NVIDIA GeForce RTX 3090

The NVIDIA GeForce RTX 3090 is a flagship Ampere-architecture GPU built on Samsung's 8 nm process, featuring a GA102 chip with 28,300 million transistors on a 628 mm² die. It launched with a launch MSRP of 1,499 USD and has since moved to end-of-life status, succeeded by the GeForce 40 series. The card is equipped with 10,496 shading units, 328 TMUs, 112 ROPs, 82 RT cores, and 328 tensor cores, delivering 35.58 TFLOPS of FP32 performance and identical FP16 throughput at a 1:1 ratio. The data shows an average benchmark score of 41,441, placing it in the 83rd percentile of all GPUs, with individual test results spanning from DirectX 9 to modern DX12 workloads.

Benchmark Performance

The RTX 3090's aggregate performance, measured by an average score of 41,441 across ten benchmark tests, positions it in the upper echelon of graphics hardware, though not at the absolute top. In 3DMark Steel Nomad DX12, the card scores 5,118, a result that indicates strong directX 12 Ultimate capability. The Geekbench OpenCL score of 191,142 and Vulkan score of 174,310 show a notable gap between compute-oriented APIs and graphics-focused workloads, with OpenCL leading by roughly 9.7%. Passmark results provide a broader picture: the G3D score of 26,645 is the highest in the Passmark suite, while GPU compute hits 15,356, suggesting the card is better optimized for rasterization than raw compute in that specific benchmark.

Comparing to nearest rivals, the RTX 3090 sits within a tight cluster. The AMD Radeon Pro 5300 posts an average score of 41,610, which is 0.4% ahead of the RTX 3090 — a negligible margin that falls within run-to-run variance. The NVIDIA GeForce RTX 3080 Ti trails by 0.5%, with an average score of 41,224, indicating the two siblings are effectively tied in overall throughput. The NVIDIA GeForce RTX 5070 leads by 0.6% at 41,687, while the NVIDIA Tesla M40 lags by 1.1% at 41,897. These deltas are all under 2%, meaning the RTX 3090 delivers performance statistically indistinguishable from its closest competitors in aggregate scoring.

Individual benchmark results reveal where the RTX 3090 excels and where it does not. The Passmark DirectX 9 score of 268 and DirectX 10 score of 182 suggest legacy API performance is adequate but not exceptional, while the DirectX 11 score of 220 and DirectX 12 score of 110 indicate increasingly lower scores on newer APIs in that suite. The G2D score of 1,063 is relatively low, reflecting that 2D workloads are not a strength. The 3DMark Steel Nomad score of 5,118, however, demonstrates that modern DX12 titles run smoothly, as this is a demanding benchmark. The overall percentile rank of 83 means the RTX 3090 outperforms 83% of all GPUs in the database, a solid but not dominant position given its flagship status.

Who Should Consider It

Benchmark results indicate the RTX 3090 is suited for high-resolution gaming at 4K and demanding settings, given its 24 GB VRAM and 936.2 GB/s bandwidth. The 3DMark Steel Nomad DX12 score of 5,118 suggests the card handles modern ray-traced titles at high settings, though the 83rd percentile rank means it is not the fastest option available. For 1440p gaming, the card is likely overkill, as the high FP32 throughput of 35.58 TFLOPS will often exceed display refresh rates in less demanding titles. At 4K, the combination of 24 GB memory and high bandwidth reduces stuttering in texture-heavy scenes, making it a viable choice for enthusiasts who prioritize resolution over frame rate.

For compute workloads, the Geekbench OpenCL score of 191,142 and Vulkan score of 174,310 indicate strong performance in GPU-accelerated applications. The Passmark GPU compute score of 15,356 further confirms this, though the gap between OpenCL and Vulkan suggests some applications may favor one API over the other. Users running machine learning or rendering tasks that leverage FP32 or FP16 (at 1:1 ratio) will find the RTX 3090 capable, but the 0.4% deficit to the AMD Radeon Pro 5300 in average score means professional users should check application-specific benchmarks. The card's 82 RT cores and 328 tensor cores make it suitable for real-time ray tracing and DLSS workloads, though the 83rd percentile rank again implies newer cards offer better performance per frame.

The RTX 3090 is not aimed at budget-conscious builders or those targeting 1080p, as the high memory capacity and bandwidth provide diminishing returns at lower resolutions. Its triple-slot width and 336 mm length also require a spacious case, which is a practical consideration. For users who need maximum VRAM for large datasets or high-resolution textures, the 24 GB GDDR6X is a standout feature, but the average benchmark scores suggest it is not the fastest card in every scenario.

Memory Subsystem

The RTX 3090 features 24 GB of GDDR6X memory on a 384-bit bus, yielding a theoretical bandwidth of 936.2 GB/s. The memory clock runs at 1219 MHz, translating to 19.5 Gbps effective. This configuration provides 50% more capacity than typical 16 GB cards and a 384-bit bus width that enables wide data transfers, critical for 4K gaming and professional workloads. The bandwidth of 936.2 GB/s is among the highest in the database, ensuring that texture streaming and high-resolution assets do not become bottlenecks.

