We overclock the AMD Radeon 780M integrated into the Ryzen 7 8700G up to 3150 MHz with the ASUS ROG Strix X670E-I Gaming WiFi motherboard and EK custom loop water cooling.

https://www.youtube.com/watch?v=yWZycOlMwNI

I already overclocked the Ryzen 7 8700G CPU cores in SkatterBencher #69. Still, today, we finally get to do the most exciting part of APU tuning: overclock the integrated graphics. The Radeon 780M was unusually tricky to overclock and is especially sensitive to memory tuning.

I hope to explain all that in this video. Let’s get started!

AMD Radeon 780M: Introduction

The Radeon 780M is not a new product from AMD, as it was also integrated into the 2023 Phoenix notebook processor lineup. For example, the Ryzen 9 7940HS already featured this integrated graphics. The Radeon 780M is only available on the desktop with the Ryzen 7 8700G APU.

The Ryzen 7 8700G is the flagship processor of AMD’s Zen 4-based Ryzen 8000 desktop APU product line codenamed “Hawk Point.” The Hawk Point processors were announced on January 8, 2024. A critical difference between the Ryzen 7000 CPUs and Ryzen 8000G APUs is that the former is a multi-chiplet-based design, and the latter features a single monolithic die. What’s also exciting about the Ryzen 8000G processors is that it’s the first AMD desktop product manufactured using the TSMC N4 process.

Hawk Point is the successor to the Ryzen 5000 Cezanne APUs launched in 2021. I overclocked the immediate predecessor of the Ryzen 7 8700G, the Ryzen 7 5700G, in SkatterBencher #24. In that guide, I covered both CPU core and IGP overclocking.

With the Radeon 780M integrated graphics, AMD finally brought the RDNA 3.0 architecture to APU, as the integrated graphics for the Ryzen 7000 processors (which I overclocked in SkatterBencher #55) was still the RDNA 2.0 architecture. This will be the first time I’m overclocking RDNA 3.0, as the last AMD discrete graphics card I overclocked was the Radeon RX 6500 XT in SkatterBencher #41.

AMD announced the RDNA 3 graphics architecture on November 4, 2022, during their Together We Advance_Gaming event. About a month later, the first RDNA 3 products entered the market with the Radeon RX 7900 XT and XTX featuring the Navi 31 chiplet. With the RX 7800 XT and RX 7600, AMD would later introduce Navi 32 and Navi 33.

Compared to RDNA 2, AMD claims a 17% IPC improvement. Combined with higher clocks and increased CU count, that translates into a 54% generational performance uplift. RDNA 3 also features the 2^nd generation ray accelerators and other architectural improvements.

The Radeon 780M has six workgroup processors (WGP) with twelve compute units (CUs) and twelve Ray Accelerators. That’s significantly less than the entry-level Radeon RX 7600 discrete graphics, which features 16 WGPs. The maximum GPU frequency is 2.9 GHz. The TDP is 65W, and the TjMax is 95 degrees Celsius.

In this video, we will cover five different overclocking strategies:

• First, we enable AMD Precision Boost Overdrive 2 and EXPO technologies,
• Second, we will tune the PBO 2 parameters to squeeze more performance,
• Third, we manually tune the frequencies relevant to IGP performance,
• Fourth, we increase the CPU core frequencies,
• Lastly, we switch to a higher-performance memory configuration.

However, before we jump into overclocking, let us quickly review the hardware and benchmarks used in this video.

AMD Radeon 780M: Platform Overview

The system we’re overclocking today consists of the following hardware.

Item	SKU	Price (USD)
CPU	AMD Ryzen 7 8700G	329
Motherboard	ASUS ROG Strix X670E-I Gaming WiFi	425
CPU Cooling	EK-Pro QDC Kit P360	750
Fan Controller	ElmorLabs EFC	20
Memory	G.SKILL Trident Z5 DDR5-6400	120
Power Supply	Antec HCP 1000W Platinum	250
Graphics Card	ASUS ROG Strix RTX 2080 TI	490
Storage	Kingston SSDNow V300 120GB SSD	50
Chassis	Open Benchtable V2	200

AMD Radeon 780M: Benchmark Software

We use Windows 11 and the following benchmark applications to measure performance and ensure system stability.

BENCHMARK	LINK
Geekbench 6 (OpenCL, Vulkan)	https://www.geekbench.com/
FurMark (1080P)	https://geeks3d.com/furmark/
AI Benchmark	https://ai-benchmark.com/
3DMark Night Raid	https://www.3dmark.com/
3DMark Speedway Stress Test	https://www.3dmark.com/
3DMark FSR Feature Test	https://www.3dmark.com/
Simple Raytracing Benchmark (1080P, Basic)	https://marvizer.itch.io/simple-raytracing-benchmark
Unigine Superposition (1080P, DirectX)	https://benchmark.unigine.com/superposition
Spaceship (1080P, High)	https://store.steampowered.com/app/1605230/Spaceship__Visual_Effect_Graph_Demo/
Handbrake (VCN, 1080P)	https://handbrake.fr/
EzBench (1080P)	https://store.steampowered.com/app/770170/EzBench_Benchmark/
Shadow of the Tomb Raider (1080P)	https://store.steampowered.com/app/750920/Shadow_of_the_Tomb_Raider_Definitive_Edition/
Returnal	https://store.steampowered.com/app/1649240/Returnal/
Final Fantasy XV (Standard, 1080P)	http://benchmark.finalfantasyxv.com/na/
OCCT	https://www.ocbase.com/

The benchmark selection is similar to the one we used in other GPU SkatterBencher guides. To add some clarification:

• For AI Benchmark, I rely again on the TensorFlow-DirectML library
• I use the 3DMark FSR Feature Test to measure the impact of overclocking on the performance improvement between FSR off and FSR on.
• I include two workloads for stress testing: the 3DMark Speedway Stress Test as a proxy for a gaming workload and the OCCT 3D Standard workload as a worst-case stress test.

