NVIDIA GeForce 6800 Ultra DDL

The NVIDIA GeForce 6800 Ultra DDL is an AGP Pro 8x card built around the NV40 chip and the Curie architecture. It was released on April 13, 2004 with a launch MSRP of 599 USD. The card carries 256 MB of GDDR3 on a 256-bit bus, runs at a 100 W TDP, is dual-slot, uses no auxiliary power connectors, and provides 2x DVI outputs. It belongs to the GeForce 6 AGP generation in the 6800 family, manufactured by TSMC on a 130 nm process with 222 million transistors on a 287 mm² die. The database record lists no benchmark scores, no nearest rivals, an average benchmark score of 0, and a 50th-percentile position among all GPUs.

Who Should Consider It

The 6800 Ultra DDL is for systems built around an AGP Pro 8x slot. No benchmark scores are recorded, so resolution and settings recommendations cannot be drawn from measured game runs. The available specification fields provide the only basis for guidance. The card is 305 mm long, or 12 inches, and takes up two slots, so case clearance is an actual installation requirement. It draws 100 W according to the TDP field, while the suggested PSU is 300 W, and there are no auxiliary power connectors. A system planning to use this card should respect that 300 W suggested PSU figure.

With 16 TMUs and 16 ROPs, the board has matching pixel and texture throughput: 6.400 GPixel/s and 6.400 GTexel/s. That symmetric setup points to a card designed for balanced fill-rate work rather than an unusual split between texture and pixel processing. The 256 MB GDDR3 frame buffer with 35.20 GB/s bandwidth is the main resource constraint for texture-heavy scenes. The two DVI outputs make this a natural fit for a dual digital monitor configuration on an AGP-era system. The DirectX 9.0c (9_3) API support places it within the DirectX 9 generation, and the OpenGL support is listed as 2.0.3 full plus 2.1 partial. Anyone shopping for this card today is dealing with an end-of-life product, so the practical context is legacy system building rather than a current-generation configuration. Because the benchmarks array is empty, no settings target such as a specific resolution or detail level can be verified from this data; the performance envelope rests on the listed pixel rate, texture rate, and memory bandwidth.

How It Compares

The nearestRivals array in the database record is empty. That means there are no rival names, no rival scores, and no deltaPct values available for comparison. This page cannot report exact percentage deltas against any specific competing product because the data does not contain those entries. The only relational information in the record is NVIDIA’s own product ordering: the predecessor field lists GeForce FX, and the successor field lists GeForce 7 AGP. That places the 6800 Ultra DDL between the GeForce FX AGP products and the GeForce 7 AGP products in NVIDIA’s sequence. The series field is null in the record, but the generation field specifies GeForce 6 AGP (6800). Without nearestRivals data, the overall percentileVsAllGpus value of 50 is not anchored to any direct rival comparison in this record. No inference such as “better than X by Y percent” can be made from the available facts, because no rival scores or deltaPct values are present.

Ray Tracing and Feature Set

The feature set is defined by the absence of modern acceleration blocks and the presence of a legacy API set. The data lists no RT cores and no tensor cores, so hardware-accelerated ray tracing is not represented in this record. No tensor core workload acceleration appears either. The API support is DirectX 9.0c with the 9_3 feature level, OpenGL 2.0.3 as full support, OpenGL 2.1 as partial support, and no Vulkan support. That makes the DirectX 9_3 feature level the upper bound for the stated API environment.

The underlying architecture is Curie, implemented on the NV40 chip. TSMC manufactured it on a 130 nm process, with 222 million transistors on a 287 mm² die, resulting in a transistor density of 773.5K per mm². The core configuration lists 16 TMUs and 16 ROPs. No shading unit count is recorded, and no FP32 or FP16 throughput figures are present. The display output configuration is 2x DVI, which matches the DDL variant name in the product title. The card is dual-slot, so installation needs two adjacent slots. No base, boost, or game clock is listed for the GPU; the only clock present in the record is the memory clock of 550 MHz with 1100 Mbps effective data rate. That memory clock describes the transfer rate of the GDDR3 memory, not the core speed. The power delivery section lists no auxiliary power connectors, while the TDP is 100 W and the suggested PSU is 300 W.

FAQ

Q: What API feature level does the card support?

A: The card supports DirectX 9.0c with the 9_3 feature level, OpenGL 2.0.3 as full, and OpenGL 2.1 as partial. Vulkan is not listed.

Q: Does it have hardware ray tracing?

A: No. The data lists no RT cores and no tensor cores.

Q: What are the memory specifications?

A: The card has 256 MB of GDDR3 on a 256-bit bus, with a memory clock of 550 MHz, an effective data rate of 1100 Mbps, and a bandwidth of 35.20 GB/s.

Q: What power supply is suggested?

A: The suggested PSU is 300 W. The TDP is 100 W, and the card uses no auxiliary power connectors.

Q: When was it released, and is it still in production?

A: It was released on April 13, 2004. The production status is end-of-life.

Q: What slot and outputs does it use?

A: It uses an AGP Pro 8x bus interface, provides 2x DVI outputs, and is a dual-slot card.

Benchmark Performance

The benchmark section of the record contains no measured results. The benchmarks array is empty, and the average benchmark score is 0. The percentileVsAllGpus value is 50, but with no score entries, that percentile is not a tested average. There are no nearestRivals entries, so no exact deltaPct comparisons can be calculated. The only performance-related quantitative values in the data are the specification-level rates: pixel rate of 6.400 GPixel/s, texture rate of 6.400 GTexel/s, and memory bandwidth of 35.20 GB/s. Those numbers describe the potential throughput of the 16 ROPs and 16 TMUs.

No base clock, boost clock, or game clock is recorded for the GPU. The only clock in the data is the memory clock, at 550 MHz with 1100 Mbps effective data rate. There is no shading unit count, and no FP32 or FP16 performance figures are listed. As a result, shader throughput cannot be quantified from this record. Any statement about the card’s benchmark result would have to rely on the empty score field, which is not usable. The percentileVsAllGpus value of 50 is the single ranking indicator in the record, but the average benchmark score of 0 does not support interpreting it as a measured performance score. The data allows a specification-based assessment, but it does not allow a benchmark-based comparison with exact percentage deltas against any rival product.

Memory Subsystem

The memory subsystem is clearly documented. The card uses 256 MB of GDDR3 on a 256-bit bus, with a memory clock of 550 MHz and an effective data rate of 1100 Mbps. These figures combine to a total memory bandwidth of 35.20 GB/s. That bandwidth is the ceiling for fetching textures, writing rendered pixels, and moving frame-buffer data. The 256-bit bus width provides the path between the GDDR3 memory and the card’s 16 ROPs. The pixel rate of 6.400 GPixel/s represents how much pixel output those ROPs can push toward the memory subsystem.

For high-resolution use, the 256 MB capacity is a hard limit on how much texture and frame-buffer data can be held on board. The card’s 2x DVI outputs mean two digital displays are supported; if both displays operate at high resolutions, they share the same 35.20 GB/s bandwidth and the same 256 MB buffer. The memory clock is listed as 550 MHz, and the effective data rate of 1100 Mbps indicates a double data rate transfer, consistent with the GDDR3 type. No memory benchmarks are present in the record, so this analysis is derived from the specified capacity, type, bus width, memory clock, and bandwidth rather than from tested memory performance.