Ryzen 7 8700G: Overclocking and Tuning Guide

The Ryzen 7 8700G is AMDs new 8 core 16 thread Zen 4 processor built on TSMCs 4nm FinFET technology, that comes with the powerful Radeon 780M integrated graphics. On the specsheet the 8700G looks similar to the Ryzen 7 7700, but with a 200 MHz lower boost clock of up to 5.1 GHz and half the L3 cache at 16 MB. Like the previous AMD G-Series processors, the new APU only has a single die instead of the multi-die solutions we find on all the other typical Ryzen processors.

For the full specification, visit amd.com.

In this guide we are trying to unlock the full potential of the Ryzen 7 8700G and the Radeon 780M iGPU through overclocking and other means of optimization to find out what AMDs top of the line APU is really capable of. I'm particularly interested on how much performance we can squeeze out of the integrated graphics and how the overclocking experience will be in general.

As a testing platform I'm using the ASUS ROG Crosshair X670E Gene motherboard paired with 2x16GB DDR5 dimms from Patriot alongside the Ryzen 7 8700G, everything setup on an open-air benchtable. The baseline memory speed will be DDR5-5200, just like stated in the official specification. We will find out later how well the CPU responds to higher memory frequencies and tighter timings. For cooling the CPU I'm using a regular Wraith Prism cooler and an additional 120mm fan pointed at the DDR5 dimms. All fans always run at 100% fan speed during all tests.

In order to establish a performance baseline, I did run a handful of synthetic benchmarks and also gathered data for some selected games. 

The average clock speed during the Cinebench R23 multi-threaded benchmark was 4850 MHz with spikes up to 5025 MHz. The CPU temperature peaked at 74.1 °C while pulling 87.7 watts of power on average with a core voltage ranging from 1.18V to 1.2V.

For the gaming benchmarks I picked The Elder Scrolls V: Skyrim and Minecraft, because of their popularity and because the Radeon 780M integrated graphics might actually be able to produce playable results in those titles.

Settings: 1440p, Ultra

Looking at the numbers in Skyrim, 34 FPS certainly is not a great gaming experience. However, I had to choose 1440p resolution in order to have enough headroom for improvements when running the benchmark with the fully tuned setup. We also see that the CPU isn't really doing that much work, looking at the relatively low CPU Clock and higher core voltage. The power draw of only 73.1 watts compared to the 87 watts we observed during a heavy CPU load, could indicate that there is a bottleneck in either the integrated graphics or memory.

Minecraft

Settings: 1440p, Optifine 1.20.1 with BLS Shaders, Render/Simulation distance 12 chunks

The customized Minecraft benchmarks puts more load on the CPU overall as we can observe from the average power draw of 86.5 watts. The effective clock of the integrated graphics is significantly lower than what we got during the Skyrim benchmark, which is probably due to the CPU hitting the the 87 watts power limit.

With the performance baseline established we can continue and start overclocking the CPU cores of the Ryzen 7 8700G. To do so we will utilize Precision Boost Overdrive (PBO) which allows me to tune the boosting behavior of the CPU depending on the current operating temperature, power and current draw. Additionally, we are going to use Curve Optimizer (CO) to apply a negative voltage offset to the CPU cores. This has proven to be an easy and effective way to overclock Ryzen CPUs in the past.

To setup PBO and CO on the ASUS ROG Crosshair X670E Gene, I entered the BIOS and on the "Advanced" Tab, "AMD Overclocking", "Precision Boost Overdrive" I configured the following settings:

Additionally, on the "Extreme Tweaker" Tab under "DIGI + VRM" I set the "CPU Load-line Calibration" to "Level 5".

Overclocking the FCLK

With the PBO settings dialed in, we will continue overclocking the Infinity Fabric clock (FCLK). Traditionally, the FCLK on G-Series Ryzen processors can clock quite a bit higher than the multi-die Ryzen CPUs and the Ryzen 7 8700G is no exception to that. The final FCLK I was able to run through my tests was 2467 MHz. Although 2500 MHz was possible in most scenarios, I had occasional crashes when using higher memory speeds, but more on that later.

