NVIDIA GeForce 8800 GTS 320

The NVIDIA GeForce 8800 GTS 320 occupies a curious position in the GeForce 8 lineup, pairing the G80 chip’s full shading capability with a deliberately trimmed memory configuration. Built on TSMC’s 90 nm process with 681 million transistors packed into a 484 mm² die, this card launched on February 11, 2007, with a suggested 300 W PSU and a 1x 6-pin power connector. Its 143 W TDP reflects the era’s brute-force approach to GPU design, and the dual-slot cooler hints at the thermal challenge posed by that transistor count. The data shows a part that is neither the flagship nor the entry point, but a specific compromise—one that trades capacity for speed in a way that reshapes its performance profile.

Benchmark Performance

The 8800 GTS 320’s raw compute numbers are defined by its 96 shading units, 24 texture mapping units, and 20 ROPs, operating at a memory clock of 792 MHz (1584 Mbps effective). This yields 228.1 GFLOPS of FP32 performance, a figure that places it firmly in the mid-range of the GeForce 8 generation. The pixel rate of 10.26 GPixel/s and texture rate of 24.62 GTexel/s are direct consequences of those unit counts, and they tell a straightforward story: this card can fill pixels and sample textures at rates that were competitive for its time, but the memory subsystem becomes the bottleneck in many scenarios.

The percentile ranking against all GPUs stands at 50, meaning the 8800 GTS 320 sits exactly at the median of the database’s tracked hardware. This is a statistical marker, not a verdict on gaming capability—it suggests that while newer cards vastly outperform it, the 8800 GTS 320 still outpaces roughly half of all GPUs ever recorded. The average benchmark score is listed as 0, which indicates a lack of standardized test data in the pack, so the analysis must rely on architectural specs and the percentile position.

Without nearestRivals data provided, the comparison framework is limited to what the FACT PACK offers. The shading unit count of 96 is modest by later standards, but it was substantial for 2007, enabling the card to handle shader-heavy workloads that stressed earlier GeForce 7 parts. The FP32 throughput of 228.1 GFLOPS is the key metric here—it represents the card’s theoretical ceiling for compute tasks, and it is roughly double what a typical mid-range card of the prior generation would deliver, though the FACT PACK does not specify those rivals. The benchmark data’s absence means the 8800 GTS 320’s real-world performance must be inferred from its pixel, texture, and shading rates, which collectively indicate a card best suited to 1280x1024 or 1600x1200 resolutions at medium settings.

Power and Cooling

The 8800 GTS 320 carries a TDP of 143 W, a figure that was typical for high-performance GPUs of its era but seems modest next to later dual-chip monsters. The suggested PSU rating is 300 W, which is notably low for a card with this transistor count—the G80 chip was known for power spikes, and the FACT PACK does not specify peak draw beyond the TDP. A single 6-pin power connector provides the electrical feed, which is a single point of connection rather than the dual connectors found on larger cards. The dual-slot cooler is a direct response to the 90 nm process’s heat output; the die size of 484 mm² spreads that heat across a large surface, but the cooler still needs substantial surface area and airflow to keep temperatures in check.

The 267 mm (10.5 inches) length means the card will fit in most mid-tower cases of the period, but it will block an adjacent slot. The power connector requirement is explicit: 1x 6-pin, and the 300 W PSU recommendation assumes a system with a single CPU and modest peripherals. Users with older power supplies from the GeForce 7 era may need to upgrade, as those units often lacked the 6-pin connector or the 12 V rail capacity to handle 143 W TDP plus the rest of the system. The data does not include efficiency figures or load testing, so the thermal solution’s effectiveness is implied by the dual-slot design rather than measured.

Ray Tracing and Feature Set

The 8800 GTS 320 predates dedicated ray tracing and tensor cores—the FACT PACK lists both rtCores and tensorCores as null, meaning this hardware has no specialized acceleration for those workloads. Its feature set is defined by the APIs it supports: DirectX 11.1 (with a 10_0 feature level) and OpenGL 3.3. The DirectX 10_0 level is critical—it means the card can run DirectX 10 titles, which were emerging in 2007, but it lacks the DirectX 10.1 features and later 11.x features that would come with subsequent generations. OpenGL 3.3 support is similarly a point-in-time capability, sufficient for games and applications released around its launch but not for later OpenGL 4.x titles.

The absence of Vulkan support (listed as null) further confines the card to legacy APIs. In practice, this means the 8800 GTS 320 is not a candidate for modern game engines that rely on Vulkan or DirectX 12; its DirectX 11.1 support is nominal, with the 10_0 feature level capping shader models and resource binding at the DirectX 10 specification. The 96 shading units are general-purpose and can process compute workloads, but without tensor cores, any AI-accelerated features are absent. The architecture is Tesla, which was NVIDIA’s first unified shader design—it merged vertex and pixel shaders into a single pool of 96 units, a major departure from the separated pipelines of the GeForce 7 series. This unification is why the FP32 rate of 228.1 GFLOPS is achievable; the units are fully utilized across vertex, geometry, and pixel tasks.

