NVIDIA Quadro K620

The NVIDIA Quadro K620 is an entry-level professional graphics card built on the Maxwell architecture, occupying the 35th percentile of all GPUs in the benchmark database. Its average benchmark score of 6286 places it in a tightly contested segment where the nearest rivals are separated by less than one percent, making it a product defined by its position in a highly competitive low-end market.

Power and Cooling

The Quadro K620 carries a 45 W TDP, a figure that reflects its efficiency-focused Maxwell design on a 28 nm process from TSMC. This low power envelope has direct practical consequences: the card requires no auxiliary power connectors, drawing all its power from the PCIe 2.0 x16 slot. The system power supply recommendation is a modest 200 W, which means this card can be installed into pre-built office workstations or small-form-factor systems without upgrading the PSU.

The cooling solution is a single-slot design, measuring 160 mm in length (6.3 inches) and 69 mm in height (2.7 inches). This compact physical profile, combined with the lack of power connectors, makes the K620 a drop-in replacement for older or integrated graphics in space-constrained chassis. The thermal design is straightforward: a 45 W GPU does not generate significant heat, so the single-slot cooler is sufficient for sustained professional workloads. The absence of a 6-pin or 8-pin PEG connector also simplifies cable management, and the card's end-of-life production status means it is typically found in refurbished or legacy systems rather than new builds.

Who Should Consider It

Benchmark data positions the Quadro K620 as a solution for basic professional acceleration rather than high-end rendering. Its Geekbench OpenCL score of 6693 and Vulkan score of 5879 indicate that it can handle compute tasks, but the 2 GB DDR3 memory and 28.80 GB/s bandwidth are limiting factors. For users working at 1080p resolution with light to moderate GPU acceleration — such as 2D CAD, basic photo editing, or spreadsheet-heavy visualization — the card provides adequate performance. The 35th percentile ranking versus all GPUs means it outperforms the majority of integrated graphics and entry-level discrete cards, but it will struggle at 1440p or 4K resolutions with complex textures or large datasets.

The card is not suited for modern gaming at high settings; the FP32 performance of 863.2 GFLOPS and 16 ROPs limit fill-rate-heavy workloads. Instead, it targets users who need certified drivers and stable output for professional applications, where the 1x DVI and 1x DisplayPort 1.2 outputs support dual-monitor setups. The 2 GB VRAM is sufficient for 1080p professional workloads but will be exhausted quickly with multi-layer compositing or high-resolution medical/satellite imagery. Users with legacy PCIe 2.0 motherboards will find the bus interface compatible, though the x16 lane width ensures no bandwidth bottleneck for this class of GPU.

How It Compares

The Quadro K620's nearest rival is the AMD Radeon R7 M350, which scores 6290 on average, a delta of -0.1% compared to the K620's 6286. This is a statistical tie; benchmark results indicate the two cards are functionally interchangeable in most workloads. The K620's Maxwell architecture offers better driver stability in professional applications, but the R7 M350 may have slight advantages in memory bandwidth depending on system configuration.

Against the AMD Radeon Pro WX 4100, which scores 6309, the K620 trails by -0.4%. The WX 4100 is a newer professional card, and while the scores are nearly identical, the WX 4100 likely benefits from more modern feature support. The data shows the K620 holding its own in raw compute, but the WX 4100's superior software ecosystem for professional ISV applications would make it the better choice for current workloads.

The NVIDIA GeForce GTX 460 SE scores 6326, placing the K620 at -0.6% behind. This is notable because the GTX 460 SE is a much older consumer GPU with a higher TDP. The K620 achieves comparable scores with far less power draw, highlighting the efficiency gains of Maxwell. However, the GTX 460 SE may have better raw fill-rate in games due to its wider memory bus.

The closest competitor below the K620 is the NVIDIA GeForce GT 1010, which scores 6241. The K620 leads by 0.7%, a small but consistent margin. The GT 1010 is a newer budget card, yet the K620's professional drivers and Quadro-specific optimizations give it an edge in compute tasks, even if the GT 1010 offers better modern API support.

FAQ

Q: What is the release date of the NVIDIA Quadro K620?

A: The production status is end-of-life, with a release date of 2014-07-21.

Q: Does the Quadro K620 require a power connector?

A: No, it has no power connectors and relies entirely on the PCIe 2.0 x16 slot for power, with a recommended 200 W system PSU.