Now let's see what performance gains we got from our CPU overclock.

Overall, we are only seeing small improvement across the board, ranging from 1% in the Cinebench R23 single test, up to 6.9% in the custom Minecraft benchmark. 

The average clock speed during the Cinebench R23 multi-threaded benchmark was 5005 MHz with spikes up to 5100 MHz. The CPU temperature peaked at 94.4°C while pulling 115.1 watts of power on average with an average core voltage of 1.28V. This represents an increase in power draw of 32% for a performance gain of just 2.2%.

During the Skyrim benchmark, the highest CPU frequency was 5150 MHz with an average power draw of 85.6 watts, compared to the 73 watts in stock condition. In the Minecraft benchmark, the peak CPU clock reached 5125 MHz with a power draw of 108 watts on average versus 86.6 watts without the overclock. Although with PBO enabled the CPU is allowed to boost up to 5350 MHz, these clocks weren't reached during the benchmark runs.

In general, these kind of improvements align with what I experienced when using other Ryzen CPUs with PBO in the past. Alternatively, you could also try and overclock the CPU manually by applying a static all core overclock and core voltage, but keep in mind that this has the drawback of disabling the 95°C maximum operating temperature throttle mechanisms and allows the CPU to reach even higher temperatures.

With the CPU and FCLK overclock dialed in, let's move on and overclock the Radeon 780M integrated graphics. Just like the CPU cores, there are two different techniques you can apply to overclock the iGPU, using PBO or manually applying a static overclock and voltage. PBO will allow us to increase the graphics clock by up to 200 MHz in the best case which will result in a theoretical maximum frequency of 3100 MHz. However, I think the clock can be pushed past that and that's why I opted to go with the static overclock instead.

The iGPU can be overclocked fairly straightforward in the "AMD Overclocking" section of the BIOS under "Manual iGPU Overclocking". I configured the following overclock settings:

Unfortunately it's not possible to get a higher voltage than 1250 mV as it seems to be the limit for the iGPU voltage. You can set a higher value in the BIOS, but it won't be applied and just fall back to 1250 mV. The highest stable overclock I could achieve and run through all my tests without crashing was at 3200 MHz. It's less than what I had hoped for, but still higher than the maximum 3100 MHz frequency limit you have when using PBO.

With the overclock settings applied, I did run the 3D benchmarks again with the following results:

The performance gains from the additional 300 MHz overclock on the Radeon 780M are again very small. Compared to the previous results with just the CPU and FCLK overclocked, there is only a marginal improvement of roughly 3%. I was a little bit disappointed with these results to be honest and was hoping that overclocking the iGPU would yield more performance.

During the Skyrim benchmark the average power draw is now 91.3 watts, about 6 watts more than without the iGPU overclock, while the custom Minecraft benchmark pulled 111 watts on average, an increase of a just 3 watts.

The Ryzen 7 8700G is now fully overclocked with the CPU cores utilizing PBO and CO for a maximum boost of up to 5350MHz, the FCLK is clocking at 2467 MHz and the Radeon 780M integrated graphics runs a speed of 3200 MHz. Now the last thing that's left to do in order to push performance higher is overclocking the DDR5 memory. During all the previous benchmarks we followed the official AMD specification for memory speed of DDR5-5200 for a single rank dual channel configuration. Now we will explore how far we can push the memory and what impact it will have on the CPUs and integrated graphics performance.

Before we start let's explain the two different modes the Ryzen 7 8700G integrated memory controller (IMC) can operate in regards to memory speed.

The IMC can either run in 1:1 mode, which means it will run at the same speed the memory modules are running at. For example, when using a DDR5-6000 configuration, this means that your memory will run at a clock of 3000 MHz. The IMC will be synchronized with the speed of the memory modules at run at the same 3000MHz. We know by now that other Zen 4 processors are capable of running memory speeds of DDR5-6000 up to DDR5 6400 in 1:1 mode, depending on the quality of the IMC, the motherboard and the memory modules and ICs they are using. 

