I recently bought a new computer. The configuration is:

file

  • ASUS TUF Gaming GeForce RTX™ 4090 24GB GDDR6X
  • ROG STRIX Z790-A GAMING WIFI
  • Intel® Core™ i9-13900KS Processor
  • SAMSUNG 990 PRO PCIe® 4.0 NVMe™ SSD 1TB

Since this hardware setup supports overclocking, I definitely want to overclock it to a safe, stable, and high-performance state with relatively low power consumption, while ensuring that the games I primarily play remain stable.

It's impossible to avoid tinkering. If I didn't want to tinker, I wouldn't have bought this setup in the first place.

Of course, I also admit that the 13th-gen i9 + 4090 combination has very little overclocking potential. After spending quite a bit of time tinkering, my conclusion is actually that the difference compared to before wasn't that significant. However, I don't regret it, as I learned a lot of knowledge from the process and got to explore the fascinating world of hardware modification.

Preparation Before Starting

There's nothing particularly special about it—just a standard, routine process:

  • Choose suitable hardware, paying attention to its overclocking potential, features, and aesthetics to ensure they meet your needs.
  • Purchase the hardware by finding appropriate vendors and shipping carriers.
  • Assemble it: install the CPU, apply thermal paste, mount the cooler, install RAM and storage drives, secure the motherboard into the case, install the GPU, and arrange the power cabling...
  • Fill the water loop with coolant, ensure there are no leaks, and power it on. (The case I purchased this time comes with a custom loop water cooling system.)
  • Install a basic operating system for temporary use during overclocking, and reinstall it later once you're done. (Considering that the overclocking process often requires downloading a lot of miscellaneous tools.)

However, note:

Before starting, it is highly recommended to search for the detailed tech specs of the equipment mentioned above. This includes how the motherboard and CPU allocate PCIe lanes, which ports are directly connected to the CPU, and which ports go through the chipset (PCH), etc.

file

Also, look up whether this motherboard has any special features or comes with bundled software. For example, my motherboard comes with a bunch of impressive and quirky features...

For example: Introduction to the Z790-A

Understanding these features is very helpful for the subsequent steps.

file

Of course, also search for information on the CPU's website to understand its architecture, whether it has a heterogeneous multi-core design, whether it includes an integrated graphics unit, whether it has TPM, what maximum memory frequency it supports, and the preset turbo boost frequencies for each core, etc. These details are helpful for later configuration.

For example: Introduction to the 13900KS

Searching for this information is to ensure that the hardware you purchase is actually compatible and works properly together. For example, the way I used my motherboard was very standard:

1x16 PCIE 5.0 完全给显卡 RTX 4090
1x4 PCIE 4.0 完全给硬盘

Here, I can note the following: the speed of a single PCIe 4.0 lane is approximately 2 GB/s. Therefore, when my hard drive is connected to a PCIe 4.0 x4 slot, the theoretical bandwidth ceiling is 8 GB/s. Since the hard drive itself can achieve around 7 GB/s, the interface will not create a bottleneck that limits the drive's performance.

Additionally, I have heard many people online state that the RTX 4090 generally performs sufficiently well even on a PCIe 5.0 x8 connection for most workloads. While I have not personally verified this, I understand that a PCIe 5.0 x16 connection is certainly more than enough to fully utilize the graphics card.

Alright, with the research completed, the structure designed, the hardware installed, and the system configured, we can now begin overclocking and tuning the computer!

file

Overall Adjustment, Monitoring, and Evaluation of PC and Motherboard

Before overclocking, it is essential to have basic monitoring capabilities for the entire computer to ensure that hardware is not damaged during the process. Additionally, we need to be able to verify the extent of the impact each setting change has.

AIDA64 Extreme

Download URL: https://www.aida64.com/products/aida64-extreme

AIDA64 is a tool for testing hardware and software system information, providing detailed insights into every aspect of a PC. In addition to features such as overclocking assistance, hardware diagnostics, stress testing, and sensor monitoring, AIDA64 also offers comprehensive performance evaluation of the processor, system memory, and disk drives.

file

However, please note: AIDA64 Extreme is paid software! It only supports annual subscriptions. I couldn't find a similar alternative, which is extremely unfriendly, so I had to finish my overclocking tasks within the trial period and then uninstall it...

ASUS ARMOURY CRATE

Download URL: https://rog.asus.com/us/armoury-crate/

ASUS ARMOURY CRATE can unlock some amazing features on many ASUS motherboards.

file

For example, if your system supports Aura Sync, you can adjust settings within it:

file

Many computers allow one-click overclocking in ARMOURY CRATE.