AMD Radeon 780M: Stock Performance

Before starting overclocking, we must check the system performance at default settings. The default Precision Boost 2 parameters for the Radeon 780M are as follows:

Here is the benchmark performance at stock:

• Geekbench 6 OpenCL: 35,497 points
• Geekbench 6 Vulkan: 44,200 points
• Furmark 1080P: 2,070 points
• AI Benchmark: 9,986 points
• 3DMark Night Raid: 35,405 marks
• 3DMark FSR Feature Test: 9.87 fps
• Simple RayTracing Benchmark: 12.41 fps
• Unigine Superposition: 5,992 points
• Spaceship: 64.4 fps
• Handbrake (VCN, 1080P): 847.6 fps
• EZBench: 1,208 points
• Shadow of the Tomb Raider: 48 fps
• Returnal: 30 fps
• Final Fantasy XV: 44.45 fps

When running the 3DMark Speedway Stability Test, the average GPU effective clock is 2881 MHz with 1.085 volts. The GPU memory clock is 2400 MHz. The average GPU ASIC Power is 50 watts.

When running the OCCT 3D Standard Stress Test, the average GPU effective clock is 2476 MHz with 0.947 volts. The GPU memory clock is 2400 MHz. The average GPU ASIC Power is 69 watts.

OC Strategy #1: PBO 2 + EXPO

In our first overclocking strategy, we take advantage of enabling Precision Boost Overdrive 2 and AMD EXPO.

Precision Boost Overdrive 2

Precision Boost Overdrive 2 is AMD’s proprietary overclocking toolkit, which enables customers to finetune the parameters governing the Precision Boost 2 algorithm. It is mainly used for CPU core overclocking but also has tools for overclocking the integrated graphics.

With the launch of Zen 3, AMD introduced an improved version of the Precision Boost Overdrive toolkit, allowing for manual tuning of even more parameters affecting the Precision Boost frequency boost algorithm. Precision Boost Overdrive 2 builds on the PBO implementation of Zen 2. In addition to the overclocking knobs from Zen+ (PPT, TDC, EDC) and Zen 2 (Boost Override and Scalar), Precision Boost Overdrive 2 also introduced Curve Optimizer.

There are essentially 3 levels of Precision Boost Overdrive

1. AMD’s stock values, which can be set by disabling PBO
2. The motherboard vendor values, which are programmed into the BIOS to match the motherboard VRM specification and can be set by enabling PBO
3. Custom values, which can be programmed by the end-user

By enabling Precision Boost Overdrive, we rely on the motherboard pre-programmed PBO parameters. We find that the following values have changed:

Increasing the PPT and, to a lesser extent, the TDC and EDC limit will help unleash the frequency in extreme workloads previously limited by the power limits. Unlocking the power limits can really make this little APU scream, especially in perhaps unrealistic workloads where we stress the CPU cores and the integrated graphics at the same time. An easy way to test this case is by using the OCCT Power test.

Here, you can see the APU run at over 260W! That’s four times the rated TDP!

EXPO – Extended Profiles for Overclocking

EXPO stands for AMD Extended Profiles for Overclocking. It is an AMD technology that enables ubiquitous memory overclocking profiles for AMD platforms supporting DDR5 memory.

EXPO allows memory vendors such as G.SKILL to program higher performance settings onto the memory sticks. If the motherboard supports EXPO, you can enable higher performance with a single BIOS setting. So, it saves you lots of manual configuration.

As we’ll see later in the video, unlocking memory performance is critical to improving APU graphics performance.

BIOS Settings & Benchmark Results

Upon entering the BIOS

• Switch to the Advanced Mode view and stay in the AI Tweaker menu
• Set Ai Overclock Tuner to EXPO I
• Enter the Precision Boost Overdrive submenu
  • Set Precision Boost Overdrive to Enabled

Then save and exit the BIOS.

We re-ran the benchmarks and checked the performance increase compared to the default operation.

• Geomean: +15.68%
• Geekbench 6 OpenCL: +12.49%
• Geekbench 6 Vulkan: +16.32%
• Furmark 1080P: +25.68%
• AI Benchmark: +19.22%
• 3DMark Night Raid: +17.06%
• 3DMark FSR Feature Test: 19.56%
• Simple RayTracing Benchmark: +19.56%
• Unigine Superposition: +20.03%
• Spaceship: +21.28%
• Handbrake (VCN, 1080P): +2.75%
• EZBench: +14.39%
• Shadow of the Tomb Raider: +26.32%
• Returnal: +11.11%
• Final Fantasy XV: +22.12%

Unlocking the power limits and enabling higher memory performance significantly impacts the integrated graphics performance. The geomean performance improvement is +16%, and we get a maximum improvement of +26.32% in Tomb Raider.

