “operational” power. Idle pieces of the Ryzen processor are downclocked or shut down to trim power, or to reallocate that power to areas of the processor that can productively use it. This technology is called “clock gating.”
As an example: We put the AMD Ryzen™ 7 1800X processor against the Core i7-6900K in the demanding POV-Ray test. This test measures the performance of a processor with raytracing, the most realistic form of 3D rendering. As you can see from our data below, the Ryzen 7 1800X enabled a better score and higher performance per watt.2
|Processor||POV-Ray Score||Average System Wall Power||Performance per Watt (Higher is better)|
|AMD Ryzen 7 1800X||3266||157.45W||20.74|
Neural Net Prediction
Where Pure Power, XFR, and Precision Boost cooperate to control power/frequency characteristics of the Ryzen processor, Neural Net Prediction is responsible for anticipating optimal pathways in the processor for the programs you’re running.
Neural Net Prediction starts with a true artificial intelligence (AI), which uses a simplified approximation of the human brain (neural net), to learn how your programs behave. Applications, and the languages used to write those applications, are human-created and have predictable patterns. Humans love patterns, and those patterns hidden in the applications can be learned!
The learned patterns form a behavioral history of an application, and that history lets the processor predict what a program is likely to do in the future. The Ryzen processor uses those predictions to pre-load certain capabilities—like storing to RAM, adding numbers, or comparing values together—so they’re ready to go before your application even makes a request. This saves processing time, and contributes to higher processor performance.
It’s important to know that the behavioral learning of Neural Net Prediction is temporary. The history is emptied when you launch a new application, or when the PC is reset or powered down. The applications you run re-train the neural net each time, and you might find that the second time you run a benchmark is a little faster than the first. That’s Neural Net Prediction at work!
Before the Ryzen processor can start to run your applications, relevant data must be brought into the processor and stored in local cache. Cache is ultra-fast memory located right on the processor, and processors like Ryzen achieve peak performance when important data fits into that cache.
It’s worth highlighting that “data” typically means “code,” where entire sub-routines of a running program are stored in cache. This can reduce or eliminate the odds that the processor has to reach across the motherboard to retrieve data from your RAM. Although the RAM is only a few inches away from the processor to your eyes, that’s a very long way from the perspective of a processor, so cache is paramount for top performance.
But feeding the cache with data is only half the battle. Getting the right data is the other half of the equation, and that’s where Smart Prefetch shines. Smart Prefetch consists of sophisticated learning algorithms that intuit what data is most used and most relevant in your applications. Smart Prefetch can then prioritize the important data, or even predict the important data, so it’s ready to go before the application needs it.
Having the next important dataset queued for execution behind the current work helps ensure that the Ryzen processor always has a consistent flow of high-quality data. And with an atypically large 20MB combined cache, Ryzen 7 1800X, 1800 and 1700 processors are uniquely equipped to handle large datasets common in scientific or creative workloads.
More cores and faster cores is well-tread ground in the PC industry (though we dare say the Ryzen™ processor is the best blend yet!), but AMD is exploring a new horizon with smarter cores. Armed with sophisticated learning algorithms, neural networks, and uncanny powers of prediction, the Ryzen processor is an incredibly intelligent and rational agent that’s ready and waiting to zero in on the exact level of performance and power efficiency you and your applications deserve.
Robert Hallock is a technical marketing guy for AMD. His postings are his/her own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
1. Not all AMD Ryzen™ processors offer every feature of AMD SenseMI technology. For specific capabilities of different processor models, please visit www.amd.com. If your system is pre-built, contact your manufacturer for additional information.
2. Default POVRay rendering preset. AMD system configuration: Ryzen 7 1800X (8C16T/3.6-4.0GHz), 2x8GB DDR4-2400, AMD Customer Reference Motherboard, NVIDIA Titan X (Pascal), NVIDA Driver 126.96.36.19933, Windows 10 x64. Intel system configuration: Core i7-6900K Extreme (8C16T/3.2-3.7GHz), 2x8GB DDR4-2400, ASUS STRIX X99 Gaming motherboard, NVIDIA Titan X (Pascal), NVIDA Driver 188.8.131.5233, Windows 10 x64. Average wall power draw: 157.45W (AMD) vs. 153.59W (Intel). POVRay scores: 3266 (AMD) vs. 2964.12 (Intel). Performance/Watt (higher is better): 3266/157.45W=20.74 score/W (AMD) vs. 2964.12/153.59W=19.29 score/W