The AMD Ryzen 5 2400G and Ryzen 3 2200G Review: Return of the APU

Raven Ridge Desktop

We look at the Ryzen 5 2400G and Ryzen 3 2200G

As we approach the one-year anniversary of the release of the Ryzen family of processors, the full breadth of the releases AMD put forth inside of 12 months is more apparent than ever. Though I feel like I have written summations of 2017 for AMD numerous times, it still feels like an impressive accomplishment as I reflect for today’s review. Starting with the Ryzen 7 family of processors targeting enthusiasts, AMD iterated through Ryzen 5, Ryzen 3, Ryzen Threadripper, Ryzen Pro, EPYC, and Ryzen Mobile.

Today, though its is labeled as a 2000-series of parts, we are completing what most would consider the first full round of the Ryzen family. As the first consumer desktop APU (AMD’s term for a processor with tightly integrated on-die graphics), the Ryzen 5 2400G and the Ryzen 3 2200G look very much like the Ryzen parts before them and like the Ryzen mobile APUs that we previously looked at in notebook form. In fact, from an architectural standpoint, these are the same designs.

Before diving into the hardware specifications and details, I think it is worth discussing the opportunity that AMD has with the Ryzen with Vega graphics desktop part. By most estimates, more than 30% of the desktop PCs sold around the world ship without a discrete graphics card installed. This means they depend on the integrated graphics from processor to handle the functions of general compute and any/all gaming that might happen locally. Until today, AMD has been unable to address that market with its currently family of Ryzen processors, as they require discrete graphics solutions.

While most of our readers fall into the camp of not just using a discrete solution but requiring one for gaming purposes, there are a lot of locales and situations where the Ryzen APU is going to provide more than enough graphics horsepower. The emerging markets in China and India, for example, are regularly using low-power systems with integrated graphics, often based on Intel HD Graphics or previous generation AMD solutions. These gamers and consumers will see dramatic increases in performance with the Zen + Vega solution that today’s processor releases utilize.

Let’s not forget about secondary systems, small form factor designs, and PCs design for your entertainment centers as possible outlets for and uses for Ryzen APUs even for the most hardcore of enthusiast. Mom or Dad need a new PC for basic tasks on a budget? Again, AMD is hoping to make a case today for those sales.

Diving into the AMD Ryzen 5 2400G and Ryzen 3 2200G

For the gear-heads in the group, let’s start with the classic table of processor specifications, comparing the Ryzen APUs to the last generation AMD part and what Intel competitive solutions are in the same price range.

The numbers and data here aren’t going to blow your socks off, but it’s important to keep in mind that these decisions made by AMD are meant to keep the Ryzen Processor with Radeon Vega Graphics inside a very tight price segment. Things like die size, ability to produce high yields with GlobalFoundries, and even packaging are all going to be balanced for a sub-$200 part.

Just like the Ryzen mobile processors we looked earlier this year, the Ryzen APU for the desktop is built with a single, monolithic die, combining both the CPU, GPU, and uncore areas of the design on single piece of silicon. This contrasts with the Intel Kaby Lake-G part that we have discussed many times that uses a discrete CPU and a discrete GPU, paired with HBM2 memory, on a single package. KBL-G is a much higher cost solution and scales up to 100-watt power levels.

The Ryzen APU with Vega graphics does not include on-die memory of any kind dedicated for the graphics portion. Instead, it uses the same system memory pool and as the CPU cores and through the same memory controller on-die.

Clock speeds for the new Ryzen 5 2400G and the Ryzen 3 2200G are higher than the previous AMD offerings (that did not include a graphics portion). This is attributed to more time with the design and with the GlobalFoundries foundry process, leading to better ability to tweak and optimize the silicon. The net result is that even though the Ryzen APU includes a larger die with integrated graphics, it should perform better than the Ryzen 1000-series comparable processors even in CPU-bound tests.

CCX design and integration is always an interesting discussion when it comes to AMD’s Ryzen processors, as we have demonstrated how the CCX-to-CCX and die-to-die latencies that exists can impact performance in gaming and other workloads. Obviously, there is no die-to-die concerns with the Ryzen APU and AMD has decided to remove the complexity of having multiple CCX’s enabled by going wit ha 4+0 configuration. This means, four cores are enabled on a CCX and utilized with these CPUs, rather than going with a 2+2 design of two cores per CCX.

This does mean that the L3 cache on the processor drops from 8MB to 4MB. AMD claims that by increasing the clock speeds and doing “other” undiscussed things to “reduce cache and memory latencies”, the Raven Ridge die should see less impact that we might otherwise think.