When running the 3DMark Speedway Stability Test, the average GPU effective clock is 2903 MHz with 1.063 volts. The GPU memory clock is 3200 MHz. The average GPU ASIC Power is 69 watts.

When running the OCCT 3D Standard Stress Test, the average GPU effective clock is 2882 MHz with 1.141 volts. The GPU memory clock is 3200 MHz. The average GPU ASIC Power is 138 watts.

OC Strategy #2: PBO Tuned

In our second overclocking strategy, we tune the Precision Boost dynamic graphics frequency using the Precision Boost Overdrive 2 toolkit.

PBO 2: GPU Boost Clock Override

Fused maximum frequency, or Fmax, is one of the Precision Boost infrastructure limiters constraining the graphics performance. The limiter determines the maximum allowed graphics frequency.

GPU Boost Clock Override or GPU Fmax Override is one of the overclocking tools available in the PBO 2 toolkit. It allows the user to override the arbitrary clock frequency limit up to +200 MHz in steps of 25 MHz. It’s important to note that the GPU Boost override only adjusts the upper ceiling of the frequency and doesn’t act as a frequency offset. Ultimately, the Precision Boost 2 algorithm still determines the actual operating frequency.

When we increase the GPU Fmax boost limit by 200 MHz, the new GPU Boost Fmax is 3100 MHz.

PBO 2: GFX Curve Optimizer

Curve Optimizer is an important new feature of Precision Boost Overdrive 2.

Curve Optimizer allows end-users to adjust their CPU’s factory-fused VFT or voltage-frequency-temperature curve. The VFT curve defines the required voltage for a given frequency at a given temperature. Higher frequencies or higher operating temperatures require higher voltage. Many parts inside your CPU have a VFT curve. Still, unfortunately, we can only finetune the curve for each of the eight CPU cores and the integrated graphics.

GFX Curve Optimizer adjusts the VFT curve of the integrated graphics by offsetting the voltages of the factory-fused VFT curve. By setting a positive offset, you increase the voltage point. Conversely, you decrease the voltage point by setting a negative offset. You can offset the entire curve by up to 50 steps in a positive or negative direction.

AMD Clocking Technologies

Experienced Ryzen overclockers are familiar with the Curve Optimizer function for CPU cores. The GFX Curve optimizer function works similarly for the Ryzen 7 8700G integrated graphics, but they are not identical. To better explain what happens, I must tell you a little about AMD’s clocking technologies, AVFS, BTC, and VAO.

Adaptive Voltage Frequency Scaling

Since the 2015 Carrizo APU, AMD has used Adaptive Voltage Frequency Scaling, or AVFS. This technology relies on Critical Path Accumulators to estimate the maximum stable frequency of circuits inside your processor – whether a CPU or a GPU. If it sounds complicated, that’s because it is. However, I’ll try to explain it as simply as possible.

Long story short, AMD adds replica paths to the circuit that serve no purpose but to assess whether the circuit is stable. The AVFS technology extracts a Gaussian distribution statistical model from the replicate paths. It infers the stability of the “real” paths using sampling statistics. This statistical data is used by the SMU to create a Voltage-Frequency-Temperature (VFT) table.

The VFT table is a part-specific lookup table with information on the optimal voltage for any combination of frequency and voltage. The higher the frequency and temperature, the higher the required voltage to ensure stability.

Boot Time Calibration

AMD uses automated test equipment (ATE) to fuse the VFT table onto a CPU. This test equipment typically provides a more robust, less noisy power supply than a consumer one. AMD uses a tester-to-platform (T2P) voltage offset to compensate for this. It is often a conservative, over-margined value.

AMD employs a boot-time calibration (BTC) process to finetune this voltage offset on each system. Essentially, during the boot process, the chip checks the quality of the power delivery. Based on the quality, it then offsets the factory-fused VFT table. If you have a plentiful power supply, the offset will be smaller. If you have a terrible power supply, the offset will be greater.

Voltage Adaptive Operation

Despite advanced technologies like AVFS, the part-specific VFT table, a T2P voltage offset, and BTC, AMD cannot predict what will happen in the real world. Sometimes, the voltage droops under transient conditions, going from idle to load or vice versa, and it is so significant that the system crashes.

Fortunately, AMD also has Voltage Adaptive Operation technology, commonly known as Clock Stretching. The technology consists of two circuits: one circuit serves as a configurable droop detector, and the second circuit functions as a configurable digital frequency synthesizer. For example, we could configure that if a voltage droop of 2.5% or more is detected, the clock period is increased by 5%.

The effect is simple: if a voltage droop is detected, the effective clock frequency is lowered to ensure continuous operation instead of a system crash. In the real world, this technology is incredibly relevant when overclocking because it may cause effective performance at a given frequency to be lower than expected.

The practical implication of voltage adaptive operation is that the effective clock frequency may differ from the configured clock frequency. How this works on AMD CPU and GPUs is slightly different, but it essentially boils down to this:

• The clock frequency is determined by the configured target frequency, which is often based on a reference clock and a multiplier. It’s usually the GPU frequency you’ll see in GPU-Z.
• The Effective Clock Frequency is the total clock cycles between two moments. This determines the actual performance as work gets done with each clock cycle. We can conveniently check the effective clock frequency with tools like HWiNFO.

AMD GPUs heavily rely on the Voltage Adaptive Operation to continuously operate at the maximum possible frequency given a certain voltage level. You’ll find that in heavier workloads, which cause more Vdroop, the difference between the set target clock and the effective clock will be greater.

