​What Computer Hardware is Needed to Play with AI?​​ (Continuously Updated)

1. Graphics Card (GPU)​

1.1 Basic Parameters for Evaluating Graphics Card Performance

​What computer hardware is needed to play with AI? Graphics cards are the most critical hardware in the AI field, and the most important parameter to focus on within graphics cards is the video memory (VRAM) specification.

1. VRAM Capacity

   1. Determines the scale of data the GPU can process at one time, directly impacting the execution capability of tasks like model training and high-resolution rendering.
   2. It’s like a water pool; if the pool is small, it can’t hold enough water, and the water will overflow, leading to a VRAM overflow (out of memory).
   3. Many AI applications have minimum VRAM requirements, similar to a computer’s RAM; if it’s too small, the application may fail to open or run very slowly.

2. VRAM Bandwidth

   1. Bandwidth = Memory Clock Speed × Memory Bus Width / 8.
   2. High bandwidth improves data throughput efficiency, which is especially crucial for AI training and scientific computing.
   3. It’s like a pipe for transporting water. If the pipe is small, water transport is slow, and efficiency is low.

3. Memory Interface Width

   1. Similar to the number of lanes on a highway; the more lanes (wider bus width), the greater the amount of data that can pass through per unit of time.
   2. Common bus widths range from 128-bit to 512-bit.
   3. For example, the RTX 5090 has a 512-bit bus width, the RTX 5080 has 256-bit, and the RTX 5060 has 128-bit.
   4. It’s very clear that Higher numerical values indicate better performance.

 

1.2 Some Suggestions When Choosing a Graphics Card

1. The 4060 Ti 16GB is the recommended minimum choice when purchasing a new graphics card.

   1. Of course, cards with slightly lower specifications can also be used; various 8GB versions of the 4000 and 3000 series are still acceptable, but there’s a difference between just getting by and performing well.
   2. If possible, buy the highest-end models like the 4090 or 5090.

2. Try to choose NVIDIA graphics cards.

   1. The vast majority of AI applications are developed based on NVIDIA, and their documentation is more comprehensive. There are more NVIDIA users worldwide, making it easier to solve problems if they arise; with AMD, you might have to figure things out on your own.
   2. It’s not that AMD cards can’t be used; many AI projects theoretically support AMD, but their efficiency is lower than NVIDIA’s, and they come with more hassles. Not recommended for non-professional users.
   3. ​This is quite straightforward, namely what is commonly referred to as “software ecosystem issues”.​​
3. NVIDIA RTX Series

   1. Regarding 30-series graphics cards:

      1. Their performance is still decent, but there is a significant risk of them being used mining cards.
      2. For example, the 3060 Ti 12GB version offers good value for money and is worth buying if you can confirm the seller is trustworthy and the price is right.

   2. Regarding 40-series graphics cards:

      1. Except for the 4090, other models do not offer a significant enough improvement over the 30-series.

   3. Regarding 50-series graphics cards:

      1. They may not offer a significant enough improvement over the 40-series, but in 2025, they will be the latest and most powerful.
4. What about AMD graphics cards?
   1. As the saying goes, “Tactics are like water, ever-changing and shapeless; situations are constantly evolving and interconverting.”
   2. On one hand, everyone knows Nvidia cards are good, so many people buy them, they are expensive, and sometimes out of stock.
   3. On the other hand, AMD might be aware that its own graphics card product line isn’t as dominant as Nvidia’s. Consequently, sometimes certain item can offer very good value, with slightly lower prices but higher performance parameters.
   4. Therefore, I believe that in 2025, AMD graphics cards will also be a good choice.

 

1.3 GPU Tier Breakdown by Brand​

1.3.1 ​ASUS​

• Budget: DUAL
• Entry-level: TUF Gaming (formerly ATS)
• Mid-range: ProArt (designer-focused)
• High-end/Sub-flagship: TUF Gaming (gamer-focused)
• Flagship: ROG Strix
• Ultra-flagship: ROG Matrix

1.3.2  MSI​

• Budget: Ventus
• Mid-range: Gaming (discontinued)
• High-end/Sub-flagship: Gaming X Trio
• Flagship: Suprim X
• Ultra-flagship: Lightning (discontinued)

1.3.3 GIGABYTE​

(Refer to ASUS/MSI tiers for analogous product lines.)