Q: What is the memory configuration of the K620?

A: It has 2 GB of DDR3 memory on a 128-bit bus, providing 28.80 GB/s of bandwidth.

Q: What display outputs are available?

A: The card offers 1x DVI and 1x DisplayPort 1.2 outputs.

Q: How does the K620 perform in Vulkan benchmarks?

A: It scores 5879 in Geekbench Vulkan, compared to 6693 in Geekbench OpenCL, indicating stronger OpenCL performance.

Q: What is the transistor count and die size?

A: The GM107 chip contains 1,870 million transistors on a 148 mm² die, fabricated on a 28 nm process.

Benchmark Performance

The average benchmark score of 6286 places the Quadro K620 in a razor-thin competitive band. The data shows a spread of only 85 points between the lowest rival (GT 1010 at 6241) and the highest (GTX 460 SE at 6326), representing a 1.4% total range. Within this context, the K620's -0.1% delta versus the R7 M350 is noise-level; the two cards are effectively identical in performance. The -0.4% gap to the Radeon Pro WX 4100 is similarly negligible, though the WX 4100's higher absolute score of 6309 suggests slightly better sustained performance in professional compute loops.

The most meaningful comparison is the +0.7% lead over the GeForce GT 1010. While 45 points is small, the K620 achieves this with a much older architecture and half the memory bandwidth of newer cards. The Geekbench OpenCL score of 6693 versus Vulkan's 5879 reveals a 13.8% advantage for OpenCL, indicating that compute workloads leveraging OpenCL will see better results than those using Vulkan. This is typical for a professional card of this era, where OpenCL was the dominant compute API in CAD and DCC applications. The FP32 throughput of 863.2 GFLOPS, combined with a texture rate of 26.98 GTexel/s and pixel rate of 17.98 GPixel/s, provides a coherent picture: the K620 is a balanced, low-throughput GPU that excels in latency-sensitive professional tasks rather than throughput-heavy rendering.

Memory Subsystem

The K620's memory subsystem is its most significant bottleneck. The 2 GB DDR3 capacity is workable for 1080p professional tasks but insufficient for 4K texture sets or large simulation grids. The 128-bit bus width, when paired with DDR3 at 1800 Mbps effective, yields 28.80 GB/s of bandwidth — a figure that is roughly one-third of what mid-range consumer cards from the same era offered. In high-resolution scenarios, this bandwidth constraint will manifest as texture pop-in or reduced frame rates in viewport-heavy applications.

However, the memory configuration is not without merit. The 2 GB capacity is sufficient for dual-monitor setups at 1080p, and the 128-bit bus is power-efficient, contributing to the 45 W TDP. The memory clock of 900 MHz (1800 Mbps effective) is conservative, which aids stability in prolonged compute tasks. For professional workloads that are compute-bound rather than memory-bound — such as finite element analysis or signal processing — the bandwidth limitation is less impactful. The data suggests that users should avoid 4K resolutions or multi-layer compositing, as the 28.80 GB/s will throttle performance well before the GPU's compute units are saturated.

Ray Tracing and Feature Set

The Quadro K620 has no dedicated ray tracing cores and no tensor cores, as it predates the RTX architecture. Its Maxwell architecture provides 384 shading units, 24 texture mapping units, and 16 ROPs. The API support includes DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.4, which means the card can run modern APIs but at baseline feature levels. The DirectX 12 (11_0) designation indicates it supports the DirectX 12 API but only with feature level 11_0, lacking some of the more advanced features like bindless resources or conservative rasterization.

For ray tracing, the card offers no hardware acceleration, meaning any ray-traced workloads would rely on compute shaders or CPU fallbacks, which would be impractically slow given the 863.2 GFLOPS FP32 throughput. The Vulkan 1.4 support is forward-looking, enabling the card to run modern Vulkan applications, but again without hardware RT extensions. The tensor core absence also means no AI-accelerated features like DLSS or denoising. In professional contexts, this limits the K620 to traditional rasterization and compute tasks. The OpenGL 4.6 support is robust for CAD applications like AutoCAD or SolidWorks, which rely heavily on this API. The 16 ROPs and 24 TMUs provide adequate texture filtering and pixel throughput for 1080p viewport rendering, but the card is not designed for photorealistic rendering or real-time ray tracing.