In some cases, we can take advantage of this technology. For example, with the Radeon RX 6500 XT, we overvolted the GPU, which caused the effective clock to be higher than the set target clock. We could reach a 3 GHz effective clock with a 2975 MHz set clock. Remember this as we’ll get back to it.

Anyway, I’m trying to make the point that the GPU’s performance is tied closely to the actual voltage it’s running at. When we use Curve Optimizer, we finetune those guardbands, which lets us finetune the effective clock frequency.

AMD Ryzen 8000G GFX Curve Optimizer Tuning

Just a couple of minutes ago, I gave the example of using overvolt on the Radeon RX 6500 XT to squeeze more frequency out of the chip. Similarly, we will use the GFX Curve Optimizer tool in the Precision Boost Overdrive 2 toolkit.

It’s counter-intuitive for experienced Ryzen CPU overclockers but bear with me. To make a long story short, when we set a positive GFX curve optimizer setting, we tell the graphics core it requires more voltage for a specific V/F point. The voltage supply will then increase the voltage accordingly, and the voltage adaptive operation technology will match the voltage with the appropriate maximum frequency.

The effect is a higher effective clock frequency at a given target clock frequency … but only a minimal difference.

For example, we set the clock to 3100 MHz with GPU Boost Override. In the 3DMark Speedway stress test, we get about 3099 MHz effective clock with a -50 curve optimizer and 3110 MHz with a +50 curve optimizer. In the OCCT 3D Standard stress test, we get a 3053 MHz effective clock with a -50 curve optimizer and 3060 MHz with a +50 curve optimizer.

So, at the end of the day, the GFX Curve Optimizer isn’t that useful for overclocking the integrated graphics.

EXPO Tweaked

EXPO Tweaked is an option available under Ai Overclock Tuner in addition to EXPO I and EXPO II. All three settings load the memory kit EXPO profile but do it slightly differently.

• EXPO I loads only the primary timings, frequency, and voltage. The secondary timings are adjusted by the motherboard auto-rules.
• EXPO II loads the complete EXPO profile, including the primary and secondary timings, the memory frequency, and the voltage.
• EXPO Tweaked loads the complete EXPO profile and adjusts various timings if possible.

Since memory performance will impact the integrated graphics performance, I tried the EXPO Tweaked setting. Everything seemed stable, so I used it for this OC Strategy.

BIOS Settings & Benchmark Results

Upon entering the BIOS

• Switch to the Advanced Mode view and stay in the AI Tweaker menu
• Set Ai Overclock Tuner to EXPO Tweaked
• Go to the Advanced Menu
• Enter the AMD Overclocking submenu
  • Click Accept
  • Enter the Precision Boost Overdrive submenu
    • Set Precision Boost Overdrive to Advanced
    • Set PBO Limits to Motherboard
    • Set GPU Boost Clock Override to Enabled
    • Set Max GPU Boost Clock Override to 200 MHz
    • Enter the GFX Curve Optimizer submenu
      • Set GFX Curve Optimizer to GFX Curve Optimizer
      • Set GFX Curve Optimizer Sign to Positive
      • Set GFX Curve Optimizer Magnitude to 50

Then save and exit the BIOS.

We re-ran the benchmarks and checked the performance increase compared to the default operation.

• Geomean: +19.20%
• Geekbench 6 OpenCL: +17.42%
• Geekbench 6 Vulkan: +19.95%
• Furmark 1080P: +26.29%
• AI Benchmark: +23.60%
• 3DMark Night Raid: +17.50%
• 3DMark FSR Feature Test: +20.23%
• Simple RayTracing Benchmark: +21.19%
• Unigine Superposition: +21.59%
• Spaceship: +22.98%
• Handbrake (VCN, 1080P): +3.59%
• EZBench: +18.18%
• Shadow of the Tomb Raider: +28.95%
• Returnal: +25.93%
• Final Fantasy XV: +23.46%

With only a slightly higher frequency of 200 MHz, we didn’t expect much of a performance improvement. Compared to the previous OC Strategy, the geomean performance improved by 3.5 percentage points. We get the highest performance improvement over stock of +28.95% in Tomb Raider.

When running the 3DMark Speedway Stability Test, the average GPU effective clock is 3099 MHz with 1.200 volts. The GPU memory clock is 3200 MHz. The average GPU ASIC Power is 83 watts.

When running the OCCT 3D Standard Stress Test, the average GPU effective clock is 3053 MHz with 1.200 volts. The GPU memory clock is 3200 MHz. The average GPU ASIC Power is 159 watts.

OC Strategy #3: Manual Tuning

In our third overclocking strategy, we’re pursuing a manual overclock. My goal is to increase the clock frequency of every part that may affect the IGP performance, including:

• The graphics clock frequency (GFXCLK)
• The infinity fabric clock frequency (FCLK)
• The memory controller clock frequency (UCLK)
• The system memory clock frequency (MCLK)

Before I show you my final overclocking settings, however, I need to cover the basics of the AMD Hawk Point clocking and voltage topology. This will help us better understand how to do our manual overclocking.

AMD Hawk Point Clocking Topology

The clocking of Hawk Point APUs is similar to the other Zen 4 desktop processors.