At high resolutions, the memory subsystem's impact is pronounced. A 24 GB pool allows games to load ultra-high-resolution texture packs without exceeding VRAM limits, which is particularly relevant for titles that exceed 16 GB at 4K. The 384-bit bus width reduces the number of memory transactions needed to move data, while the GDDR6X type offers higher transfer rates than standard GDDR6. The pixel rate of 189.8 GPixel/s and texture rate of 556.0 GTexel/s complement the memory bandwidth, allowing the card to fill large frame buffers efficiently.

However, the data shows that memory capacity alone does not guarantee top-tier performance. The RTX 3090's average score of 41,441 trails the RTX 5070 by 0.6%, despite having 24 GB versus the newer card's likely smaller pool. This suggests that memory bandwidth and capacity are necessary but not sufficient for performance leadership; architecture efficiency and clock speeds also play roles. For users running multi-GPU setups or massive compute tasks, the 24 GB capacity is a clear advantage, but for mainstream gaming, the bandwidth of 936.2 GB/s is more than adequate, and the card's 83rd percentile rank reflects that it competes well but does not dominate.

Power and Cooling

The RTX 3090 has a TDP of 350 W, requiring a suggested PSU of 750 W. The power is delivered via a single 12-pin connector, which may necessitate an adapter for standard PSUs. The card is triple-slot in width, measuring 336 mm in length, 140 mm in height, and 61 mm in width, making it one of the larger GPUs in the database. This physical footprint demands a case with ample clearance and good airflow, as the 350 W TDP generates substantial heat.

The 350 W TDP is high but consistent with the card's performance class. The 750 W PSU recommendation provides headroom for the rest of the system, though users with power-hungry CPUs may need more. The single 12-pin connector is a departure from the dual 8-pin connectors common on rival cards, so builders must verify their PSU's compatibility. The triple-slot cooler is designed to dissipate heat effectively, but it also blocks adjacent PCIe slots, reducing expansion options. The card's dimensions — 336 mm length — mean it will not fit in compact cases, so buyers should measure their chassis before purchase.

Thermal performance is not directly measured in the FACT PACK, but the 350 W TDP and triple-slot design imply a robust cooling solution. The pixel rate of 189.8 GPixel/s and texture rate of 556.0 GTexel/s generate significant heat under load, and the card's size is a direct response to that. The end-of-life production status means availability may be limited, and potential buyers should be aware of the power and space requirements before committing.

How It Compares

AMD Radeon Pro 5300: The Radeon Pro 5300 posts an average score of 41,610, which is 0.4% higher than the RTX 3090's 41,441. This margin is effectively negligible, indicating that the two cards perform at parity in aggregate benchmarks. However, the Radeon Pro 5300 is a professional workstation GPU, whereas the RTX 3090 is a consumer flagship, so the comparison favors the RTX 3090 in gaming scenarios despite the statistical tie.

NVIDIA GeForce RTX 3080 Ti: The RTX 3080 Ti scores 41,224 on average, trailing the RTX 3090 by 0.5%. This is a smaller difference than typically expected between a Ti variant and its higher-end sibling, suggesting that the extra 24 GB of VRAM and wider memory bus on the RTX 3090 do not translate into significant performance gains in the tested workloads. The RTX 3090's advantage is likely more pronounced in memory-bound tasks, but the benchmark average shows near-parity.

NVIDIA GeForce RTX 5070: The RTX 5070 leads the RTX 3090 by 0.6%, with an average score of 41,687. This indicates that a newer generation card can match or slightly exceed the older flagship in overall performance, despite likely having less VRAM. The RTX 3090's 83rd percentile rank is the same as what the RTX 5070 would achieve, but the 0.6% delta shows that generational improvements in architecture have closed the gap.

NVIDIA Tesla M40: The Tesla M40 scores 41,897 on average, which is 1.1% ahead of the RTX 3090. This is surprising given the M40's older architecture, but the Tesla M40 is a compute-focused card that likely excels in specific workloads. The RTX 3090's broader feature set — including RT cores, tensor cores, and modern API support like DirectX 12 Ultimate and Vulkan 1.4 — makes it a more versatile choice, even though the M40 edges it out in aggregate score.

FAQ

Q: What is the RTX 3090's average benchmark score?

A: The average benchmark score is 41,441, which places it in the 83rd percentile of all GPUs.

Q: How does the RTX 3090 compare to the RTX 3080 Ti?

A: The RTX 3080 Ti averages 41,224, which is 0.5% lower than the RTX 3090's score, indicating near-identical performance.

Q: What memory configuration does the RTX 3090 use?

A: It uses 24 GB of GDDR6X memory on a 384-bit bus, providing 936.2 GB/s of bandwidth at 19.5 Gbps effective.

Q: What power supply is recommended for the RTX 3090?

A: The suggested PSU is 750 W, and the card has a TDP of 350 W with a single 12-pin power connector.

Q: Is the RTX 3090 faster than the AMD Radeon Pro 5300?

A: No, the Radeon Pro 5300 scores 41,610, which is 0.4% higher than the RTX 3090, though this difference is within margin of error.

Q: What is the card's physical size?

A: The RTX 3090 is 336 mm long, 140 mm high, and 61 mm wide, requiring a triple-slot cooler and a case with ample space.