ATI Radeon 9800 PRO Mac Edition

The ATI Radeon 9800 PRO Mac Edition is an end-of-life AGP Pro 8x graphics card released on 2004-03-03. It is built on the R300 architecture using the R360 chip, produced on TSMC's 150 nm process. The die contains 117 million transistors on 218 mm², a density of 536.7K transistors per mm². The card packs 128 MB of DDR memory behind a 256-bit bus, with a 338 MHz memory clock and 676 Mbps effective data rate; bandwidth is 21.63 GB/s. Its rendering pipeline includes 8 TMUs and 8 ROPs, delivering 3.024 GTexel/s and 3.024 GPixel/s. The board is single-slot, consumes 60 W, uses no auxiliary power connectors, and is paired with a 250 W suggested PSU. The database lists a 50th percentile against all GPUs and an average benchmark score of 0.

How It Compares

The nearestRivals array for this entry is empty, so the fact pack provides no named rivals, scores, or deltaPct values. As a result, this card cannot be positioned against specific competing products using the comparative data normally used for that analysis. The only positional marker is the database-wide percentileVsAllGpus value of 50, placing it at the midpoint of all GPUs in the database. The average benchmark score of 0 is a separate data point, but it does not combine with any rival score to produce a percentage differential.

In the absence of nearestRivals, the product tree offers context: the predecessor is Radeon R200, the successor is Radeon R400 AGP, and the generation field is Radeon R300 (9800). That places the card sequentially between those two listed families. The 50th percentile standing is the only rank visible in the data. There are no rival-specific paragraphs to write because no nearestRivals entries exist in the fact pack.

Ray Tracing and Feature Set

The fact pack is explicit in its null values for rtCores and tensorCores: neither ray tracing cores nor tensor cores are reported. Therefore, this card does not have a listed hardware path for ray tracing or tensor-based acceleration. The feature set that is documented centers on the R300 architecture and the R360 chip. API support is limited to DirectX 9.0 and OpenGL 2.0; Vulkan support is absent from the list.

The fixed-function hardware is represented by 8 TMUs and 8 ROPs, and the corresponding texture and pixel rates are both 3.024 GTexel/s and 3.024 GPixel/s. This symmetry suggests a balanced rasterization stage for the DirectX 9-era feature set. The R360 chip is manufactured on a 150 nm process at TSMC, with 117 million transistors on 218 mm². Transistor density is 536.7K per mm². No shader unit count, FP32, or FP16 values are listed, so the programmable pipeline cannot be quantified beyond the API list and TMU/ROP counts.

Memory Subsystem

The memory subsystem is one of the most concrete parts of this listing. The card has 128 MB of DDR VRAM, connected over a 256-bit bus. The memory clock is 338 MHz, which translates to 676 Mbps effective data rate, and the resulting bandwidth is 21.63 GB/s. At high resolutions, both capacity and bandwidth can become limiting factors: 128 MB is a finite storage pool for textures, depth buffers, and framebuffer data, while 21.63 GB/s bounds the speed at which that pool can be fed.

The 256-bit bus width is a wide pathway, but the DDR memory type and 338 MHz clock keep the effective data rate at 676 Mbps. This configuration is a 2004-era memory design rather than a modern high-bandwidth arrangement. No FP32 or FP16 throughput numbers are listed, so the memory subsystem cannot be tied to compute-oriented performance metrics. For multi-display use, the 2x DVI outputs allow two digital monitors to be driven, and the memory bandwidth must serve whatever resolutions are active on those outputs.

FAQ

Q: What kind of memory and bus does the card use?

A: It uses 128 MB of DDR memory on a 256-bit bus. The memory runs at 338 MHz, or 676 Mbps effective, giving 21.63 GB/s bandwidth.

Q: Does the fact pack list hardware ray tracing or tensor cores?

A: No. The rtCores and tensorCores fields are both null, so no RT or tensor core counts are present in the data.

Q: Which APIs does it support?

A: DirectX 9.0 and OpenGL 2.0 are listed. Vulkan is not listed.

Q: What are the power requirements?

A: The TDP is 60 W, the board is single-slot, it has no power connectors, and the suggested PSU is 250 W.

Q: What is the bus interface and what display outputs does it have?

A: The bus interface is AGP Pro 8x, and the display outputs are 2x DVI.

Q: When was it released and what is its production status?

A: It was released on 2004-03-03 and is listed as end-of-life.

Who Should Consider It

The database's 50th percentile rank puts this card at the median of the recorded GPU distribution, while the average benchmark score of 0 adds no positive benchmark evidence. That combination points toward a user with a legacy AGP Pro 8x system and a workload built around DirectX 9.0 or OpenGL 2.0. The 128 MB memory and 21.63 GB/s bandwidth indicate that lower resolutions and smaller texture sets are a better fit than high-resolution, high-texture scenarios. At higher resolutions, the VRAM capacity can be exhausted quickly and bandwidth can become a bottleneck.

The 8 TMUs and 8 ROPs, with 3.024 GTexel/s and 3.024 GPixel/s, place a throughput ceiling that is most appropriate for older titles or desktop use. The 2x DVI outputs make it suitable for a dual-digital monitor setup in a machine that accepts AGP Pro 8x. Conversely, anyone expecting support for ray-traced effects or tensor-accelerated workloads should look elsewhere, because no RT cores or tensor cores are listed. The end-of-life status further narrows its audience to systems that specifically require this AGP-era card.

Power and Cooling

The card has a TDP of 60 W, which is the primary thermal design figure. It is a single-slot design, so it occupies one slot width. There are no auxiliary power connectors on the listing, which means the power draw is delivered without supplemental connectors. The suggested PSU is 250 W, giving a system-level recommendation rather than a board-level connector requirement. The absence of a power connector is notable because the data lists no external power input.

The single-slot cooler and 60 W TDP are modest enough that the fact pack lists no power connector. No dimensions are provided, so the physical length, height, and width cannot be quantified. The production status is end-of-life, which may affect availability of support. The release date of 2004-03-03 is the only temporal reference associated with the design.

Benchmark Performance

The benchmarks array is empty, and the nearestRivals array is also empty. Consequently, there are no individual benchmark scores, no rival names, and no deltaPct values in the data. The only performance-position fields are avgBenchmarkScore (0) and percentileVsAllGpus (50). A 50th percentile means the card ranks at the median of all GPUs in the database. The zero average score, taken alone, would normally indicate no performance, but in the context of an empty benchmarks array it is a placeholder rather than a measured result.

Because no rival scores are supplied, exact percentage deltas against other cards cannot be computed. What can be quantified are the fixed-function throughput rates: 3.024 GTexel/s texture fill and 3.024 GPixel/s pixel fill. These provide upper bounds for legacy workloads, but they do not translate into a percentage comparison without rival data. The 50th percentile is the sole comparative percentage available. Therefore, the benchmark analysis ends at the database-wide percentile, not at a card-to-card delta.