If we want to push higher memory clocks, we can configure the IMC to operate in 1:2 mode. This simply means the integrated memory controller will run at half the speed of the DDR5 modules. However, using 1:2 mode will introduce a small performance penalty, but opens the door for higher memory speeds.

The ICs found on your memory modules will be the major limiting factor in how far you can push the memory clock and timings. Let's have a short look at the most commonly found ICs on DDR5 modules used today and there overclocking potential:

Samsung B-Die: These IC's are mostly found on modules running DDR5-6000 CL36-36-36 at 1.35V or similar. Especially the lower tRCD and tRP of 36 are a good indicator that a memory kit uses Samsung B-Die ICs. In terms of performance they are a solid choice for running 1:1 mode on Ryzen CPUs, but you will have a hard time reaching speeds of DDR5-6600 or above. In terms of overclocking headroom and performance they aren't too bad, but there are better options available.

SK-Hynix M-Die: M-Die was the go-to memory IC when DDR5 modules first arrived, because they offered the best overclocking potential at that time. 2x16GB memory kits running DDR5-5600 CL36-38-38 at 1.25V, DDR5-6000 CL36-38-38 are well known for having SK-Hynix M-Die ICs on them. The easiest distinction are again the tRCD and tRP timing values, because M-Die cannot run them quite as low as Samsung B-Die. However, these ICs can clock all the way up to 7200 MHz if you are pushing them and are also capable of running lower sub-timings which ultimately makes them superior over Samsung B-Die ICs.

SK-Hynix A-Die: A-Die is the second iteration of SK-Hynix DDR5 ICs and sets new standards in regards to memory speed. Any 2x16GB memory kit running DDR5-6800 or faster will most likely have SK-Hynix A-Die on them. These ICs are capable of reaching memory speeds of DDR-8000 or more, but only if your CPU and motherboard are able to handle these kind of speeds. If you are serious about overclocking DDR5 memory, than these ICs are definitely what you are looking for when shopping for memory kits.

The memory modules I'm using in this test, the 2x16GB Patriot Viper XTREME 5 rated at DDR5-8200, also use SK-Hynix A-Die. Of course, such a highly rated kit is not necessary and kits with ratings like DDR5-6800 and DDR5-7200 can be able to produce similar results, because they are using the same A-Die ICs under the hood. For the sake of this test however, I wanted to make sure the memory isn't limiting me in any kind during the test. As always when overclocking, your milage may vary and each memory kit will overclock slightly different.

Our first goal will be to push the memory clock as high as possible while having the memory controller in 1:1 mode and then optimize the memory timings for that speed. In the BIOS, I entered the "Extreme Tweaker" tab again and configured the following settings:

Let's explain these settings a little bit. The CPU SOC voltage is very important as it powers the CPUs IMC. I like to start at a relatively low value like 1.20V and increase it as I raise the memory voltage. The highest you can set your CPU SOC voltage is 1.30V, but I usually try to stay slightly below that if possible.

The DRAM VDD voltage is the main voltage for the memory ICs and you need to increase it as you raise the memory frequency. XMP / EXPO profiles commonly use DRAM VDD voltages between 1.35V and 1.45V. DRAM VDDQ is the voltage that powers the I/O of the memory ICs. You can keep it in sync with VDD, but I found that 1.35V works well for me so I won't run it unnecessarily high if I don't have to. CPU VDDIO / MC is the voltage used to transfer data between the memory controller and the dimms. I didn't change any other voltages manually and left them on auto.

With the voltages and memory frequency configured let's enter the "DRAM Timing Control" page and tighten up the timings. I won't go over every timing in detail but will share them anyway for you as a reference point and for transparency.

When overclocking your memory, it's recommended to only change one settings at a time and immediately run a stability test afterwards. This is especially important if you are relatively inexperienced in memory overclocking, because it will make it harder to identify issues in your configuration when you encounter errors during testing.

After all the memory timings are tightened and the memory overclock settings dialed in, let's see how the overclock affects the performance of the Ryzen 7 8700G.