Possible features include:

  • Viewing the frequency, temperature, and usage of each CPU core
  • One-click overclocking and silent mode
  • Viewing the frequency, temperature, and usage of the graphics card
  • Viewing fan speeds
  • Adjusting fan speeds
  • Adjusting Aura Sync
  • Adjusting monitor color profiles
  • Configuring the workload distribution between the discrete GPU and integrated GPU

These features vary depending on the device. You can download it to explore and test it yourself.

Of course, if you do not own an ASUS device, there is no need to use this.

BIOS Settings

Of course, BIOS settings are very important. On the ASUS Z790-A motherboard, you can do the following in the BIOS:

  • Enable AI one-click overclocking
  • Adjust fan speed curves
  • Adjust memory XMP settings (to switch to the ideal designed frequency)
  • Adjust detailed overclocking settings
  • Enable voltage droop prevention settings

In most cases, simply using ASUS's AI overclocking is sufficient. Its overclocking performance is quite good and not too conservative.

Of course, if you want to push the limits and enjoy fine-tuning, you can adjust detailed parameters bit by bit in the BIOS.

file

The general way to enter the BIOS is to repeatedly press F2 or Delete during startup.

Generally, I recommend modifying the following:

  • XMP -> Set according to the memory's recommended settings
  • Above 4G Decoding -> Enable
  • Resizable bar -> Enable
  • Intel CPU Virtualization Technology -> Enable
  • TPM -> Enable
  • CSM Support -> Disable
  • UEFI -> Enable
  • Secure Boot -> Enable (Windows OS)
  • CPU calibration load line -> Level 4-6

Most other settings are actually fine with the defaults. Of course, you can adjust them as needed, for example, if you need to split PCIe lanes, you can come back and configure them.

OpenHardwareMonitor

OpenHardwareMonitor can display the CPU's temperature and load, as well as the GPU's temperature and frequency.

file

It can also stay visible in a corner of your desktop like a widget! It's very convenient for monitoring your computer's performance.

file

What's great about it is that it's open source! You can find the source code here:

https://github.com/openhardwaremonitor/openhardwaremonitor

The only downside is that it doesn't show the operating frequency of each individual CPU core. That's a bit frustrating.

HWMonitor

HWMonitor fills the gap left by the shortcomings mentioned above! It is similar to OpenHardwareMonitor, but the good news is that it provides more information!

file

As shown in the image above, my CPU has been overclocked to 6.3 GHz on the single large core, 5.8 GHz on all cores, and 4.6 GHz on the small cores.

CPU Overclocking: Concepts and Knowledge

file

CPU overclocking is, of course, the most critical aspect, as it directly impacts daily usage. However, compared to adjusting hardware settings, it is actually more important to understand exactly who is using your CPU and how these programs are utilizing it. By reasonably assigning tasks to the big cores and little cores, ensuring that your big cores focus on running your highest-priority tasks, you can achieve an excellent experience.

Additionally, some people might mislead you into believing that only Windows 11 can schedule workloads between big and little cores, while Windows 10 simply uses them blindly. This is a complete misconception. Windows 11 merely provides additional APIs that allow processes to be aware of which core a thread is running on. The actual scheduling is handled by Intel, using hardware-accelerated scheduling based on what the threads are doing. Windows 10 can also correctly schedule workloads between big and little cores and will not foolishly assign gaming tasks to the little cores.

Understanding Multi-Core Heterogeneity and Hyper-Threading

Before getting started, note the following: Since Alder Lake, many of Intel's high-end CPUs have adopted a hybrid multi-core architecture (commonly referred to as big.LITTLE cores). This means that, compared to previous generations, we also need to consider:

  • Which tasks should run on Performance cores (P-Cores)
  • Which tasks should run on Efficient cores (E-Cores)
  • Overclocking Performance cores and Efficient cores separately

Typically, I tend to assign my most focused and critical work, such as video encoding/decoding, gaming, and compilation, entirely to Performance cores, while placing background processes, chat applications, and cloud storage clients on Efficient cores.

Intel implements hardware-accelerated scheduling for its hybrid multi-core architecture. You can simply ignore it and let the system handle scheduling, but it may occasionally migrate single-threaded applications that clearly belong on either Performance or Efficient cores back and forth, thereby degrading performance.


At the same time, it is also necessary to understand Hyper-Threading.