The standard Hawk Point platform has a 48 MHz crystal input to the integrated CGPLL clock generator. The CGPLL then generates a 48 MHz clock for the USB PLL and a 100 MHz reference clock for the FCH, which contains the CCLK PLL for the CPU cores and several SOC PLLs.

The CCLK PLL 100MHz reference clock drives the 200 MHz VCO, which is then multiplied by an FID and divided by a DID. As a whole, this provides a CPU clock frequency granularity of 25 MHz. As with Raphael, each CCX has its own PLL, with the cores within that CCX running at the same frequency. This isn’t particularly relevant since there’s only one CCX, featuring eight Zen 4 cores.

The SOC PLLs include a wide range of PLLs present on the die. The ones most relevant for overclocking are:

• FCLK for the data fabric,
• UCLK for the memory controller,
• MCLK for the system memory,
• GFXCLK for the integrated graphics,
• IPUCLK for the inference processor.

FCLK is the term used for the infinity fabric clock frequency. The default FCLK frequency is 2000 MHz, but some motherboards have auto-rules setting it to 2400 MHz. I was able to push it to 2500 MHz.

UCLK is the term used for the unified memory controller clock frequency. It runs by default at the same frequency as the system memory. However, motherboard auto-rules may drop to half the frequency if the system memory frequency is too high. It is relatively inflexible as it can run at the same or half the system memory frequency. I could run the UCLK in sync with the system memory up to DDR5-6400 on my particular system.

MCLK is the term used for the system memory clock frequency. It is, by default, either the same or double the memory controller frequency. AMD recently improved its memory overclocking capabilities, and Hawk Point APUs can run well over DDR5-8000.

GFXCLK is the term used for the integrated graphics core clock frequency. It will go up to 2.9 GHz during a 3D load at stock. The Precision Boost algorithm manages the graphics clock, even when you set a manual target frequency.

IPUCLK is the term used for the inference processing unit clock frequency. It should be able to go up to 1.6 GHz. However, I’ve only seen it go up to 1028 MHz. It can currently not be overclocked.

With the launch of Raphael, we also received a functional eCLK mode. ECLK stands for external clock and is precisely what the term suggests: an external clock generator. It was previously available on Ryzen 2000 Pinnacle Ridge processors but was removed afterward.

In addition to the standard internal CGPLL, Hawk Point supports up to two external clock modes. They’re called eCLK0 Mode and eCLK1 Mode.

• In eCLK0 Mode, also called synchronous mode, an external 100MHz reference clock is used for both the CPU PLL and SOC PLLs. In other words, it’s a reference clock that affects the CPU core clocks as well as the PCIe and SATA clocks.
• In eCLK1 Mode, also referred to as asynchronous mode, there are two distinct external 100MHz reference clocks. One clock provides the 100MHz input for the CPU PLL, and another provides the 100MHz reference clock for the SOC PLLs.

AMD suggests up to 140 MHz can be expected for the CPU core reference clock, but your mileage may vary. At the launch of Raphael, I showed that 170 MHz is possible (https://valid.x86.fr/ilmj1v)

AMD Hawk Point Voltage Topology

AMD Hawk Point’s voltage topology is similar to that of previous Ryzen processors but with one significant change. In fact, I got this one wrong in my 8700G CPU core overclocking video! 

As usual, the processor relies on an internal and external power supply to generate the processor voltages. There are four primary power supplies from the motherboard VRM to the processor: VDDCR, VDDCR_SOC, VDDCR_MISC, and VDDIO_MEM.

The VDDCR voltage rail provides the external power for three internal voltage regulators: VDDCR_CPU, VDDCR_VDDM, and the VDDCR_GFX.

• VDDCR_CPU provides the voltage for the CPU cores within the CCX. The voltage rail can work in either regular or bypass mode, but it is always in bypass mode on Hawk Point. That means the voltage of the cores is always equal to the VDDCR external voltage. The end user can change the voltage in the BIOS.
• VDDCR_VDDM provides the voltage for the L2, L3, and, if present, 3D V-Cache on a CCX. This rail cannot work in bypass mode; therefore, it is always internally regulated from the VDDCR external voltage rail. We can also not adjust this voltage.
• VDDCR_GFX provides the voltage for the integrated graphics. In the past, this voltage would be provided by the VDDCR_SOC voltage rail. Still, it’s likely due to the high current requirements of the powerful integrated graphics, using the typically beefier VDDCR voltage plane is safer. The voltage rail can technically work in regular or bypass mode, but only regular mode is available. In regular mode, the voltage is managed by the integrated voltage regulator and derived from the VDDCR voltage rail. This voltage is limited to 1.25V under ambient conditions and requires LN2 mode for a higher range.

The VDDCR_SOC voltage rail provides the external power for multiple internal voltage regulators on SOC for the various IP blocks, including but not limited to the memory controller, SMU, PSP, etc. It is essential to know that the VDDCR_SOC voltage must always be lower than VDDIO_MEM_S3 + 100mV. The default VDDCR_SOC voltage is 1.05V and can be set to 1.30V under ambient conditions. Again, we need LN2 mode enabled for higher voltages.

The VDDCR_MISC voltage rail provides the external power for the internally regulated VDDG voltage rail. VDDG is the voltage supply for the infinity fabric data path. Previously, you could manually tune the voltages for each infinity fabric connection. However, this doesn’t seem to be available for Hawk Point.

