Low-profile Nvidia RTX 3060 graphics card shows limited performance due to bus power constraints

The low-profile Nvidia RTX 3060 graphics card, which draws power exclusively from the PCIe slot, exhibits significantly reduced performance compared to standard, externally powered versions of the same GPU. While the card features 12GB of VRAM, making it theoretically capable of handling local large language model (LLM) tasks, its power consumption is strictly limited by the PCIe bus. This constraint leads to thermal throttling and reduced computational throughput during intensive workloads. In testing, the card achieved approximately 70 frames per second in Cyberpunk 2077 at 1080p on low settings, a result notably lower than a standard RTX 3060. Furthermore, the card scored 4,846 in the 3DMark Time Spy benchmark, which is lower than the performance of a standard GTX 1080. When running LLM tasks, the card's performance is further bottlenecked by its inability to draw additional power, causing it to slow down other components in a multi-GPU setup. The card is not recommended for users seeking optimal performance, as its high cost of entry and limited capabilities make it an inefficient choice for both gaming and AI applications.

The low-profile Nvidia RTX 3060 graphics card, which draws power exclusively from the PCIe slot, exhibits significantly reduced performance compared to standard, externally powered versions of the same GPU. While the card features 12GB of VRAM, making it theoretically capable of handling local large language model (LLM) tasks, its power consumption is strictly limited by the PCIe bus. This constraint leads to thermal throttling and reduced computational throughput during intensive workloads. In testing, the card achieved approximately 70 frames per second in Cyberpunk 2077 at 1080p on low settings, a result notably lower than a standard RTX 3060. Furthermore, the card scored 4,846 in the 3DMark Time Spy benchmark, which is lower than the performance of a standard GTX 1080. When running LLM tasks, the card's performance is further bottlenecked by its inability to draw additional power, causing it to slow down other components in a multi-GPU setup. The card is not recommended for users seeking optimal performance, as its high cost of entry and limited capabilities make it an inefficient choice for both gaming and AI applications.

The low-profile RTX 3060 is limited by its reliance on bus power, which restricts its overall computational performance. In the 3DMark Time Spy benchmark, the card achieved a score of 4,846, underperforming compared to a standard GTX 1080.

The card's 12GB of VRAM allows it to run large language models, but the limited power supply results in slow token generation speeds. During stress testing, the card exhibited significant thermal throttling, reaching temperatures of 79 degrees Celsius.

The card's performance is further constrained in multi-GPU configurations, as it can bottleneck faster cards by slowing down shared compute tasks.

Chapter guide

Worth noting

  • The creator purchased the graphics card with their own funds and states that the opinions expressed are their own.
  • The test system used for evaluation is five years old, which may impact the performance results compared to modern hardware.

Watch the original video ↗