NVIDIA Quadro FX 350

The NVIDIA Quadro FX 350 is a professional workstation graphics card built on the Curie architecture, utilizing the G72 chip fabricated on a 90 nm process at TSMC. It targets legacy CAD and visualization workloads, and its benchmark percentile of 50 places it at the midpoint of all GPUs tracked in the database, indicating it is firmly an entry-level part even within its own generation. The card’s average benchmark score of 0 suggests that no standardized performance tests have been recorded for this model, meaning all analysis must derive from its architectural specifications and memory subsystem rather than direct performance metrics.

Benchmark Performance

Without a recorded benchmark score, the FX 350's performance must be inferred from its fixed-function pipeline and clock rates. The card operates with a memory clock of 405 MHz, translating to 810 Mbps effective, which feeds a 64-bit memory bus. This configuration yields a bandwidth of 6.480 GB/s, a figure that severely constrains fill-rate-heavy tasks. The pixel rate is 1.100 GPixel/s, and the texture rate is 2.200 GTexel/s, based on 4 texture mapping units and 2 raster output units. In practical terms, these numbers indicate the card can handle basic 2D and light 3D workloads at low resolutions, but it will struggle with any modern 3D application that requires substantial texture fetching or pixel shading.

Comparing this to typical entry-level cards of its era, the FX 350’s bandwidth is a decisive bottleneck. The 6.480 GB/s figure is roughly a quarter of what mid-range professional cards offered at the time, meaning any scene with large texture sets or high-resolution framebuffers will see significant performance degradation. The texture rate of 2.200 GTexel/s, while adequate for simple geometry, means that multi-textured surfaces or detailed environments will cause the pipeline to stall. The data suggests that the FX 350 is best suited for wireframe views, basic solid modeling, and 2D schematics, where its 1.100 GPixel/s fill rate is less stressed. In direct contrast, the lack of any shading unit count in the fact pack implies that the card relies entirely on fixed-function T&L (transform and lighting) rather than programmable shaders, which caps its feature support and performance in shader-heavy applications.

Ray Tracing and Feature Set

The FX 350 does not include ray tracing cores or tensor cores, as these are absent from the fact pack. This is consistent with its Curie architecture, which predates hardware-accelerated ray tracing by over a decade. The card’s API support is limited to DirectX 9.0c (9_3) and OpenGL 2.1, with no Vulkan support listed. This means the card cannot run any modern DirectX 12 or Vulkan titles, and its OpenGL 2.1 capability restricts it to legacy professional applications that were written for that specification. For ray tracing specifically, the absence of dedicated hardware means any such workload would be entirely software-based, which is impractically slow on a 21 W card with 2 ROPs.

The feature set is further constrained by the lack of programmable shaders, as indicated by the null shading units field. This points to a card that is strictly fixed-function, meaning it cannot execute pixel or vertex shader programs. Consequently, it is incompatible with most 3D games released after 2004 and many professional visualization tools that rely on shader model 3.0 or higher. The DirectX 9.0c support is nominal; without shader units, the card cannot fully utilize the API’s features. The OpenGL 2.1 support similarly offers no advantage in practice, as the hardware lacks the flexibility to implement modern GLSL programs. For a professional user, this means the FX 350 is only viable for very specific, legacy software that targets fixed-function pipelines, such as early 2000s CAD packages.

Who Should Consider It

Given the performance profile, the FX 350 is only appropriate for users running software that explicitly supports fixed-function rendering. At resolutions of 1024x768 or lower, the card can manage basic 3D wireframes and simple shaded views in CAD applications that predate 2006. The 128 MB memory capacity is sufficient for small models, but any assembly with more than a few thousand polygons will likely exceed the framebuffer, forcing texture thrashing. For 2D desktop work or spreadsheet-heavy tasks, the card is perfectly adequate, as its 1.100 GPixel/s fill rate is unnecessary for 2D blits.