The VDDIO_MEM voltage rail provides the external power for the VDDP_DDR internal voltage regulator. VDDP is the voltage for the DDR bus signaling or DRAM PHY. So it can help achieve higher memory frequencies. As a rule, the external VDDIO_MEM should always be higher than the internal VDDP_DDR + 100mV. Furthermore, the external VDDCR_SOC voltage rail should be lower than the external VDDIO_MEM + 100mV. When memory overclocking, you may need to manually increase the VDDP voltage as it does not automatically adjust when changing the VDDIO_MEM voltage.

AMD Radeon 780M Manual Overclocking

Manual overclocking of the Radeon 780M is pretty straightforward, as it’s a process of trial and error. For example, for the FCLK, I tried increasing the frequency beyond 2500 MHz, but the system didn’t boot.

Overclocking the memory and memory controller are tied together. In my case, I could run them in sync up to DDR5-6400. For higher memory frequencies, I had to reduce the UCLK to half the speed of the MCLK. Since DDR5-6800 with a 1700 MHz memory controller resulted in the best performance, that’s what I stuck with.

The graphics overclocking was very underwhelming. There are three things to take into account.

1. You can set a manual target boost frequency and voltage in the AMD overclocking menu. This frequency and voltage are now the target for the Precision Boost graphics frequency.
2. The maximum temperature in my worst-case workload is already 90.3 degrees Celsius for 3.1 GHz at 1.2V. So, there’s minimal thermal headroom to increase the frequency.
3. As mentioned, the maximum graphics voltage is limited to 1.25V under ambient conditions. You can set it higher in the BIOS, but it won’t apply higher than 1.25V. You can only exceed this voltage if you enable LN2 mode, which coincidentally has a pretty strict temperature condition.

Considering all these three things, I could only try to increase the frequency with the standard voltage of 1.2V. To make things even less exciting, the maximum stable frequency I could achieve was only 3.15 GHz … only 50 MHz more than when we use Precision Boost Overdrive! That isn’t very pleasant.

In the Speedway Stress Test, we have a little more thermal headroom to try the maximum voltage of 1.25V. A quick test revealed we could clock the graphics up to 3.3 GHz. That’s higher than before, but still minimal headroom.

BIOS Settings & Benchmark Results

Upon entering the BIOS

• Switch to the Advanced Mode view and stay in the AI Tweaker menu
• Set Ai Overclock Tuner to EXPO Tweaked
• Set Memory Frequency to DDR5-6800MHz
• Go to the Advanced Menu
• Enter the AMD Overclocking submenu
  • Click Accept
  • Enter the Manual iGPU Overclocking submenu
    • Set GFX Clock Frequency to 3150
    • Set GFX Voltage to 1200
  • Leave the Manual iGPU Overclocking submenu
  • Enter the DDR and Infinity Fabric Frequency/Timings submenu
    • Enter the Infinity Fabric Frequency and Dividers submenu
      • Set Infinity Fabric Frequency and Dividers to 2500 MHz
      • Set UCLK DIV1 MODE to UCLK=MEMCLK/2
    • Leave the Infinity Fabric Frequency and Dividers submenu
  • Leave the DDR and Infinity Fabric Frequency/Timings submenu
  • Enter the Precision Boost Overdrive submenu
    • Set Precision Boost Overdrive to Advanced
    • Set PBO Limits to Motherboard

Then save and exit the BIOS.

We re-ran the benchmarks and checked the performance increase compared to the default operation.

• Geomean: +22.31%
• Geekbench 6 OpenCL: +22.45%
• Geekbench 6 Vulkan: +23.53%
• Furmark 1080P: +29.51%
• AI Benchmark: +39.09%
• 3DMark Night Raid: +21.32%
• 3DMark FSR Feature Test: +24.77%
• Simple RayTracing Benchmark: +24.47%
• Unigine Superposition: +25.58%
• Spaceship: +26.37%
• Handbrake (VCN, 1080P): +2.66%
• EZBench: +31.58%
• Shadow of the Tomb Raider: +29.63%
• Returnal: +29.63%
• Final Fantasy XV: +28.16%

Despite the tiny improvement in GPU core frequency, we still get a little bit of a performance boost thanks to the improvements in fabric and memory frequency. Compared to the previous OC Strategy, the geomean performance improved by 3.1 percentage points. We get the highest performance improvement over stock of +39.09% in the AI Benchmark.

When running the 3DMark Speedway Stability Test, the average GPU effective clock is 3141 MHz with 1.200 volts. The GPU memory clock is 3400 MHz. The average GPU ASIC Power is 84 watts.

When running the OCCT 3D Standard Stress Test, the average GPU effective clock is 3086 MHz with 1.200 volts. The GPU memory clock is 3400 MHz. The average GPU ASIC Power is 156 watts.

OC Strategy #4: CPU Overclocking

We rely on the Ryzen 7 8700G CPU core overclocking information from SkatterBencher #69 in our fourth overclocking strategy. We apply the settings from OC Strategy #2, which leverages the Precision Boost Overdrive 2 toolkit.

I won’t spend any time explaining these settings as you can easily refer to my SkatterBencher guide for all information related to CPU core overclocking of the Ryzen 7 8700G. So, let’s go straight to the BIOS configuration and the performance results.