 

1.3.4 Colorful​

Colorful is a reputable second-tier brand with reliable quality and after-sales support. Its iGame series is the flagship lineup:

• Budget: Netral/Lei Feng
• Entry-level: BattleAx
• Mid-range: Advanced (triple-fan) / Ultra (dual-fan)
• Flagship: iGame Vulcan/Neptune
• Ultra-flagship: Kudan

1.3.5 ​ZOTAC​

A Hong Kong-based manufacturer with strong R&D capabilities (Sapphire, its subsidiary, dominates AMD’s GPU market).

• Budget: Twin Edge
• Mid-range: X-Gaming
• High-end: Trinity OC
• Flagship: AMP Extreme/PGF

1.3.6 ​GALAX​

• Budget: Black/White General
• Mid-range: Metal Master
• Flagship: HOF (Hall of Fame)

---

 

1.4 Personal Take

1. What we commonly refer to as a “graphics card” encompasses the core components: GPU chip, VRAM, circuit board, backplate, casing, cooling system, and fans.
2. ​The GPU chip reigns supreme—exclusively fabricated by industry giants Nvidia and AMD.
3. Brands like ASUS and MSI procure these chips from Nvidia/AMD, then layer on proprietary cooling solutions, aesthetic designs, and engineering tweaks to create consumer-ready products.
4. Therefore, graphics cards with the same GPU chip but from different brands generally has similar performance. The differences in products mainly stem from the business strategy, design, and focus of different manufacturers.
5. ​For instance, both first-tier and second-tier brands manufacture RTX 4090 GPUs, yet their actual performance gap is likely under 10%, or even as little as 5%, sometimes negligible, while price tags can differ by hundreds of dollars.
6. ​​Thus, in certain scenarios, strong second-tier or regular second-tier brands offer significantly better value.
7. A Price-Performance Example​
   

   Metric	ASUS ROG Strix 4090	ZOTAC AMP Extreme 4090
   Boost Clock	2640 MHz	2580 MHz
   VRM Phases	24+4	20+3
   Price	$1,999	$1,599
   Performance gap: <5%	Price gap: 25%

Although ZOTAC is a second-tier brand, and i Know ASUS ROG Strix 4090 is a good item, for these two specific products, it might actually be the better choice, especially for budget-conscious gamers.

 

1.5 GPU Benchmark Hierarchy

1.6 Nvidia Card Specs

 

1.6.2 Compare 40 Series Specs

 

