NVIDIA GeForce RTX 2070 SUPER

The NVIDIA GeForce RTX 2070 SUPER is a GeForce 20-series card built around the TU104 GPU and Turing architecture. TSMC manufactures the chip on a 12 nm process, with 13,600 million transistors on a 545 mm² die and a transistor density of 25.0M per mm². It was released on 2019-07-08 with a launch MSRP of 499 USD and is currently listed as end-of-life. The card connects via PCIe 3.0 x16 and provides one HDMI 2.0, three DisplayPort 1.4a, and one USB Type-C output. Its average benchmark score is 22206, placing it in the 65th percentile of all GPUs in this database, between the GeForce 10 and GeForce 30 generations.

Benchmark Performance

The average score of 22206 is the central data point for judging this GPU. Against its nearest listed rivals, the RTX 2070 SUPER trails the RTX A400 by 0.5%, trails the RTX 4060 Mobile by 2.3%, leads the GTX TITAN Z by 0.9%, and leads the AMD Radeon RX 5700 by 2.5%. These are narrow margins; the largest gap in either direction is only 2.5%. In aggregate terms, the 2070 SUPER sits inside a tight performance cluster with all four of those cards.

Individual benchmark results give more texture to that picture. In 3DMark Steel Nomad DX12, the card scores 1651. Geekbench OpenCL returns 98316, and Geekbench Vulkan returns 94918. PassMark G3D is 18169, PassMark G2D is 878, and PassMark GPU compute is 7557. PassMark’s DirectX-specific scores are 223 for DX9, 151 for DX11, 132 for DX10, and 67 for DX12. Those separate DirectX tests are not directly comparable to the 3DMark Steel Nomad result, but they do show how the card behaves across different API benchmarks.

The compute-oriented specifications are also part of the performance picture. The GPU is rated for 9.062 TFLOPS FP32 and 18.12 TFLOPS FP16 (2:1). It carries 2560 shading units, 160 TMUs, 64 ROPs, 40 RT cores, and 320 tensor cores. Pixel rate is 113.3 GPixel/s and texture rate is 283.2 GTexel/s. The base clock is 1605 MHz and the boost clock is 1770 MHz. These figures are consistent with the OpenCL and Vulkan scores being substantially higher than the PassMark GPU compute score, since those tests stress different execution paths.

Power and Cooling

The RTX 2070 SUPER has a TDP of 215 W. It requires one 6-pin and one 8-pin PCIe power connector, and the suggested power supply rating is 550 W. That makes the connector layout straightforward to plan for: a PSU with both connector types is required, and the 550 W guidance is the specification to check against.

The card is dual-slot and measures 267 mm (10.5 inches) in length, 116 mm (4.6 inches) in height, and 35 mm (1.4 inches) in width. Those dimensions set the physical clearance requirements inside a case. The 215 W TDP is the thermal envelope that the dual-slot design must handle, and the 550 W PSU suggestion is the vendor’s stated recommendation for a system built around it.

How It Compares

RTX A400: The RTX A400 has an average score of 22307. The RTX 2070 SUPER is 0.5% behind, which is the smallest delta among the four nearest rivals. On the basis of average benchmark score, the two cards are effectively equivalent.

GTX TITAN Z: The GTX TITAN Z averages 22006. The RTX 2070 SUPER leads by 0.9%. That is a thin margin, and it puts both cards in the same performance band according to the aggregate data.

RTX 4060 Mobile: The RTX 4060 Mobile averages 22729, making the RTX 2070 SUPER 2.3% slower. Despite the “Mobile” designation, this is the only rival in the group that clearly outranks the 2070 SUPER in average score.

AMD Radeon RX 5700: The RX 5700 averages 21670. The RTX 2070 SUPER is 2.5% higher, and this is the largest delta in the group. Still, a 2.5% average lead is modest rather than decisive.

FAQ

Q: What chip and architecture does the RTX 2070 SUPER use?

A: It uses the TU104 chip on the Turing architecture, fabricated by TSMC on a 12 nm process. The die contains 13,600 million transistors on a 545 mm² area.

Q: How much memory does it have, and what kind?

A: It has 8 GB of GDDR6 on a 256-bit bus. The memory clock is 1750 MHz, listed as 14 Gbps effective, and bandwidth is 448.0 GB/s.

Q: What power supply and power connectors are required?

A: The card has a 215 W TDP, requires one 6-pin and one 8-pin PCIe power connector, and lists a suggested PSU rating of 550 W.

Q: What APIs and display outputs does it support?

A: It supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. Display outputs are one HDMI 2.0, three DisplayPort 1.4a, and one USB Type-C.

Q: How does it compare with its nearest rivals?

A: Its average score is 22206. That is 0.5% below the RTX A400, 0.9% above the GTX TITAN Z, 2.3% below the RTX 4060 Mobile, and 2.5% above the AMD Radeon RX 5700.

Q: Is the card still in production?

A: No. Production status is end-of-life. The release date was 2019-07-08, and the card sits between the GeForce 10 and GeForce 30 generations.

Who Should Consider It

The 65th-percentile ranking shows that a majority of tracked GPUs score higher, but the RTX 2070 SUPER is firmly inside a competitive peer group. The data indicates it is best suited to users who run modern-API workloads. Its 3DMark Steel Nomad DX12 score is 1651, and its Geekbench Vulkan score is 94918; the PassMark DX9 and DX10 scores are 223 and 132, respectively. That split points toward DirectX 12 and Vulkan being the stronger environment for this card.

The memory figures matter for high-resolution use. With 8 GB of GDDR6 and 448.0 GB/s of bandwidth, the card has enough frame buffer and transfer rate for high-resolution textures that stay within that capacity. Its 113.3 GPixel/s pixel rate and 283.2 GTexel/s texture rate define the throughput limits for pixel-heavy and texture-heavy scenes. The aggregate deltas to the nearest rivals are small: 0.5% behind the RTX A400, 0.9% ahead of the GTX TITAN Z, 2.3% behind the RTX 4060 Mobile, and 2.5% ahead of the RX 5700. A buyer should treat the card as a mid-upper-tier modern-API GPU whose practical ceiling is set by the 8 GB frame buffer and the bandwidth figures, not by raw compute headroom.

Memory Subsystem

The memory subsystem is built around 8 GB of GDDR6 on a 256-bit bus. The memory clock is 1750 MHz, listed as 14 Gbps effective, and the resulting bandwidth is 448.0 GB/s. Those three numbers — capacity, bus width, and bandwidth — determine how quickly textures and geometry data can be fed to the 2560 shading units.

At high resolutions, the amount of data staged in the frame buffer rises, and the 8 GB capacity becomes the first constraint. Within that limit, 448.0 GB/s of bandwidth keeps the 160 TMUs and 64 ROPs supplied. The texture rate of 283.2 GTexel/s and pixel rate of 113.3 GPixel/s are the upper bounds for how fast textures can be sampled and pixels written. The memory subsystem is therefore a balanced match for the GPU’s compute throughput: enough bandwidth to support the shader load, and enough capacity to handle high-resolution scenes that do not exceed 8 GB.