Users considering this card for gaming or modern 3D modeling should look elsewhere, as the lack of shader units and the 6.480 GB/s bandwidth make it functionally obsolete. The data indicates that even entry-level integrated graphics from the same era would outperform it in shader-based workloads. For professional environments, this card is best suited as a secondary display adapter for a multi-monitor setup running legacy software, or for diagnostic purposes on old workstations. It is not a viable choice for any current 3D application, regardless of settings, because the fixed-function pipeline cannot process modern geometry or lighting models. The 21 W TDP and lack of power connectors make it easy to install in any system with a 200 W power supply, but that convenience does not offset the performance limitations.

FAQ

Q: What is the maximum supported DirectX version for the NVIDIA Quadro FX 350?

A: The card supports DirectX 9.0c (9_3), which is the highest DirectX version listed in its specifications.

Q: Does the NVIDIA Quadro FX 350 support hardware ray tracing?

A: No, the card does not include ray tracing cores, and its Curie architecture predates hardware-accelerated ray tracing entirely.

Q: How much memory bandwidth does the Quadro FX 350 have?

A: The card has a memory bandwidth of 6.480 GB/s, derived from a 64-bit bus and 405 MHz memory clock running at 810 Mbps effective.

Q: What is the power consumption of the Quadro FX 350?

A: The card has a TDP of 21 W and requires no additional power connectors, with a suggested power supply rating of 200 W.

Q: Can the Quadro FX 350 run Vulkan applications?

A: No, Vulkan support is not listed for this card; its only supported APIs are DirectX 9.0c and OpenGL 2.1.

Q: What is the process node of the GPU used in the Quadro FX 350?

A: The G72 chip is fabricated on a 90 nm process at TSMC, containing 112 million transistors on an 81 mm² die.

How It Compares

The fact pack lists no nearest rivals and no benchmark scores, so direct comparisons cannot be made using numerical deltas. However, the FX 350’s position in the database’s 50th percentile suggests it sits exactly in the middle of all GPUs, which is misleading given its fixed-function nature. In the context of its own generation, the card is clearly at the bottom of the professional stack, as its 2 ROPs and 4 TMUs are minimal for even the entry-level Quadro FX line. Without rival data, one can infer that any contemporary card with programmable shaders would outperform it in anything but the most basic 2D tasks. The lack of a successor in the fact pack’s nearestRivals field indicates that the database does not track comparative performance for this model, leaving only architectural analysis as a guide.

Memory Subsystem

The memory subsystem is the most critical limiting factor for the FX 350. It is equipped with 128 MB of DDR2 memory, a capacity that was small even in 2006, and is paired with a 64-bit memory bus. This bus width is half of what was standard for even low-end cards at the time, resulting in a bandwidth of 6.480 GB/s. For high resolutions such as 1600x1200 or higher, this bandwidth is grossly insufficient, as the framebuffer alone would consume a significant portion of the available bandwidth, leaving little for texture reads or vertex data. The 128 MB capacity also means that large textures cannot be stored locally, forcing the card to either downsample textures or stream them from system memory over the PCIe 1.0 x16 interface, which is slower than the local bus.

The effective memory speed of 810 Mbps, while typical for DDR2, does not compensate for the narrow bus. In practice, this means the FX 350’s memory subsystem is suitable only for low-resolution (1024x768 or lower) and low-color-depth (16-bit) framebuffers. The 6.480 GB/s bandwidth is roughly sufficient for a 1024x768 32-bit framebuffer at 60 Hz, but any 3D rendering with depth buffering will double the memory traffic, pushing the card beyond its limits. For a professional user, this memory configuration dictates that the card is only usable for 2D CAD drawings or basic 3D models with simple textures. The absence of any error-correcting code (ECC) support, while not listed, is implied by the consumer-grade DDR2 type, which further reduces its suitability for mission-critical visualization tasks where data integrity is paramount.