Intel® Hyper-Threading Technology is a hardware innovation that allows multiple threads to run on each core. More threads mean more work can be completed in parallel. How does Hyper-Threading work? When Intel® Hyper-Threading Technology is enabled, the CPU exposes two execution contexts on each physical core. This means that one physical core now acts like two "logical cores," capable of handling different software threads. For example, when Hyper-Threading is enabled, a 10-core Intel® Core™ i9-10900K processor has 20 threads.

The API exposed to the system by Hyper-Threading Technology makes it appear as if one core has become two. For instance, CPU0 and CPU1 that you see in Task Manager are actually two Hyper-Threading threads of the same physical core:

file

If a single-threaded application fully occupies CPU0, the performance of CPU1 will drop significantly.

Generally:

  • For Intel's own overclocking tools, if they display: P-Core 0, P-Core 1, this refers to physical cores.
  • For Windows' own CPU management tools, if they display: CPU0, CPU1, CPU2, CPU3... this may not refer to physical cores, but rather to the exposed Hyper-Threading threads.

Task Manager

Yes, Task Manager is simply the most useful computer tuning tool!

For example, I recently noticed that my computer was behaving inexplicably, often failing to achieve high benchmark scores for unknown reasons. In such cases, I open Task Manager to investigate:

file

Wow, CPU0 is running at full capacity! With one of the large cores gone, no wonder the system feels laggy...

I can simply kill the related processes and services, and that’s it. Alternatively, I can use the Set affinity feature to force it to run on the small cores.

file

As shown in the image, I moved the Lighting Service directly from the large cores to the small cores, thereby freeing up one large core.

file

Intel Extreme Tuning Utility

Download Intel Extreme Tuning Utility

Intel Extreme Tuning Utility is a one-click and advanced overclocking tool released by Intel. However, based on my experience, its one-click overclocking is a bit too conservative.

Its one-click overclocking directly pulled my single-core clock down to 6.1 GHz. It’s way too conservative.

file

Of course, I definitely don’t like letting it handle one-click overclocking for me. So naturally, I prefer to adjust these settings myself:

Note: Modifying these settings may require you to close other overclocking software on your computer, especially ASUS AI Overclocking.

file

Additionally, it is also a decent monitoring tool that can show the usage of each core. Previously, while running a single-threaded CPU stress test, I noticed that Core 0 always had a utilization of 50%. It was only after using this tool that I discovered my CPU’s single core actually has two threads.

file

As shown in the image: I was running a two-thread stress test using CPU-Z. Both Core 0 and Core 2 showed a utilization of 50%, indicating that the two threads were likely assigned to Core 0 and Core 2 respectively.

If I forcibly use Task Manager to pin both threads to Core 0, I will see the performance drop by half. This is because both threads are running on the same core.

file

CPU-Z

Download CPU-Z here: https://www.cpuid.com/softwares/cpu-z.html

There’s no need to say much about CPU-Z. However, one thing I really like about it is:

  • CPU-Z surprisingly has benchmarking capabilities, supporting both multi-core and single-core benchmarks, and you can even specify the number of threads to use
  • CPU-Z also has a stress test feature that can max out the CPU with a single click

These features are notably fast, providing scores almost instantly. Of course, if you want an objective score, you should close all other workloads. However, if you don’t close them, it actually reflects how much CPU resources a new, ordinary process can grab under real-world conditions, which is more representative of actual usage scenarios.

Therefore, compared to some specialized benchmarking software, many times I just want to quickly find out what this CPU is capable of, so I run the benchmark directly. It’s very simple and straightforward.

file

By using the Set-Affinity feature in Task Manager, you can even test the performance of individual small cores, allowing you to achieve overclocking targeted at specific cores.

As shown in the image, my final tuned 13900KS achieved a single-core score of 980 and a multi-core score of 17100. For reference only.

Cinebench R20\R23

A professional benchmarking tool. No further explanation needed.

You can download it directly from the app store.

Although its benchmarking process is slow, it is more accurate. Specifically: we often use whether a configuration can pass the Cinebench R20 benchmark to determine if the setup is usable.

CPU Overclocking Approach

file

The CPU is the most direct component affecting performance. Therefore, CPU overclocking is the most important part of computer overclocking. Above, we only briefly introduced the basic concepts and common knowledge about overclocking. Next, we will detail the approach to overclocking.

There are four main steps in CPU overclocking:

  • Step 1: Survive the stress test
  • Step 2: Maximize cooling
  • Step 3: Efficient output
  • Step 4: Reduce noise

CPU Overclocking Stage 1 - Surviving the Stress Test 🐔

file

First, ensure your CPU can survive the stress test. In the following steps, if you encounter any Blue Screen of Death (BSOD), screen flickering, boot failures, Windows exceptions, or benchmarking issues, you must roll back to this step and ensure your system passes the stress test.