	GeForce RTX 4090	GeForce RTX 4080 SUPER	GeForce RTX 4080	GeForce RTX 4070 Ti SUPER	GeForce RTX 4070 Ti	GeForce RTX 4070 SUPER	GeForce RTX 4070	GeForce RTX 4060 Ti	GeForce RTX 4060
GPU Engine Specs:
NVIDIA CUDA® Cores	16384	10240	9728	8448	7680	7168	5888	4352	3072
Shader Cores	Ada Lovelace 83 TFLOPS	Ada Lovelace 52 TFLOPS	Ada Lovelace 49 TFLOPS	Ada Lovelace 44 TFLOPS	Ada Lovelace 40 TFLOPS	Ada Lovelace 36 TFLOPS	Ada Lovelace 29 TFLOPS	Ada Lovelace 22 TFLOPS	Ada Lovelace 15 TFLOPS
Ray Tracing Cores	3rd Generation 191 TFLOPS	3rd Generation 121 TFLOPS	3rd Generation 113 TFLOPS	3rd Generation 102 TFLOPS	3rd Generation 93 TFLOPS	3rd Generation 82 TFLOPS	3rd Generation 67 TFLOPS	3rd Generation 51 TFLOPS	3rd Generation 35 TFLOPS
Tensor Cores (AI)	4th Generation 1321 AI TOPS	4th Generation 836 AI TOPS	4th Generation 780 AI TOPS	4th Generation 706 AI TOPS	4th Generation 641 AI TOPS	4th Generation 568 AI TOPS	4th Generation 466 AI TOPS	4th Generation 353 AI TOPS	4th Generation 242 AI TOPS
Boost Clock (GHz)	2.52	2.55	2.51	2.61	2.61	2.48	2.48	2.54	2.46
Base Clock (GHz)	2.23	2.29	2.21	2.34	2.31	1.98	1.92	2.31	1.83
Memory Specs:
Standard Memory Config	24 GB GDDR6X	16 GB GDDR6X	16 GB GDDR6X	16 GB GDDR6X	12 GB GDDR6X	12 GB GDDR6X	12 GB GDDR6 / 12 GB GDDR6X	16 GB GDDR6 or 8 GB GDDR6	8 GB GDDR6
Memory Interface Width	384-bit	256-bit	256-bit	256-bit	192-bit	192-bit	192-bit	128-bit	128-bit
Display Support:
Maximum Digital Resolution (1)	4K at 240Hz or 8K at 60Hz with DSC, HDR	4K at 240Hz or 8K at 60Hz with DSC, HDR	4K at 240Hz or 8K at 60Hz with DSC, HDR	4K at 240Hz or 8K at 60Hz with DSC, HDR	4K at 240Hz or 8K at 60Hz with DSC, HDR	4K at 240Hz or 8K at 60Hz with DSC, HDR	4K at 240Hz or 8K at 60Hz with DSC, HDR	4K at 240Hz or 8K at 60Hz with DSC, HDR	4K at 240Hz or 8K at 60Hz with DSC, HDR
Standard Display Connectors	HDMI(2), 3x DisplayPort(3)	HDMI(2), 3x DisplayPort(3)	HDMI(2), 3x DisplayPort(3)	HDMI(2), 3x DisplayPort(3)	HDMI(2), 3x DisplayPort(3)	HDMI(2), 3x DisplayPort(3)	HDMI(2), 3x DisplayPort(3)	HDMI(2), 3x DisplayPort(3)	HDMI(2), 3x DisplayPort(3)
Multi Monitor	up to 4(4)	up to 4(4)	up to 4(4)	up to 4(4)	up to 4(4)	up to 4(4)	up to 4(4)	up to 4(4)	up to 4(4)
HDCP	2.3	2.3	2.3	2.3	2.3	2.3	2.3	2.3	2.3
Technology Support:
NVIDIA Architecture	Ada Lovelace	Ada Lovelace	Ada Lovelace	Ada Lovelace	Ada Lovelace	Ada Lovelace	Ada Lovelace	Ada Lovelace	Ada Lovelace
Ray Tracing	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
NVIDIA DLSS	DLSS 3	DLSS 3	DLSS 3	DLSS 3	DLSS 3	DLSS 3	DLSS 3	DLSS 3	DLSS 3
NVIDIA Reflex	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
NVIDIA Broadcast	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
PCI Express Gen 4	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Resizable BAR	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
NVIDIA App	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
NVIDIA Ansel	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
NVIDIA FreeStyle	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
NVIDIA ShadowPlay	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
NVIDIA Highlights	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
NVIDIA G-SYNC®	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Game Ready Drivers	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
NVIDIA Studio Drivers	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
NVIDIA Omniverse	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
RTX Remix	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Microsoft DirectX® 12 Ultimate	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
NVIDIA GPU Boost™	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
NVIDIA NVLink™ (SLI-Ready)	No	No	No	No	No	No	No	No	No
Vulkan RT API, OpenGL 4.6	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
NVIDIA Encoder (NVENC)	2x 8th Generation	2x 8th Generation	2x 8th Generation	2x 8th Generation	2x 8th Generation	1x 8th Generation	1x 8th Generation	1x 8th Generation	1x 8th Generation
NVIDIA Decoder (NVDEC)	5th Generation	5th Generation	5th Generation	5th Generation	5th Generation	5th Generation	5th Generation	5th Generation	5th Generation
AV1 Encode	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
AV1 Decode	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
CUDA Capability	8.9	8.9	8.9	8.9	8.9	8.9	8.9	8.9	8.9
VR Ready	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Card Dimensions:
Length	304 mm	304 mm	304 mm	Varies by manufacturer	Varies by manufacturer	244 mm	244 mm	244 mm	Varies by manufacturer
Width	137 mm	137 mm	137 mm	Varies by manufacturer	Varies by manufacturer	112 mm	112 mm	112 mm	Varies by manufacturer
Slot	3-Slot	3-Slot	3-Slot	Varies by manufacturer	Varies by manufacturer	2-Slot	2-Slot	2-Slot	2-Slot
SFF-Ready Enthusiast GeForce Card	–	Varies by manufacturer	Varies by manufacturer	Varies by manufacturer	Varies by manufacturer	Founders Edition Yes Varies by manufacturer	Founders Edition Yes Varies by manufacturer	–	–
Thermal and Power Specs:
Maximum GPU Temperature (in C)	90	90	90	90	90	90	90	90	90
Total Graphics Power (W)	450	320	320	285	285	220	200	165 or 160	115
Required System Power (W) (5)	850	750	750	700	700	650	650	550	550
Required Power Connectors	3x PCIe 8-pin cables (adapter in box) OR 1x 450 W or greater PCIe Gen 5 cable	3x PCIe 8-pin cables (adapter in box) OR 1x 450 W or greater PCIe Gen 5 cable	3x PCIe 8-pin cables (adapter in box) OR 1x 450 W or greater PCIe Gen 5 cable	2x PCIe 8-pin cables (adapter in box) OR 300 W or greater PCIe Gen 5 cable	2x PCIe 8-pin cables (adapter in box) OR 300 W or greater PCIe Gen 5 cable	2x PCIe 8-pin cables (adapter in box) OR 300 W or greater PCIe Gen 5 cable. Certain manufacturer models may use 1x PCIe 8-pin cable	2x PCIe 8-pin cables (adapter in box) OR 300 W or greater PCIe Gen 5 cable Certain manufacturer models may use 1x PCIe 8-pin cable	1x PCIe 8-pin cables (adapter in box) OR 300 W or greater PCIe Gen 5 cable. Certain manufacturer models may use 1x PCIe 8-pin cable	1x PCIe 8-pin cables (adapter in box) OR 300 W or greater PCIe Gen 5 cable. Certain manufacturer models may use 1x PCIe 6-pin cable or 1x PCIe 8-pin cable

