Modern CPU’s do not have static clocks anymore. Rather, they have a dynamic clock speed on top of the guaranteed base clock frequency. This is true whether we are talking about the Intel Core i9 9900K or AMD’s Threadripper 2990WX, these are all the same in that they all have a secondary advertised clock speed frequency. This is known as the boost clock. The Ryzen 9 3900X sample is advertised as having a clock speed of 3.8GHz/4.6GHz. This is also known as base clock and boost clock.
It’s important to note that boost clocks are never guaranteed and that they are form of automatic overclocking. This isn’t the type of overclocking that voids the warranty as this is a feature built into the CPU’s to improve performance under ideal conditions. There are two basic concepts for these higher clocks. There is what’s known as the single core boost and all core boost. AMD actually states that these are outdated concepts that do not exist regarding their CPU’s. This isn’t true, but here is what AMD says:
AMD Ryzen Precision Boost 2
The automatic overclocking algorithm that governs boost clocks on AMD Ryzen CPU’s is Precision Boost 2. The part we are really focusing on is where AMD says it does not have an “all-core boost” or a “single-core boost.” This isn’t accurate in the strictest sense. If you run a lightly threaded or single-threaded application, you will get one core that will clock much higher than the others will. Which core this is varies by individual processor.
You can see which ones are the likely candidates in Ryzen Master as it will usually be one marked with a gold star. However, this single core is not guaranteed to hit the maximum advertised clock speed. It certainly might happen, and we’ve seen it happen, but this didn’t occur during our initial review.
As the paragraph from AMD above states, Precision Boost 2 (automatic overclocking) is an algorithm that sets a loaded core to the highest possible frequency. Some of the variables that govern this are mentioned above. Under Precision Boost 2 (PB2 from now on) there are pre-defined OEM programmed limits that the processor will adhere to.
You can see in the image below what happens when you hit these limits. In our POV-Ray test, (multi-threaded) the PPT, TDC and EDC values are at 100%, 99%, and 99% respectively. This resulted in a reduction of CPU core frequency across all the cores to 4,036MHz. Note that our CPU temperature in this example was still only 68c.
Essentially, the clock speed will only hit the advertised maximum boost clock speed of 4.6GHz if the PPT, EDC and TDC values are within those preset OEM limits and the workload demands it. These are values set for the CPU by AMD. This is controlled by the CPU and the BIOS using AMD’s AGESA code. This is base code that the motherboard makers cannot touch or alter. AMD’s AGESA Combo PI code is sent to the motherboard manufacturers as needed to solve issues or improve performance, add features, etc. This code can do things like improve memory clocking or improve compatibility. There is a bit of a juggling act as sometimes a newer AGESA code isn’t necessarily better. You could pick up clock speed while loosing actual performance.
All CPU Cores AMD Ryzen
The same algorithm that governs boost clocks for single cores also applies to all CPU cores. Where AMD says that it doesn’t have specific all core or single core overclocking comes down to the fact that the PB2 algorithm is opportunistic. In other words, if the CPU can actually clock all the cores in heavily threaded workloads to that maximum boost clock, it will. Things are never that good on air or water cooling, but in theory, if it could be done, it would. This is why AMD doesn’t make an all core or single core distinction. In lightly threaded, or single-threaded workloads, it is possible to see one or more cores at 4.6GHz using our 3900X as an example.
The thing is, when placing a heavily threaded workload on the CPU, it’s impossible for conditions to allow the CPU to clock all its cores at the full 4.6GHz clock frequency. What you are looking at is how frequencies behave on this model CPU (typical) when more cores are used. You can see how a single thread can run at 4.6GHz and as you add cores and threads, the clocks drop until around 4.3GHz. You can also see slightly slower clocks than that as well.
Again, it depends on specific environmental and system configurations as to what you will see specifically. The main issue we are talking about here is how single-threaded performance may have suffered because we never saw the theoretical 4.6GHz clocks during our review. That’s the crux of what people are on about in forums and on social media. I think many people understand you won’t see 4.6GHz on all cores, but they expected to see it on a single core at least.