The "stress test" refers to running the CPU under a heavy load for a period of time. This confirms that your settings are stable enough to support daily use.

Generally, a newly purchased computer should be able to pass the stress test without issues.

file

To pass the stress test, you may consider downclocking! You can directly adjust the core operating frequencies using Intel Extreme Tuning Utility or within the BIOS settings. For example, my i9 13900KS can be adjusted to 5.6 GHz for the P-cores and 4.4 GHz for the E-cores. These lower frequencies will definitely allow the system to pass the stress test.

CPU Overclocking Phase 2 - Pushing Cooling to the Limit

The purpose of this step is to ensure that the CPU is not idling, is indeed performing at its highest capacity, and has reached the thermal limit of the cooler. This way, when we proceed to overclock, the cooler will consistently be the bottleneck. To further improve performance, the only option is to upgrade the cooling solution.

Before fully stressing the cooling system, first ensure the following:

  1. Your water pump is set to maximum flow rate (the pump has a greater impact than fans)
  2. All your fans are set to maximum speed (configured in Armoury Crate)

Armoury Crate has a feature called Fan Expert; simply set it to Full!

file

Simultaneously open HWMonitor and Cinebench R23, and start the stress test. Pay close attention to the CPU temperature in HWMonitor.

Typically, this value should be 100. At this point, the CPU is under heavy load and will quickly hit the thermal wall. If the temperature does not reach 95, it indicates that the CPU is not performing at full capacity.

file

If the CPU is not performing at full capacity, you should investigate why it is not pushing its limits. Common reasons include: frequency set too low, power adapter malfunction, etc. Appropriately increase the frequency.

If the CPU consistently reaches 100 degrees during benchmark runs, with fans and water pump at maximum settings, this proves that the CPU is not idling and that you have fully utilized the cooling limit of your cooler. This is a straightforward principle: if my cold plate, fans, water pump, etc., are all working at full capacity, and the CPU temperature remains steady at 100 degrees during benchmarks, then this represents the limit of this CPU's working capability.

CPU Overclocking Phase 3 - Precision Output

At this stage, we have reached the limit of our thermal capacity. We can still make some adjustments to enable the CPU to deliver higher performance.

First, we can increase the anti-voltage drop settings. This setting is called: CPU Calibration Load Line. When configuring it, a recommended system value will be displayed in the bottom right corner of the screen. Simply use the recommended value.

<img src="/download/markdown-images/img-367d0e4d-4fb5-44c6-89d6-c833a488202c.png" style="width: 30%">

Also, don't forget to Override the CPU voltage. It will also provide suggested values. For the 13900KS, I can directly enter 1.440 or 1.500.

After that, precisely adjust the frequencies for both the large cores and the small cores. Ensure that the system passes stress testing after the adjustments.

file

This step is very tedious; it involves constantly tweaking parameters, seeing which one is set too high and causing instability, then lowering it. If you lower it too much, increase it again... It is expected to take an afternoon.

Basically, CPU overclocking involves tuning these three parameters: frequency, voltage, and Load-Line Calibration (LLC).

Additionally: Do not forget the small cores! They are equally important for multi-core benchmark scores!

CPU Overclocking Stage 4 - Reduce Noise

Regular users can just click here and be done with it.

file

Of course, you can also manually adjust the curves below.

Around 30 degrees, the system is under no load. At 70 degrees, it starts handling heavy workloads. You can set the minimum speed at 30 degrees and full load at 70 degrees. See the figure:

file

The two bars on the right refer to setting acceleration. If you want to prevent the CPU from ramping up too quickly under sudden loads, you can lower the acceleration rate. If you want to prevent the fans from slowing down abruptly at high speeds, you can also lower the deceleration rate.

Finally, after reducing the noise, don't forget to run a benchmark!

The 13900KS can score over 40,000 points in R23!

file

As shown in the figure above, the 13900KS can score 39,836 points with VBS and all security features enabled. In fact, the score can be even higher if VBS is disabled.

My Little Experience

On my first day after successfully overclocking, everything seemed to be running perfectly fine. However, my monitor would randomly and inexplicably flicker. I thought it was a system issue, so I reinstalled the OS. I thought it was a driver issue, so I reinstalled the drivers. I thought it was a cable issue, so I swapped the cables. I thought there might be a problem with the monitor or GPU ports, so I switched ports... But no matter what I changed, it would still flicker from time to time... It was truly very frustrating.