 

 

 

 

 

 

 

 

 

1.8 Graphics Card Buying Reference

 

2. Processor (CPU)

 

1. In the field of artificial intelligence, CPUs are also important, but not as crucial as GPUs, which represent a do-or-die matter.

2. I3 processor may suffice for basic use but is not recommended. It is advised to use at least an i5 or preferably an i7 or i9.

3. While the CPU is not the most critical component, some AI-related applications can utilize CPU computing power. Naturally, a more powerful CPU provides a better overall experience.

2.1 Intel CPU Performance Degradation Issues

1. Intel 13th and 14th generation processors are known to suffer from performance degradation. This typically manifests as a significant drop in performance after extended use, potentially leading to system instability and even the Blue Screen of Death (BSOD).

2. Performance loss can reach up to 30%, and this decline is irreversible.

3. The issue does not always occur, but it is more likely under prolonged high-load conditions. For light-use scenarios like daily office work, the probability of encountering such issues is significantly lower.

4. Generally speaking, the choice comes down to either Intel’s 12th-gen CPUs or AMD’s.

    

2.2 Can AMD Processors Be Used for Artificial Intelligence?

1. Due to its historically dominant position in both consumer and server markets, Intel has a higher market share. As a result, certain software and frameworks include optimizations specifically tailored for Intel CPUs, potentially offering a slight advantage.

2. However, the performance degradation issues present in Intel’s 13th and 14th generation CPUs are a serious concern that cannot be ignored.

3. Therefore, in 2025, AMD processors have become a very viable and competitive choice for AI-related tasks.

 

 

3. Memory (RAM)

1. Start with at least 32GB of RAM. Dual-channel 32GB (16GB + 16GB) is recommended, and 64GB (32GB + 32GB) is better, 128GB or 256 GB is  much beter— more capacity is always better for performance.
2. Nowadays, many applications load data into RAM to boost efficiency, as RAM read/write speeds significantly exceed those of hard drives.
3. You can even use software like RAMDisk to simulate excess RAM as a virtual hard disk. Such a virtual RAM drive is particularly well-suited for specific scenarios, such as serving as a ​cache area for download software​.
4. Memory modules are easy to purchase​ and well-protected within the motherboard (similar to CPUs), with minimal risk of damage. Focus on distinguishing between ​DDR4/DDR5​ and verifying maximum capacity compatibility
5. There’s no need to buy so-called “geek-grade” memory modules, nor is it worth paying a premium for slightly tighter timings from hand-picked memory chips. A standard 64GB memory kit is far more practical than a 32GB top-tier one.
6. DDR5 offers superior performance. While the actual improvement may not be groundbreaking, it’s still worth considering because in the semiconductor industry, the latest technology is generally the optimal choice. However, ​don’t forget to verify whether your motherboard supports DDR5.

 

3.1 Memory Buying Reference

 

4. ​Solid State Drives (SSD)​

 

4.1 ​Everything You Need to Know About SSDs​

First, let’s clarify some common technical terms. I understand that this involves many specialized terms and acronyms like ​M.2, NVMe, SATA, PCIe, AHCI, which can be overwhelming for general consumers. I will dedicate a separate article to explain these in detail later.

