NVIDIA GeForce RTX 5070

The NVIDIA GeForce RTX 5070 is a GeForce 50-series graphics card built around the GB205 chip and Blackwell 2.0 architecture. TSMC's 5 nm process is the manufacturing node, and the die contains 31,100 million transistors on a 263 mm² surface, giving a transistor density of 118.3M per mm². The card carries 12 GB of GDDR7 memory on a 192-bit bus. Its average benchmark score is 41687, which places the RTX 5070 at the 83rd percentile of all GPUs in the database. Production status is Active.

Power and Cooling — TDP, PSU recommendation, connector requirements

The RTX 5070 has a TDP of 250 W. NVIDIA's suggested power supply for the card is 600 W. Power input is provided through a single 1x 16-pin connector, so the physical power path is a single connector rather than multiple 8-pin inputs. The card is dual-slot in width and measures 245 mm (9.6 inches) in length, 115 mm (4.5 inches) in height, and 40 mm (1.6 inches) in width. These dimensions are the case-compatibility figures listed for the card. The host interface is PCIe 5.0 x16. For system planning, the relevant numbers are the 250 W TDP and the 600 W suggested PSU; the 1x 16-pin connector is the required power connection, and the dual-slot footprint defines its installation space.

Ray Tracing and Feature Set

The RTX 5070 is a Blackwell 2.0 part with 48 RT cores and 192 Tensor cores. The shader setup includes 6144 shading units, 192 TMUs, and 80 ROPs. FP32 compute throughput is 30.87 TFLOPS, and FP16 throughput is also 30.87 TFLOPS with a 1:1 ratio. Pixel fill rate is 201.0 GPixel/s, and texture fill rate is 482.3 GTexel/s. On the API side, the card supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. Display output is handled by 1x HDMI 2.1b and 3x DisplayPort 2.1b. The 48 RT cores and 192 Tensor cores are the listed hardware resources for RT and tensor workloads, and the API list covers the modern graphics entry points for this card.

Benchmark Performance

The RTX 5070's aggregate average benchmark score is 41687, with a percentile vs all GPUs of 83. In 3DMark Steel Nomad DX12, the card scores 5077. Geekbench OpenCL returns 185269, while Geekbench Vulkan returns 179413. The Passmark results are more varied across API generations: DirectX 9 scores 320, DirectX 10 scores 180, DirectX 11 scores 277, DirectX 12 scores 108, G2D scores 1305, G3D scores 29137, and GPU compute scores 15787. The base clock is 2325 MHz, and the boost clock is 2512 MHz. The Passmark DirectX 12 result is lower than the DirectX 11 result, while the 3DMark Steel Nomad DX12 figure is substantially higher; these are different workloads that exercise different parts of the GPU. The average score is the summary figure that places the RTX 5070 at the 83rd percentile.

The nearest rivals in the database form a tight cluster. AMD Radeon Pro 5300 has an average score of 41610 and a deltaPct of 0.2. NVIDIA Tesla M40 has an average score of 41897 and a deltaPct of -0.5. NVIDIA GeForce RTX 3090 has an average score of 41441 and a deltaPct of 0.6. AMD Radeon Pro 580X has an average score of 41991 and a deltaPct of -0.7. The RTX 5070's average of 41687 sits inside this group, and the rival averages range from 41441 to 41991. The deltaPct values are small, indicating close aggregate performance in all four pairings. Against the AMD Radeon Pro 5300 and the GeForce RTX 3090, the positive deltaPct values correspond to the RTX 5070 having the higher aggregate score. Against the Tesla M40 and the Radeon Pro 580X, the negative deltaPct values correspond to those cards having the higher aggregate score.

FAQ

Q: What power supply is recommended for the RTX 5070?

A: The suggested PSU is 600 W, the card TDP is 250 W, and power is delivered through a single 1x 16-pin connector.

Q: What memory configuration does the RTX 5070 use?

A: It has 12 GB of GDDR7 on a 192-bit bus, with a memory clock of 1750 MHz / 28 Gbps effective and a bandwidth of 672.0 GB/s.

Q: How does the RTX 5070 compare to its nearest rivals?

A: The average score is 41687. The nearest rivals are AMD Radeon Pro 5300 with 41610 and a deltaPct of 0.2, NVIDIA Tesla M40 with 41897 and -0.5, NVIDIA GeForce RTX 3090 with 41441 and 0.6, and AMD Radeon Pro 580X with 41991 and -0.7.

Q: What is the launch MSRP?

A: The launch MSRP is 549 USD.

Q: Which APIs are supported?

A: The card supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Q: How many RT and Tensor cores are present?

A: There are 48 RT cores and 192 Tensor cores.

Who Should Consider It

The RTX 5070 is positioned at the 83rd percentile of all GPUs, with an average benchmark score of 41687. That places it in the upper band of the database's tracked GPU pool. The 3DMark Steel Nomad DX12 score of 5077 is the stronger modern DirectX 12 result in the recorded set, and the Passmark G3D score of 29137 points to solid rasterization throughput. Geekbench OpenCL at 185269 and Vulkan at 179413 indicate compute-capable performance across both APIs. For high-resolution rendering, the 672.0 GB/s memory bandwidth is the key throughput figure, while the 12 GB frame buffer is the fixed capacity for texture-heavy workloads. The API support for DirectX 12 Ultimate and Vulkan 1.4 makes it a match for modern rendering paths. The 80 ROPs and 482.3 GTexel/s texture fill rate are the output-stage figures behind high-resolution frame generation. Users who rely heavily on legacy DirectX 9/10 paths should look at the Passmark DirectX 9 score of 320 and DirectX 10 score of 180 as those specific API results. The aggregate average score remains the best single summary of the card's position.

Memory Subsystem

The memory subsystem consists of 12 GB of GDDR7 on a 192-bit bus. The listed memory clock is 1750 MHz / 28 Gbps effective, and total bandwidth is 672.0 GB/s. This bandwidth figure is the rate at which data can move across the memory bus, which becomes especially important at high resolutions where frame buffers and texture sets grow large. The 12 GB capacity is the maximum amount of resident data available, and the 672.0 GB/s bandwidth is the throughput that feeds it. The 192-bit bus width is the interface between the GPU and the memory chips. The pixel rate of 201.0 GPixel/s and texture rate of 482.3 GTexel/s are the output throughput numbers associated with the 80 ROPs and 192 TMUs. Together with the GDDR7 memory type and the effective speed of 28 Gbps, these figures define the memory and output profile of the RTX 5070.