Recently, a special display card has emerged in the Chinese market during the Tesla T10 launch event — a GPU originally designed by NVIDIA exclusively for cloud gaming services, primarily used in the GeForce NOW cloud gaming platform. These retired display cards have now entered the secondary market, with current prices on Chinese platforms around 1,350 RMB (approximately 190 USD). Due to their affordability, I purchased two to examine their performance.
Table of Contents
Hardware specifications and efficiency
- 16GB GDDR6 Memory
- 150W TDP Design
- Full High-End Design
- Original Factory Overclocked Design
- PCIe 3.0 x16 Interface
- TU102 GPU
- Base Clock: 1065 MHz
- Max Boost Clock: 1590 MHz
- Memory Clock: 1575 MHz
- Memory bandwidth: 256-bit
- Memory: 16GB GDDR6
Efficiency testing
Test Environment
Test environment is VM.
Linux:
- Ubuntu 24.04 kernel 6.8.0-51-generic
- Nvidia Driver: 550.127.08
- CUDA Version: 12.4
- CPU: Eypc 7413 16-core vCPU
- RAM: 16GiB DDR4 3200 MHz ECC REG
Windows:
- Windows 11 24H2 OS Build 26100.2894
- Nvidia Driver: 560.81 (AWS Cloud Gaming Driver)
- CPU: Eypc 7413 16-core vCPU
- RAM: 16GiB DDR4 3200 MHz ECC REG
Game Performance
3DMark Time Spy GPU score:10092
3DMark Steel Nomad Score:2338
Performance close to RTX 2070 Super and RTX 4060.
AI Computing Performance
Usage Llama 3 8B Model and benchmark using llama-bench:
Q4_K Quantized Version (4.58 GiB)
| Test Environment | Generation Speed (tokens/sec) |
|---|---|
| 512 tokens | 62.10 |
| 1024 tokens | 60.41 |
| 4096 tokens | 52.43 |
| 8192 tokens | 41.46 |
F16 Full Precision Version (14.96 GiB)
| Test Environment | Generation Speed (tokens/sec) |
|---|---|
| 512 tokens | 24.10 |
| 1024 tokens | 23.85 |
| 4096 tokens | 22.53 |
8192 token due to insufficient cooling capacity cannot be tested.
Power, cooling, and temperature performance
Peak performance is 150W, while P8 mode (performance mode) power consumption is around 18W.
Due to full load, the system requires sufficient airflow to maintain temperature.
The graphics card is installed in a Dell PowerEdge R7515 server, and during load testing, a high-speed 89% PWM fan allows the graphics card to stay at 82–83°C.
Usage notes
Currently, two T10 cards are installed on the Dell PowerEdge R7515 server—one for Windows environment as a remote gaming machine, and the other for Linux environment as a GPU compute node for Kubernetes.
In real-world applications, whether playing mid-tier games or running large AI models like phi-4, performance is quite impressive. However, the main issue lies in thermal control: if the third-party PCIe card’s LFM (Linear Feet per Minute) cooling mode is disabled, the GPU temperature easily reaches 86°C, causing throttling; at this point, system fans remain at low speeds. But if the LFM mode is enabled, fans continuously run at 89% PWM speed, increasing system power draw by approximately 100W—unacceptable in power-constrained colocation environments. The current solution is to manually adjust fan speeds only when needed.
At a price of 50 (about 190 USD), this graphics card offers excellent value for money, making it a worthwhile consideration for enthusiasts.
Advantages
- Nvidia vGPU support
- 16GB large memory capacity with ECC support
- Dedicated design saves space
- AI computing performance advantage
Drawbacks
- Requires additional cooling solution
- High cooling requirement
- 16GB VRAM limit
- No output port
Conclusion
The Tesla T10 has perfectly showcased the second wave of server-grade hardware in the consumer market. For users requiring high compute power, strong cooling capacity, or solutions for VDI environments, this is a compelling choice. Particularly recommended for users with solid hardware knowledge who plan to optimize performance over time.
Purchase recommendations
Suitable for:
- Limited AI hobbyist
- Professional GPU for large display rendering
- User requiring VDI GPU acceleration for decoding
Not suitable for:
- General consumer
- User seeking instant plug-and-play experience