CPU performance barley changed at all with the tuned memory configuration, while 3D performance increased by about 20% across all workloads tested with up to 25.4% in the custom Minecraft benchmark. The average power draw during the Skyrim benchmark is now up at 106.7 watts and 124.7 watts in the Minecraft test.

It's incredible to see how effective the memory overclock is for 3D performance. One of the major differences between the integrated graphics and discrete GPUs is, that it doesn't have any dedicated video memory, but relies entirely on the system memory instead. Because of that, it's no surprise that overclocking the system memory vastly improves the overall performance of the Ryzen 7 8700G integrated graphics.

At this point it's already clear that memory overclocking is the main performance driver for the Zen 4 APU. The Radeon 780M iGPU simply cannot operate effectively when it's limited by the DDR5 system memory. Armed with that new knowledge, I am very curios how much there is left to gain when pushing the memory higher.

DDR5-6400 was the highest possible speed I managed to run in 1:1 mode. To push higher frequencies we will put the memory controller in 1:2 mode and overclock the memory further. I went into the BIOS and adjusted the memory overclocking settings as follows:

CPU SOC voltage increased slightly to 1.275V as well as the CPU VDDIO / MC voltage. The DRAM VDD voltage also increased to 1.45V. If you are wondering, the voltage levels in this configuration are still at a reasonable level and nothing you wouldn't also see on a high performance XMP / EXPO profile.

To put the memory controller in 1:2 mode, you have to set the "UCLK DIV1 MODE" setting to "UCLK=MCLK/2"

With the memory controller now operating in 1:2 mode, but a faster memory speed of DDR5-7200, it will be interesting to see how the performance compares against the optimized 1:1 settings with DDR5-6400. 

Regarding CPU performance, the new DDR5-7200 overclock in 1:2 mode didn't really move the needle much. 3D performance however received another performance uplift and is now sitting at +31.1% in the Skyrim benchmark test. The average power draw only increased a little bit at this points, reaching 109.5 watts in the Skyrim test and 126.9 watts in the custom Minecraft benchmark. The Radeon 780M iGPU seems to be very happy about every small memory bandwidth improvement it can get, so I will continue to push the memory speed further.

Again in the BIOS, I adapted the overclock settings and set the memory frequency to DDR5-8000, leaving the voltages the same as in the DDR5-7200 configuration. At this point I also have to remind you that I am using a high end, two-dimm motherboard which will help tremendously with running high memory frequencies.

For the memory timings, I just had to loosen the primary timings a little bit as well as tRFC2 in order to stabilize the system.

With a memory speed of DDR5-8000 as well as optimized timings, these are the final results I got for the benchmarks:

At DDR5-8000, the Ryzen 7 8700G with its Radeon 780M integrated graphics really shows what it's truly capable of and how much potential for improvements it has left. In this final overclocked configuration, the highest performance uplift was achieved in the Skyrim benchmark with an improvement of +44.3% on average frames per second. During this run, the CPU consumed a total of 115.3 watts on average. In the custom Minecraft benchmark, the power draw is averaging 130.3 watts now, an increase of +51.1% compared to the power draw at stock settings.

The overall overclocking experience of the Ryzen 7 8700G and Radeon 780M iGPU itself was actually somewhat disappointing. Overclocking the CPU cores itself isn't any different from other Ryzen 7000 series processors I have worked with and it didn't really have a significant impact on the CPUs performance.

The overclocked Radeon 780M iGPU also left something to desire. Even with a 300 MHz frequency increase the actual impact on performance was only small and although it's very easy to overclock the integrated graphics, you also don't have a lot of different overclocking techniques available.

The crazy performance uplift in 3D workloads with optimized memory configurations however, is really impressive and I believe this is what makes the Ryzen 8700G with the Radeon 780M iGPU actually usable in a lot more workloads. It also makes the CPU cooler less relevant, as the CPU and iGPU itself aren't the main performance drivers. This fact could make the 8700G even a possible candidate for a fully passive system, as most of the improvements come from overclocking the system memory. On the other side, overclocking DDR5 is significantly more challenging and takes a considerable amount of time and effort to really get it right.