BIOS Settings & Benchmark Results

Upon entering the BIOS

• Switch to the Advanced Mode view and stay in the AI Tweaker menu
• Set Ai Overclock Tuner to EXPO Tweaked
• Set Memory Frequency to DDR5-6800MHz
• Go to the Advanced Menu
• Enter the AMD Overclocking submenu
  • Click Accept
  • Enter the Manual iGPU Overclocking submenu
    • Set GFX Clock Frequency to 3150
    • Set GFX Voltage to 1200
  • Leave the Manual iGPU Overclocking submenu
  • Enter the DDR and Infinity Fabric Frequency/Timings submenu
    • Enter the Infinity Fabric Frequency and Dividers submenu
      • Set Infinity Fabric Frequency and Dividers to 2500 MHz
      • Set UCLK DIV1 MODE to UCLK=MEMCLK/2
    • Leave the Infinity Fabric Frequency and Dividers submenu
  • Leave the DDR and Infinity Fabric Frequency/Timings submenu
  • Enter the Precision Boost Overdrive submenu
    • Set Precision Boost Overdrive to Advanced
    • Set PBO Limits to Motherboard
    • Set Precision Boost Overdrive Scalar Ctrl to Manual
      • Set Precision Boost Overdrive Scalar to 10X
    • Set CPU Boost Clock Override to Enabled (Positive)
      • Set Max CPU Boost Clock Override(+) to 200
    • Enter the Curve Optimizer submenu
      • Set Curve Optimizer to Per Core
      • Set Core 0 to Core 7 Curve Optimizer Sign to Negative
        • Now, I set the Curve Optimizer for each core according to my test result.
      • Set Core 0, 1, and 2 Curve Optimizer Magnitude to 50
      • Set Core 3 Curve Optimizer Magnitude to 25
      • Set Core 4, 5, and 6 Curve Optimizer Magnitude to 40
      • Set Core 7 Curve Optimizer Magnitude to 35

Then save and exit the BIOS.

We re-ran the benchmarks and checked the performance increase compared to the default operation.

• Geomean: +22.69%
• Geekbench 6 OpenCL: +24.92%
• Geekbench 6 Vulkan: +23.93%
• Furmark 1080P: +29.75%
• AI Benchmark: +39.74%
• 3DMark Night Raid: +22.40%
• 3DMark FSR Feature Test: +26.98%
• Simple RayTracing Benchmark: +24.86%
• Unigine Superposition: +25.24%
• Spaceship: +26.37%
• Handbrake (VCN, 1080P): +2.51%
• EZBench: +31.58%
• Shadow of the Tomb Raider: +29.63%
• Returnal: +29.63%
• Final Fantasy XV: +27.47%

Overclocking the CPU cores hardly improves the graphics performance of this APU. Compared to the previous OC Strategy, the geomean performance improves by only 0.3 percentage points. We get the highest performance improvement over stock of +39.74% in the AI Benchmark.

I didn’t re-run the stress tests as the integrated graphics configuration is identical to the previous overclocking strategy.

OC Strategy #5: High-Performance Memory

In our fifth overclocking strategy, I want to show you how extensive memory tuning can significantly impact the APU’s graphics performance. To illustrate this, I will do something I never do in a SkatterBencher guide: swap out the memory kit!

I stick with G.SKILL, but instead of using the EXPO-6400 kit I have had since the launch of the Ryzen 7000, I switch to an XMP-7800 kit I used for the 13900KS Intel overclocking demonstration. We would expand that increasing the memory frequency from EXPO-6400 to XMP-7800 should immediately impact performance. However, that’s not really the case. The two configurations yield about the same level of performance.

The most significant performance improvements for APU graphics can be found when tuning the memory timings. Now, I’m far from a DDR5 tuning expert, so I will take a shortcut by using ASUS Memory Presets instead.

ASUS Memory Presets

ASUS Memory Presets is an ASUS overclocking technology that provides a selection of memory-tuning presets for specific memory ICs. The presets will adjust the memory timings and voltages.

The technology was first introduced in 2012 on Z77 and has been on select ASUS ROG motherboards ever since. The memory profiles available differ from platform to platform. This X670E-I Gaming motherboard with BIOS version 2012 features no less than fourteen DDR5 overclocking profiles for Hynix, Samsung, and Micron ICs.

• Hynix 6200MHz 1.4V 2x16GB SR
• Hynix 6400MHz 1.4V 2x16GB SR
• Hynix 6200MHz 1.35V 2x32GB DR
• Hynix 6400MHz 1.4V 2x32GB DR
• Hynix 6400MHz 1.5V 2x16GB SR
• Hynix 5600MHz 1.4V 4x16GB SR
• Samsung 6200MHz 1.4V 2x16GB SR
• Samsung 6400MHz 1.4V 2x16GB SR
• Samsung 6400MHz 1.4V 2x32GB DR
• Samsung 5400MHz 1.3V 4x16GB SR
• Micron 5600MHz 1.35V 2x16GB SR
• Micron 5200MHz 1.3V 2x32GB DR
• Hynix 8000MHz 1.45V 2x16GB SR
• Hynix 8000MHz 1.45V 2x24GB SR

I pick the Hynix 8000MHz 1.45V 2x16GB SR profile for this OC Strategy. This automatically sets the memory primary and secondary timings. For those interested in the memory timings and performance, here’s a couple of screenshots with Zentimings and AIDA64 from different memory configurations. From left to right: DDR5-6400 EXPO, DDR5-7800 XMP, and DDR5-7800 Memory Preset

It’s also best to double-check if the VDDIO, DRAM VDD, and DRAM VDDQ voltages are set correctly. Lastly, be sure to also select the correct memory frequency, as the memory presets only adjust the memory timings and voltages.