Finally, it suddenly dawned on me: If the CPU is overclocked too aggressively, it can cause calculation errors in the GPU drivers, leading to monitor flickering! As a result, I slightly lowered the overclock, and the issue was resolved.

Simplified Overclocking?

If you really don't feel like tuning it yourself, you can also use the ASUS motherboard's beginner-friendly AI one-click overclocking feature.

file

However, please note that using AI one-click overclocking may not guarantee stable operation at this frequency. You must:

  • Train the fan performance before AI overclocking. This can be adjusted in BIOS settings under Main tweaks -> AI Feature.
  • Manually adjust the memory frequency!
  • Enable anti-voltage droop to the recommended level!
  • If you frequently experience blue screens after AI one-click overclocking, you must adjust the confidence level of the AI overclocking. This can be adjusted in BIOS settings under Main tweaks -> AI Feature. You can lower it to 97 or 96 and observe for a period of time.

GPU/Graphics Card Adjustment

How well you tune your graphics card directly determines how enjoyable your visual experience is. This includes whether you can fully leverage your monitor's color accuracy, brightness, resolution, and refresh rate; whether hardware acceleration is enabled during video playback; and whether you can optimize bottlenecks that cause stuttering in games.

Of course, if you don't have a dedicated graphics card or don't play games, you can skip this section.

System Settings

You read that right. The most important settings are all in the system settings:

Set the resolution to the monitor's recommended resolution:

file

Hardware acceleration scheduling:

file

Set the refresh rate to the maximum:

file

Enable HDR:

file

Enable other features as well. For example, AutoHDR:

file

Note: When HDR is enabled, Windows will instruct the GPU to output an HDR signal. This can improve the dynamic range of the display. However... many applications do not support or understand HDR, and are not optimized for it. (For example, the Windows desktop does not support it.) Therefore, Windows will map this SDR content to an HDR signal, which can make the content appear somewhat artificial or unsaturated. Furthermore, enabling HDR makes screen sharing, taking screenshots, and live streaming more difficult. Here is my recommendation:

  • Only enable HDR right before you open HDR game content or play HDR movies/shows. Disable HDR before sharing your screen, taking screenshots, or live streaming.

GPU-Z

GPU-Z is a great tool for checking graphics card status.

It provides a comprehensive view of various GPU parameters and operational conditions, making it easy to compare later.

file

Download GPU-Z here: https://www.techpowerup.com/gpuz/

NVIDIA Control Panel

(NVIDIA GPUs only) NVIDIA Control Panel. It may seem unnecessary, but it is incredibly useful.

For example, if your monitor claims to support native 10-bit color but you cannot find the 10-bit option in Windows display settings, NVIDIA Control Panel is the place to go!

file

Here, you can make more advanced changes to display settings, adjust G-Sync, modify rendering settings for specific applications to improve the visual quality of older games, configure multi-monitor surround, set up PhysX, and more.

I remember previously enjoying the classic game Need for Speed. Since that game is quite old and even its highest graphics settings do not put much strain on the GPU, you can use NVIDIA Control Panel to adjust its anti-aliasing settings.

NVIDIA GeForce Experience

file

Download NVIDIA GeForce Experience here: https://www.nvidia.com/en-us/geforce/geforce-experience/

(For NVIDIA graphics cards only) Many people choose not to install NVIDIA GeForce Experience, but it is actually very useful. It can:

  • Optimize your game graphics settings (This feature is a bit flawed; it even dared to lower my resolution. The one thing I cannot tolerate is a reduced resolution)
  • Monitor your GPU status in real-time while gaming
  • Simple overclocking
  • Screenshot, recording, shadow play, and streaming (even supports HDR screenshots and screenshots at resolutions higher than your screen)

I was once very puzzled as to why my GPU power consumption was so low (it stayed at 200+ W in many games), yet I heard from some review agencies that it often hits 400+ W. So I used GeForce Experience to run a 3DMark Time Spy benchmark and successfully captured a moment where it hit 500 W:

file

For example, while playing Cyberpunk, I pressed Alt+F1 to get a high-definition HDR screenshot:

file

MSI afterburner

Download MSI Afterburner here: https://www.msi.com/Landing/afterburner/graphics-cards

A very popular GPU overclocking tool. You can use it freely even if you don't have an MSI graphics card. For many GPUs, you can even directly adjust the fan speed curve.

file

GPU overclocking feels much easier than CPU overclocking... Just pull the core frequency and memory frequency sliders up, and you're done.

The only thing that confuses me is: overclocking this 4090, it's easy to hit 3000, but performance doesn't really improve much. I don't know why; I guess there must be some other bottleneck.

