ATI FireGL 9600 T2-64S

The ATI FireGL 9600 T2-64S is a workstation-oriented graphics card from AMD, built on the RV350 chip using the R300 architecture and a 130 nm process at TSMC. It carries 60 million transistors on a 76 mm² die, resulting in a transistor density of 789.5K per mm². This card occupies a specific historical niche, and its benchmark data positions it at the 50th percentile among all GPUs, with an average benchmark score of zero, indicating it serves as a baseline reference point rather than a performance leader.

Benchmark Performance

The benchmark results for the FireGL 9600 T2-64S are stark: the average benchmark score is zero, and the nearestRivals list is empty. This absence of comparative data is itself informative. The 50th percentile ranking among all GPUs suggests that, in the broader historical database, this card sits exactly at the median, meaning half of all recorded GPUs score higher and half score lower. However, without specific rival scores or deltaPct values, the raw numbers cannot be contextualized against direct competitors.

The pixel rate is 1.296 GPixel/s, and the texture rate matches at 1.296 GTexel/s. These figures are identical because the card has 4 TMUs and 4 ROPs, operating in a balanced configuration where each pixel output corresponds to a texture fetch. The memory clock is 203 MHz, with an effective data rate of 406 Mbps, which is a doubled rate typical of DDR memory. This low clock speed, combined with the 128-bit bus, yields a bandwidth of 6.496 GB/s. In practical terms, these numbers indicate a card designed for early 2000s CAD and DCC workloads, where polygon throughput and fill rates were the primary metrics, not modern shader complexity.

The lack of a fp32 or fp16 value means the card does not have a meaningful unified shader performance figure, as it predates the unified shader architecture. The R300 architecture used separate pixel and vertex pipelines, and the data shows no shading units listed, only fixed-function TMUs and ROPs. This makes direct comparisons to modern GPUs impossible, but within its own era, the 50th percentile ranking implies it was a mid-pack workstation card, neither entry-level nor top-tier.

Who Should Consider It

Given the benchmark data, this card is not suitable for modern gaming or high-resolution workloads. The 64 MB DDR memory and 6.496 GB/s bandwidth are grossly insufficient for any resolution above 800x600 in contemporary titles, and even then, texture-heavy scenes would exceed the memory capacity. The 1.296 GPixel/s fill rate limits the card to very low polygon counts and simple shading effects.

The intended audience is professionals using early 2000s software that relied on DirectX 9.0 (9_0) and OpenGL 2.0 APIs. For wireframe CAD modeling, 2D drafting, or basic 3D visualization in applications from that period, the card provides hardware acceleration for viewport transformations and lighting. Users working with large assemblies or complex scenes would find the 64 MB memory a hard constraint, as geometry and texture data would spill into system memory over the AGP 8x bus, causing stutters.

The card is end-of-life, so no current software vendor targets it. Those who require it are likely maintaining legacy systems or running specific industrial software that has not been updated. The 50th percentile ranking suggests it was adequate for mainstream workstation tasks of its day but was not a high-end solution. Users should expect it to handle light 3D modeling and 2D UI acceleration, but not simulation, rendering, or any GPU-accelerated compute.

Ray Tracing and Feature Set

The FireGL 9600 T2-64S has no ray tracing cores and no tensor cores, as these technologies did not exist in the R300 architecture. The API support is limited to DirectX 9.0 (9_0) and OpenGL 2.0, with no Vulkan support listed. This means the card cannot accelerate any modern ray-traced effects, and even the DirectX 9.0 support is at the baseline 9_0 feature level, which lacks later shader model enhancements.

The fixed-function pipeline is evident in the absence of shading units. The 4 TMUs and 4 ROPs are the only processing elements, and they operate at the same clock rate as the core (implied by the identical pixel and texture rates). The card supports AGP 8x bus interface, which provides a theoretical 2 GB/s transfer rate, but the actual bottleneck is the 6.496 GB/s memory bandwidth.

For workstation use, the feature set is adequate for OpenGL 2.0-based CAD software of the era, which relied on fixed-function transformations and lighting. The lack of programmable shaders means it cannot run early shader model 2.0 effects efficiently, though it may support them in a limited software fallback. The single LFH60 display output is a legacy connector that requires a splitter cable for dual displays, but the card itself is single-slot with no power connectors, simplifying installation.

How It Compares

As the nearestRivals list is empty, there are no direct competitor comparisons available in the data. The 50th percentile ranking is the only positional reference, indicating that the card sits at the midpoint of the historical GPU performance distribution. This suggests it was outclassed by high-end workstation cards of its time, such as those with larger memory pools or higher fill rates, but it also outperformed budget and integrated solutions.