FAQ

Q: What is the memory size and type of the 8800 GTS 320?

A: The card has 320 MB of GDDR3 memory on a 320-bit bus, yielding a bandwidth of 63.36 GB/s.

Q: Does this GPU support ray tracing?

A: No, the FACT PACK lists rtCores as null, meaning there are no dedicated ray tracing cores on this architecture.

Q: What power supply is recommended for this card?

A: The suggested PSU rating is 300 W, and the card requires a single 6-pin power connector.

Q: What is the launch MSRP of the 8800 GTS 320?

A: The launch MSRP is 269 USD.

Q: Which API versions does the card support?

A: It supports DirectX 11.1 (with a 10_0 feature level) and OpenGL 3.3, but no Vulkan support is listed.

Q: How many shading units does the 8800 GTS 320 have?

A: It has 96 shading units, which are unified across vertex and pixel processing in the Tesla architecture.

Who Should Consider It

The 8800 GTS 320’s performance profile, defined by 228.1 GFLOPS of FP32 compute and a 63.36 GB/s memory bandwidth, points to a specific use case: 2007-era gaming at 1280x1024 or 1600x1200 resolutions. The pixel rate of 10.26 GPixel/s is sufficient for those resolutions at medium to high detail settings in DirectX 9 and early DirectX 10 titles. The 320 MB memory capacity is the limiting factor—at 1600x1200, some games with high-resolution textures will exceed that capacity, forcing the card to swap data over the 320-bit bus at 63.36 GB/s. This bandwidth is respectable for the time, but capacity constraints will show up as stuttering in texture-heavy scenes.

For users with a 300 W PSU and a single 6-pin connector, the card is a drop-in upgrade from a GeForce 7 PCIe part, as the predecessor field indicates. It will not handle modern games that require DirectX 11 or Vulkan, so its audience is collectors, retro-build enthusiasts, or those running legacy software. The percentile rank of 50 against all GPUs suggests that, in a historical context, it sits above the median—but that median includes integrated graphics and low-end cards from the past decade. The 90 nm process and 143 W TDP mean it runs hot by modern standards, so adequate case airflow is essential.

Memory Subsystem

The 8800 GTS 320’s memory configuration is its defining characteristic: 320 MB of GDDR3 on a 320-bit bus, running at 792 MHz (1584 Mbps effective). This produces a bandwidth of 63.36 GB/s, which is a balanced figure for the card’s compute capability—the 228.1 GFLOPS FP32 rate requires a certain level of memory throughput to keep the shading units fed. The 320-bit bus is wide, but the 320 MB capacity is the unusual choice. This was a common strategy in the mid-2000s: reduce capacity to hit a price point, relying on bandwidth to mitigate the loss.

At 1600x1200 and above, 320 MB becomes a bottleneck. Frame buffers for high resolutions consume memory quickly, and texture-heavy games can exceed 320 MB, forcing the card to evict and reload data. The 63.36 GB/s bandwidth helps minimize the impact, but it cannot compensate for a hard capacity ceiling. The pixel rate of 10.26 GPixel/s is lower than what the memory bandwidth could theoretically support, indicating that the 20 ROPs are the limiting factor in fill-rate-bound scenarios. The texture rate of 24.62 GTexel/s, driven by 24 TMUs, is more robust, suggesting that games with heavy texture sampling will perform better than those with complex pixel shaders.

How It Compares

The FACT PACK lists no nearestRivals data, so a direct comparison table is impossible. However, the predecessor and successor fields provide context: the GeForce 7 PCIe series was the prior generation, and the GeForce 9 series followed. Against a typical GeForce 7 card, the 8800 GTS 320 offers the unified shader architecture—96 shading units that can be allocated dynamically to vertex or pixel work, whereas GeForce 7 parts had fixed separate pipelines. This architectural shift is why the 8800 GTS 320 achieves 228.1 GFLOPS, a figure that would be significantly higher than any GeForce 7 card, though the FACT PACK does not provide those numbers.

Compared to the GeForce 9 successor, the 8800 GTS 320 is a smaller chip with fewer transistors (681 million vs. later G92 variants that are not specified here). The 90 nm process is also a generation behind, as GeForce 9 parts moved to 65 nm, but the FACT PACK does not confirm that. The 320 MB memory capacity is smaller than what the GeForce 9 series typically offered, and the DirectX 10_0 feature level is the same, so API support is not a differentiator. The percentile rank of 50 places it below the average of the GeForce 9 series, which likely sits higher in the distribution, but again, exact scores are not in the pack. In essence, the 8800 GTS 320 is a bridge between generations—more capable than its predecessor in raw compute, but held back by memory capacity and an older process node relative to its successor.