Anyway, I ended up lowering it back down. I figure if this card had such easy performance gains, NVIDIA would have already pushed it to its limits before shipping it to me.

TimeSpy Extream

TimeSpy is a scenario in 3DMark. You need to download 3DMark first: https://www.3dmark.com/

file

TimeSpy Extreme is a benchmarking tool. It simulates a complex, real-world DirectX 12 game scene to measure how well a graphics card performs.

file

TimeSpy also supports conveniently sharing your results. For example, my benchmark score is: https://www.3dmark.com/3dm/88914779

SSD/Hard Drive Adjustment

SAMSUNG Magician

(Samsung hard drives only)

Download SAMSUNG Magician here: https://semiconductor.samsung.com/consumer-storage/support/tools/

I actually find this quite strange. Hard drives should ideally be tuned to optimal performance right out of the box. But Samsung insists that you need to install this additional Magician software to unlock peak performance.

After installation, it didn't recognize my drive on the first try. It took several retries before it worked. Simply enable performance mode directly.

file

It requested a restart and also allocated some space from my C drive. However, since it recommended this action to improve performance, I won't complain.

CrystalDiskInfo

Of course, to get the most comprehensive hard drive information, you still need CrystalDiskInfo.

Download CrystalDiskInfo here: https://crystalmark.info/en/software/crystaldiskinfo/.

It is simply a disk status viewer. Importantly, it can tell you how much data has been written to the disk, which is very helpful for estimating its lifespan.

file

CrystalDiskMark

Download CrystalDiskMark here: https://crystalmark.info/en/software/crystaldiskmark/

Now that the hard drive has been configured, it’s natural to run some tests. It is recommended to use CrystalDiskMark to simulate real-world usage scenarios.

If you have an NVME SSD, it is recommended to select NVME in the settings in advance, and choose Peak Performance +Mix in the Profile section:

file

For example, my disk achieved 7000 MB/s sequential read, 6700 MB/s sequential write, 4900 MB/s random read, and 5500 MB/s random write. You can compare these figures with the drive's design specifications, or check them again later to see how much performance has degraded over time.

HDTune Pro

HDTune Pro can test hard drive speeds with a more professional approach. Compared to CrystalDiskMark, its tests focus more on theoretical accuracy, but they may not fully reflect real-world usage conditions.

file

Note that HDTune Pro is not free software.

Game Settings

After completing basic hardware overclocking, what has the biggest impact on the gaming experience?

It is the game settings!

Incorrect game settings may cause you to hit the power limit, run out of VRAM, encounter a CPU bottleneck, encounter a GPU bottleneck, or encounter a disk bottleneck... Properly adjusting game settings is also key to achieving a good gaming experience.

Solutions for Out-of-Memory Errors

VRAM overflow occurs when the GPU's memory is completely filled, causing severe game stuttering. In fact, under VRAM overflow conditions, the GPU is often largely idle.

For example, consider the following scenario:

<img style="max-width:300px" src="/download/markdown-images/img-40680484-d0a1-42b0-971a-90d00368cea3.png">

Pay attention to the data above:

  • CPU usage is extremely low.
  • GPU power consumption is far below the graphics card's TDP. (A normal 4090 should draw around 400W; anything under 300W indicates it's idling, while peak power may exceed 500W.)
  • The frame rate is in the single digits, making it unplayable.

At this point, you can check in Task Manager to see if there is a video memory overflow issue:

<img style="max-width:700px" src="/download/markdown-images/img-f2e6660d-c688-4b5f-9790-4b29a6140b24.png">

Clearly, the VRAM is already full.

In this situation, adjusting settings such as anti-aliasing, DLSS, resolution, ambient lighting, or ray tracing won’t make much of a difference. Once VRAM is full, tweaking environment and rendering settings won’t solve the issue of the GPU waiting for available VRAM. The priority is to free up VRAM.

The process is straightforward. Gradually lower the following settings:

  • Render distance
  • Number of world objects
  • Texture quality

These are all effective ways to reduce VRAM usage. After making changes, aim to keep VRAM usage around 70–80%. Setting it too low may result in blurry textures and a short draw distance. Try to increase these settings as much as possible without exceeding VRAM limits.

file

Solution for Poor Screen Quality

"Terrible graphics" refers to the situation where, even after you've cranked the game's default settings to the maximum, your hardware still feels like it's barely working. A classic example is Minecraft.

file

Even if you max out the graphics in Minecraft, it remains pixelated and uses almost no VRAM...

To fix this, you need to find a third-party shader pack in the community.