The predecessor is listed as "Fire GL," which indicates a lineage of professional 3D accelerators, and the successor is "FirePro Terascale," which represents a major architectural shift with unified shaders and much higher raw compute. The FireGL 9600 T2-64S sits between these two generations as a transitional product that maintained the fixed-function approach while adopting the newer AGP 8x interface.

Without rival data, one can only infer that the empty nearestRivals field means the benchmark database does not have comparable scores from other cards in the same performance tier, likely due to the card's age and low sampling. The 50th percentile is therefore a default mid-point assignment rather than a measured median against active peers.

Power and Cooling

The card does not list a specific TDP in the data, which is typical for early 2000s workstation cards that often operated within the AGP slot's power budget. The suggested PSU is 200 W, which is a modest requirement by modern standards, indicating that the card draws very little power. The slot width is single-slot, and there are no power connectors, meaning the card relies entirely on the AGP slot for power delivery. This simplifies installation in legacy systems with small power supplies.

Cooling is not explicitly specified, but the lack of power connectors and the 130 nm process node suggest a passive or small active cooler would suffice. The 200 W PSU recommendation is for the entire system, not just the card, and it implies a low overall system draw. Users building a period-correct machine should ensure their power supply meets this 200 W figure, but any modern PSU far exceeds this.

The single-slot design and lack of additional power requirements make this card easy to fit in compact workstations. The AGP 8x interface is backward compatible with AGP 4x slots, though performance would be limited by the older bus standard. The card's end-of-life status means no current driver support, so cooling solutions are obsolete, but the low power draw reduces heat output, allowing for a simple heatsink.

FAQ

Q: What is the memory size and type of the ATI FireGL 9600 T2-64S?

A: The card has 64 MB of DDR memory with a 128-bit bus width, providing a bandwidth of 6.496 GB/s.

Q: Does this card support ray tracing?

A: No, the card has no ray tracing cores or tensor cores, and its API support is limited to DirectX 9.0 (9_0) and OpenGL 2.0 with no Vulkan support.

Q: What power supply is recommended for this card?

A: The suggested PSU is 200 W, and the card requires no power connectors, drawing power entirely from the AGP 8x slot.

Q: What is the pixel and texture fill rate?

A: Both the pixel rate and texture rate are 1.296 GPixel/s and 1.296 GTexel/s, respectively, based on 4 ROPs and 4 TMUs.

Q: How does this card compare to other GPUs?

A: The card is at the 50th percentile among all GPUs in the database, but has an empty nearestRivals list, so no direct percentage comparisons are available.

Q: What display outputs does it have?

A: It has a single LFH60 output, which is a legacy connector that typically requires a splitter for dual-display setups.

Memory Subsystem

The memory subsystem consists of 64 MB of DDR memory on a 128-bit bus, operating at 203 MHz with an effective data rate of 406 Mbps. The bandwidth is calculated as 6.496 GB/s, which is the product of the bus width and effective clock rate. This bandwidth figure is extremely low by modern standards, but for early 2000s workstation workloads, it was sufficient for moderate texture sizes and framebuffer operations.

The 64 MB capacity is the most limiting factor for high resolutions. At 1920x1080 with 32-bit color, the framebuffer alone would consume approximately 8 MB, leaving little room for textures and geometry. This forces the card to use the AGP 8x bus for texture paging, which is far slower than local memory. For 1280x1024 or lower resolutions, the card can hold a reasonable working set, but any scene with high-detail textures would exceed the 64 MB limit.

The DDR type is first-generation double data rate, which transfers data on both clock edges, hence the 406 Mbps effective rate from the 203 MHz clock. This is a significant improvement over single data rate SDRAM, but the 128-bit bus is the key factor in achieving 6.496 GB/s. A 64-bit bus would have halved this figure, so the 128-bit width is a deliberate choice to maximize bandwidth within the memory technology constraints of the era.

The memory clock is not linked to the core clock in this architecture, and the lack of a listed core clock suggests a fixed operating point. The 6.496 GB/s bandwidth is consumed by both the pixel rate (1.296 GPixel/s) and texture rate (1.296 GTexel/s), meaning each pixel and texture fetch has approximately 5 bytes of bandwidth available. This is adequate for basic color and depth operations, but not for multi-texturing or high dynamic range effects, which would cause immediate bandwidth saturation.