I want to make a quick note here to say that these memory timings are not entirely stable. We could spot that from the graphical artifacts visible on the screen when running a 3D workload.

I could probably figure out which timings are the root cause with additional time spent. Still, this OC Strategy aims to demonstrate the performance impact of tuning the memory timings, not to get a fully stable overclock.

Upon entering the BIOS

• Switch to the Advanced Mode view and stay in the AI Tweaker menu
• Set Memory Frequency to DDR5-7800MHz
• Enter the DRAM Timing Control submenu
  • Enter the Memory Presets submenu
    • Select Load Hynix 8000MHz 1.45v 2x16GB SR
  • Leave the Memory Presets submenu
• Leave the DRAM Timing Control submenu
• Set CPU SOC Voltage to Auto
• Set CPU VDDIO / MC Voltage to 1.4
• Set DRAM VDD and VDDQ Voltage to 1.5
• Go to the Advanced Menu
• Enter the AMD Overclocking submenu
  • Click Accept
  • Enter the Manual iGPU Overclocking submenu
    • Set GFX Clock Frequency to 3150
    • Set GFX Voltage to 1200
  • Leave the Manual iGPU Overclocking submenu
  • Enter the DDR and Infinity Fabric Frequency/Timings submenu
    • Enter the Infinity Fabric Frequency and Dividers submenu
      • Set Infinity Fabric Frequency and Dividers to 2500 MHz
      • Set UCLK DIV1 MODE to UCLK=MEMCLK/2
    • Leave the Infinity Fabric Frequency and Dividers submenu
  • Leave the DDR and Infinity Fabric Frequency/Timings submenu
  • Enter the Precision Boost Overdrive submenu
    • Set Precision Boost Overdrive to Advanced
    • Set PBO Limits to Motherboard
    • Set Precision Boost Overdrive Scalar Ctrl to Manual
      • Set Precision Boost Overdrive Scalar to 10X
    • Set CPU Boost Clock Override to Enabled (Positive)
      • Set Max CPU Boost Clock Override(+) to 200
    • Enter the Curve Optimizer submenu
      • Set Curve Optimizer to Per Core
      • Set Core 0 to Core 7 Curve Optimizer Sign to Negative
        • Now, I set the Curve Optimizer for each core according to my test result.
      • Set Core 0, 1, and 2 Curve Optimizer Magnitude to 50
      • Set Core 3 Curve Optimizer Magnitude to 25
      • Set Core 4, 5, and 6 Curve Optimizer Magnitude to 40
      • Set Core 7 Curve Optimizer Magnitude to 35
    • Leave the Curve Optimizer submenu
  • Leave the Precision Boost Overdrive submenu
  • Enter the SoC Voltage submenu
    • Set SoC Voltage to 1300

Then save and exit the BIOS.

We re-ran the benchmarks and checked the performance increase compared to the default operation.

• Geomean: +37.75%
• Geekbench 6 OpenCL: +37.11%
• Geekbench 6 Vulkan: +36.63%
• Furmark 1080P: +61.14%
• AI Benchmark: +52.08%
• 3DMark Night Raid: +30.36%
• 3DMark FSR Feature Test: +27.21%
• Simple RayTracing Benchmark: +55.49%
• Unigine Superposition: +45.15%
• Spaceship: +46.33%
• Handbrake (VCN, 1080P): +7.26%
• EZBench: +34.85%
• Shadow of the Tomb Raider: +50.00%
• Returnal: +48.15%
• Final Fantasy XV: +27.20%

The performance improvement from tuning the memory timings exceeded my wildest expectations. Clearly, our integrated graphics craves faster memory access. Compared to the previous OC Strategy, the geomean performance improved by over 15 percentage points. We get the highest performance improvement over stock of +61.14% in Furmark 1080P.

As I indicated earlier, this configuration isn’t entirely stable, as seen from the graphics artifacts. Since I don’t consider this a stable configuration, I didn’t entertain running stability tests.

AMD Radeon 780M: Conclusion

Alright, let us wrap this up.

I’ll be honest: I was extremely excited to try APU integrated graphics overclocking. When I started testing this platform, IGP overclocking wasn’t working yet. I had to eagerly await a new BIOS release. Hence, this guide is later than my Ryzen 7 8700G guide.

Unfortunately, my excitement was soon squished by the extremely poor overclocking headroom. That’s in part thanks to the excellent work of the AMD engineers to leverage AVFS and Voltage Adaptive Operation to maximize the frequency out of the box. But perhaps there are other tricks they didn’t do that could help increase the overclocking headroom.

The memory tuning side of APU graphics performance tuning is simply incredible. There’s much performance to be unlocked by digging through the memory timings and tuning the memory subsystem. I can’t wait to see true DDR5 overclocking experts drive the APU 3DMark scores to unreal heights.

Anyway, that’s all for today!

I absolutely love the Hawk Point overclocking challenge so far. Hence, I’ll likely have a closer look at overclocking the IPU and overclocking the APUs with Phoenix2 dies, which have Zen4c cores. I want to thank my Patreon supporters for supporting my work. If you have any questions or comments, please drop them in the comment section below. 

See you next time!