For example, my solution for Minecraft's poor visuals was installing the Sildur's+Vibrant+Shaders+v1.50+Extreme-VL.zip shader pack. You can download it here: download.

The visual result after installation looks realistic! Reference video: https://videos.aiursoft.cn/view?m=ebIGGp5Cs

Solutions for CPU Bottlenecks

A CPU bottleneck refers to a situation where the CPU is nearly at its limit, while the GPU is underutilized. In this scenario, the game's logic may be struggling, causing rendering to lag as well. However, the visuals are not particularly complex, and both VRAM and GPU utilization remain low, while CPU usage is high.

Note: A key characteristic of a CPU bottleneck is NOT that the CPU utilization is at 100%! Many games have poor multi-core optimization, meaning that if the specific core handling critical threads is saturated, the game will start to stutter! Therefore, relying solely on the overall CPU usage rate in the performance panel is not an accurate way to identify a CPU bottleneck! You need to analyze both the overall multi-core usage and the single-core usage, combined with GPU and VRAM utilization, to determine if you are experiencing a CPU bottleneck.

Another way to identify a CPU bottleneck is that lowering the resolution does not improve game performance.

For example, games that typically encounter CPU bottlenecks include:

  • Transport Fever 2
  • Cities Skylines
  • SimCity
  • Microsoft Flight Simulator (single-core bottleneck; do not focus on overall usage)

For games with CPU bottlenecks, the only current solution is to enable DLSS.

This is because DLSS can leverage the idle GPU compute power to generate additional frames, creating a smoother experience. When there is a CPU bottleneck, the GPU is often underutilized, so having it perform frame generation can often double the frame rate!

Solutions for GPU Bottlenecks

A GPU bottleneck refers to the situation where the GPU's ray tracing units, optical flow units, and other components are pushed to their limits. In this scenario, the game application itself does not lag, and the operating system remains responsive enough to switch out of the game. However, the visual rendering feels choppy. GPU 3D utilization is very high, but VRAM usage remains low. Neither single-core nor multi-core CPU usage reaches full capacity.

A GPU bottleneck is likely the most common issue encountered in gaming. The natural solutions are:

  • Enable DLSS/FSR
  • Lower anti-aliasing settings
  • Reduce render distance
  • Lower shadow quality
  • Lower environmental quality
  • Lower resolution (not recommended)

This is based on the consideration that:

Deciding to Use DLSS or FSR

Deep Learning Super Sampling (DLSS) is an image scaling technology developed by Nvidia that can be used in real-time in video games. This technology utilizes deep learning to upscale images from a lower resolution to a higher resolution, enabling them to be displayed on high-resolution computer monitors.

DLSS itself allows the graphics card to first render at a lower resolution and then use algorithms to reconstruct a high-resolution image. In many cases, it can achieve a realistic effect that is indistinguishable from native rendering. Furthermore, this technology significantly reduces the computational load on the graphics card.

Enabling DLSS can significantly improve frame rates when the GPU is the bottleneck, with almost no cost.

Reduce Anti-Aliasing Decision

Cited from https://zh.wikipedia.org/zh-cn/%E5%8F%8D%E9%8B%B8%E9%BD%92

Full Scene Anti-Aliasing (FSAA) refers to the use of anti-aliasing technology to upscale, sample, and reconstruct the full-screen image signal output to the display, rather than applying anti-aliasing techniques to only a portion of the screen. FSAA has a significant impact on 3D game visuals, making the entire 3D game scene appear smoother, sharper, and more realistic. It has become a standard feature in virtually all modern commercial games and represents an important technological application. FSAA can be implemented using the following methods:

Supersampling Antialiasing

Super-Sampling Anti-aliasing (SSAA) is an early anti-aliasing method. It is resource-intensive but simple and direct: the image is first mapped to a buffer and upscaled, then super-sampling is applied to the pixels of the upscaled image. Typically, two or four neighboring pixels are selected, and these samples are blended to generate the final pixel. This allows each pixel to possess characteristics of its neighboring pixels, resulting in similar transition colors between pixels and smoother color transitions along graphic edges. The final pixels are then downscaled back to the original image size and saved to the frame buffer (i.e., video memory), replacing the original image storage. Finally, the result is output to the display, showing a single frame. This process is equivalent to taking a blurry large image, refining it, and then shrinking it into a clear small image. If anti-aliasing is applied to every frame, all visuals in games or videos will have the anti-aliasing effect. [2] When mapping the image to a buffer and upscaling it, the scaling factor determines the level of anti-aliasing effect. For example, in Figure 1, the x2, x4, and x8 following "AA" represent the upsampling factors relative to the original image. The sampling methods generally used in Super-Sampling Anti-aliasing are two types:

  • Ordered Grid Super-Sampling (OGSS), which selects two neighboring pixels during sampling.
  • Rotated Grid Super-Sampling (RGSS), which selects four neighboring pixels during sampling.
Multisample Anti-Aliasing

Multisampling Anti-Aliasing (MSAA) is a specific type of Supersampling Anti-Aliasing (SSAA). MSAA originated in OpenGL. Specifically, MSAA applies supersampling anti-aliasing only to data in the Z-Buffer and Stencil Buffer. This can be simply understood as applying anti-aliasing only to the edges of polygons. As a result, compared to SSAA, which processes all data in the image, MSAA significantly reduces resource consumption, though it may offer slightly lower image quality. [3]

This is the most common form of anti-aliasing, and nearly all games support MSAA.

Lowering the anti-aliasing setting can significantly improve frame rates when the GPU is the bottleneck, with generally minimal visual cost.

Making Decisions in Vulkan and OpenGL

Many games support freely choosing different rendering engines. The recommended priority here is:

  • DirectX
  • Vulkan
  • OpenGL

Both Vulkan and OpenGL have excellent cross-platform compatibility. However, they have weaker features compared to DirectX.

OpenGL is generally easier for developers to implement.

OpenGL typically has a rich ecosystem of community extensions.

Monitor Settings

Yes, many people don't realize that monitors also require specific calibration!

Use a Professional Colorimeter (Optional)

In certain professional graphics workflows, you may need to rent a professional colorimeter to ensure accurate on-screen colors. However, most gamers likely won't care about color accuracy.

If you don't care about precise color reproduction, you can open your monitor's settings menu and choose one of the preset profiles that you prefer. These include:

file

  • Default
  • Vivid
  • Movie
  • Game
  • Reading
  • Eye Care

Adjusting SDR Content Brightness in HDR Mode

You can learn more about HDR in this blog post. Learn about HDR

After enabling HDR, don't forget to adjust the SDR content brightness for HDR mode in Windows.

In the Windows 10 operating system, you can find this option by following these steps:

  1. Click the Windows button and select "Settings" (gear icon).
  2. Click "System".
  3. Select the "Display" tab.
  4. Under "Windows HD Color", find the "SDR content" option.
  5. Drag the "SDR content brightness" slider.

file

If it is an LCD or MiniLED display

  • If the brightness is set too low, the SDR content may become hard to see clearly.
  • If the brightness is set too high, it may be blindingly bright, like staring at a white notepad.

If Using an OLED Display

This setting doesn't make much of a difference; just set it to the middle. OLED displays naturally don't get too bright when showing full-screen white content.

file

Enable HDR

You may need to set up a keyboard shortcut to quickly toggle HDR on and off.

file

For LCD Displays

  • I recommend enabling HDR only when viewing photos, videos, or playing games with HDR content. Otherwise, please turn off HDR.

This is because LCD displays have poor contrast capabilities, resulting in minimal differences between dark and bright areas. Enabling HDR offers little to no benefit and may even lead to reduced saturation.

Additionally, I still recommend replacing your LCD monitor with a better one.

If using a MiniLED display

  • I recommend enabling HDR in all situations. This is because, based on SDR to HDR conversion, traditional SDR video and desktop content will also achieve higher contrast on MiniLED displays, delivering a stronger visual impact.

If You Have an OLED Display

  • HDR must be enabled under all circumstances.

This is to prevent OLED from forcing all brightness levels into a very dark safe brightness range when in SDR mode, which would result in a dim viewing experience.

Change More Display Settings

The monitor has many settings, including:

file

  • Enable GSync-Adaptive Sync. This can help mitigate screen tearing if supported.
  • If you have an LCD or MiniLED monitor, enable low latency mode with caution. This may cause inexplicable display issues. You can set it to a lower level. High-level low latency settings may result in an inability to effectively control the response.

Remember to perform HDR calibration!

Of course, if the game supports HDR, remember to enable HDR and perform HDR calibration first! This can help avoid encountering inexplicable brightness and shadow issues in-game.

https://support.microsoft.com/en-us/windows/calibrate-your-hdr-display-using-the-windows-hdr-calibration-app-f30f4809-3369-43e4-9b02-9eabebd23f19

Continue Learning HDR

Using HDR features requires a certain level of foundational knowledge. You need to continuously learn a lot of information. Learn more here: https://anduin.aiursoft.com/post/2023/3/26/hdr