AMD mentions in its reviewer’s guide for this part that it did thorough testing, specifically in gaming (where the company has suffered previously) and found that it was essentially a “net-zero” on going with a 4+0 or 2+2 design. While some games benefited more from the larger L3 cache of the 2+2 implementation, others saw more gain from the latency reductions of having all threads on the same CCX. All things being equal then, going with a 4+0 design allows AMD to make the die more compact, saving cost and complexity in the production step.

Another way to reduce die size, cost, and lower uncore complexity was to move from a 16-lane implementation of PCI Express down to 8 lanes for discrete graphics. Considering the cost of these processors, and the level of GPU likely to be paired with them in consumer builds, there is little chance that this will create any kind of bottleneck for mainstream gamers. It does remove the ability to run multiple graphics cards on an AM4 platform, but I think that is again something a Raven Ridge user would be extremely unlikely to want to do. There remain 8 additional lanes for chipset connectivity (using 4 of them) and for NVMe/accessories. (Yes, that means a discrete GPU, on a B350 motherboard, and an NVMe SSD will use all available PCIe on the processor.)

An interesting physical change on the Ryzen APU is with the package itself. To save money, AMD is going with a lower cost, non-metallic thermal interface between the die and the heatspreader. This might raise some eyebrows as the same kind of thing that Intel did with its Core processor family a couple of generations back – and it’s basically the same idea. AMD is doing this only on its lowest priced processors (today), while Intel is using standard TIM for all enthusiast class products.

This new package supposedly changes enough of the electrical pattern for AMD to officially support JEDEC 2933 MHz DDR4 memory as well, which should be a good upgrade for the integrated graphics portion of the processor.

One of the only major feature improvements on the design is a move to Precision Boost 2, the company’s upgraded algorithm to monitor telemetry of the processor including temperature, power, and load, to adjust the clock speed for optimal performance. This was available in the first generation of Ryzen but was relatively crude in implementation. Engineers this time around have the ability to be more granular, adjusting the clock speed in the same 25 MHz increments but more in line with the cores and threads being utilized. This should equate to better average clock speeds based on similar workloads to the first iteration.

Interestingly, my understanding from the initial Ryzen launch was that the implementation we see in Precision Boost 2 is what AMD had envisioned all along. It simply ran out of time in a rushed and hectic launch window.

For power delivery, the Raven Ridge design has two dedicated power rails. One is for the CPU section and the other is combined and controls the GPU and memory section. This is likely a compromise of sorts in the architecture as ideally you want this to be as granular as possible as well. We’ll see in our testing (today and going forward) how much impact this has on overclocking and power efficiency, but I am guessing that the need to raise power to the graphics sub-section when the memory controller may only be active for CPU tasks will have some effect.

For those of you looking for more information on the graphics system at play with the Ryzen APU, there are no major changes from the Vega architecture that we have come to understand at its release in the summer of last year. You get the same features and capabilities, including consumer features like FreeSync and developer-focused capabilities like rapid packed 16-bit math acceleration.

The Ryzen 5 2400G has an 11 CU (compute unit) implementation with 704 stream processors, 44 texture units, and a peak theoretical compute rate of 1.76 TFLOPS. The Ryzen 3 2200G lowers that to 8 CUs, 512 stream processors, 32 texture units, and a compute rate of 1.12 TFLOPS.

Both share 16 ROPs, 2 hardware schedulers, 4 asynchronous compute engines, and 2 render backends. Based on the numbers alone, it looks like AMD has built these parts to be as much compute bottlenecked as possible, leaning towards overcompensation in the way of scheduling and async compute. In a system that may end up memory starved (as most integrated graphics configurations are) there emphasis on memory-saving capabilities could be a plus.

Though I will openly admit to being hesitant to accept AMD Infinity Fabric as the magical construct the company often portrays it, the implementation on a product like the Ryzen APU and Raven Ridge seems idyllic. The Infinity Fabric on this part is used to service the six unique clients in the processor design that include: the Zen core complex, the Vega graphics, multimedia engines, display engine, DDR4 memory controllers, and the I/O and system hub. These are all monitored and managed by the IF and are included in the heuristics of the part for power management and even Precision Boost 2.

Finally, looking at the SoC design for its I/O capability, the Ryzen APU supports 4x USB 3.1 Gen 2 ports, one additional USB 3.1 Gen 1 and USB 2.0 port, 8 lanes of PCIe 3.0 for discrete graphics,8 lanes of PCIe 3.0 general use (four of which are utilized for the chipset connection), and 2 SATA ports. Anything beyond that is going to be provided by the chipset and motherboard implementation itself.