For now, we’ll focus on the core SSD-related considerations:

1. ​Larger capacity is better, as it allows storing more data.
   1. Larger SSDs provide more storage space for data, applications, and multimedia.
   2. High-capacity drives (e.g., 2TB vs. 1TB) also tend to have higher Total Bytes Written (TBW) ratings, which directly correlate with lifespan.
2. ​Faster read/write speeds are better.
   1. Prioritize SSDs with PCIe 4.0/5.0 interfaces and NVMe protocols, which deliver sequential speeds exceeding 7,000 MB/s.
   2. Avoid SATA-based SSDs (max ~550 MB/s) for performance-critical tasks like gaming or video editing.
3. Longer design lifespan is better, which is related to the following metrics:

1. 1. ​Total Bytes Written (TBW)​，Higher TBW values indicate greater durability. For example, a 2TB SSD often has double the TBW of a 1TB model.
   2. ​Over-Provisioning (OP)​​
   3. ​Controller algorithms，while specific algorithms are rarely disclosed, mid-to-high-tier SSDs typically employ advanced wear-leveling and garbage collection mechanisms to optimize lifespan.​
   4. ​NAND flash type​ (e.g., SLC, MLC, TLC, QLC)
      1. Hierarchy: SLC > MLC > TLC > QLC in terms of durability and performance.
      2. SLC/MLC: Primarily used in enterprise-grade SSDs due to high costs.
      3. TLC: Dominates consumer markets with balanced cost and reliability.
      4. QLC: Marketed as cost-effective but lacks long-term user validation.
   5. After-sales Policy
      1. Although it can be very frustrating when SSD damage or data loss occurs even under normal use, it is still a consolation if the manufacturer or seller can provide good after-sales service, such as free replacement.

 

4.2 SSD Buying Reference

 

5. Cooling System

As the performance of computer hardware such as graphics cards, processors, and PCIe 5.0 solid-state drives (SSDs) becomes increasingly powerful, the heat generation from these hardware components is also going up a lot. 

Therefore, the cooling system is becoming increasingly important. At best, it affects the hardware’s performance; at worst, it could damage the hardware.

5.1 CPU Coolingn

Liquid Cooler (Water Cooling Radiator)

• Advantages

  • Theoretically, liquid cooling systems are better, offering higher cooling efficiency.

  • More space-saving; some ITX builds can only use liquid cooling.

• Disadvantages

  • Uncertain lifespan; product quality, the computer’s usual workload, and vibrations from moving or shipping can all affect its usable years. It could last two to three years, or even seven to eight years.

  • In certain special circumstances, there might be a risk of liquid leakage, which could then lead to further damage to other hardware components.

Air Cooler (Air Cooling Radiator)

• Advantages

  • Durable and reliable.

• Disadvantages

  • High-end air coolers, designed for better performance, are often has a large volume, and some computer cases may not be able to accommodate them.

  • Cooling efficiency is not as good as liquid cooling.

My Choice

1. I personally prefer air cooling. Buying a top-tier air cooler means it will likely not break and can be used for many years.

5.2 Solid State Drive (SSD) Cooling

Some high-performance SSDs nowadays are very powerful but generate an exaggerated amount of heat. There are now products specifically designed for SSD cooling. If the case and motherboard space allow for installation, it is recommended to install one.

5.3 Graphics Card Cooling

The cooling system for a graphics card is designed and determined by the manufacturer. There are no separate, additionally installed coolers for graphics cards available on the market. Therefore, regarding graphics card cooling, the first step is to understand the graphics card’s cooling design, and the second is to optimize the cooling design of the computer case, such as its airflow.

 

6. Operating System

For general users, Windows 10 & Windows 11 are preferred.

Unlike graphics cards, the operating system is not a life-or-death issue. Most AI applications also support Linux and macOS. Furthermore, programs can be run within containers using virtualization technologies like virtual machines (VMs) or Docker, but this can be more troublesome for ordinary users.

 

7. Pre-built PC vs. DIY Assembly

It’s generally much better to buy individual components and assemble the PC yourself. Aspects like user experience and computer hardware upgradeability are far superior compared to pre-built systems.

As commercial entities, pre-built PC manufacturers always have profit as their primary consideration. Some manufacturers might sacrifice quality to squeeze out more profit.

 

8. Computer Hardware Priority

• Graphics Card > Memory >= SSD > CPU > Motherboard >= Power Supply
• Although I rank the power supply last, I am also very clear that if the power supply is insufficient or unstable, there is a risk of damaging other hardware.
• My meaning is: ensure you have enough to eat first, then consider eating well. (This is a metaphorical way of saying: meet basic needs first, then pursue higher quality.)