NVIDIA GeForce GTX 1080 Ti vs NVIDIA GeForce RTX 2070 Comparison

The GeForce RTX 2070 and the GTX 1080 Ti sit one generation apart, and the recorded data shows a classic clash between a newer architecture and raw, older muscle. The 1080 Ti wins 8 of the 10 head-to-head benchmarks, including every single gaming test in the database, yet the RTX 2070 dominates the newer Vulkan compute workload by more than double and carries hardware the Pascal card simply lacks. Both are end-of-life parts now, but both remain relevant on the used market, so the split matters.

FAQ

Q: Which card is faster overall in the recorded benchmarks?

A: The GTX 1080 Ti. It wins 8 of 10 head-to-head tests, including the 3DMark Steel Nomad DX12 test at 2231 versus 1569 and PassMark G3D at 18600 versus 16094. Its overall average benchmark score is 15548, compared with 18789 for the RTX 2070, but the head-to-head record favors the 1080 Ti decisively.

Q: Does the RTX 2070 win anything?

A: Yes, two compute tests. Geekbench Vulkan is a blowout: 82521 versus 40511, a 103.7 percent lead for the RTX 2070. It also leads Geekbench OpenCL at 79966 versus 67929, a 17.7 percent advantage.

Q: How do they compare in power consumption and PSU requirements?

A: The RTX 2070 is far lighter on power: a 175 W TDP with a single 8-pin connector and a suggested 450 W PSU. The 1080 Ti runs a 250 W TDP, needs a 6-pin plus an 8-pin connector, and calls for a 600 W PSU.

Q: What are the launch MSRPs?

A: The RTX 2070 launched at an MSRP of 499 USD. The GTX 1080 Ti launched at an MSRP of 699 USD.

Q: Which has more memory?

A: The 1080 Ti, with 11 GB of GDDR5X on a 352-bit bus delivering 484.4 GB/s of bandwidth. The RTX 2070 has 8 GB of faster GDDR6 on a narrower 256-bit bus, at 448.0 GB/s.

Q: Does the 1080 Ti support ray tracing hardware?

A: No. It has no RT cores and no tensor cores. The RTX 2070 has 36 RT cores and 288 tensor cores, and it supports DirectX 12 Ultimate (12_2) versus the 1080 Ti's DirectX 12 (12_1).

Architecture Differences

The GTX 1080 Ti is the GP102 chip on Pascal, built at TSMC on a 16 nm process, with 11,800 million transistors packed into a 471 mm² die at a density of 25.1M per mm². The RTX 2070 is the TU106 chip on Turing, built at TSMC on a 12 nm process, with 10,800 million transistors in a smaller 445 mm² die at 24.3M per mm². The newer node lets Turing do more with fewer transistors in a smaller footprint, but Pascal compensates with sheer scale: 3584 shading units, 224 TMUs and 88 ROPs on the 1080 Ti against 2304 shading units, 144 TMUs and 64 ROPs on the 2070.

That raw spec gap shows up in the theoretical numbers. The 1080 Ti hits 11.34 TFLOPS FP32, 139.2 GPixel/s and 354.4 GTexel/s. The 2070 manages 7.465 TFLOPS FP32, 103.7 GPixel/s and 233.3 GTexel/s. But FP16 tells the opposite story: the RTX 2070 delivers 14.93 TFLOPS at a 2:1 ratio thanks to its tensor cores, while the 1080 Ti manages only 177.2 GFLOPS at a crippling 1:64 ratio. Pascal was simply never designed for mixed-precision work.

Turing also introduces the RT cores (36 of them) and tensor cores (288) that Pascal lacks entirely, plus support for the DirectX 12 Ultimate (12_2) feature level rather than 12_1. Both support OpenGL 4.6 and Vulkan 1.4. The 2070's 2018 launch followed the 1080 Ti's 2017 debut by a generation, with the 10-series preceding the 20-series directly.

The Verdict

For pure gaming performance, the data is unambiguous: the GTX 1080 Ti wins every gaming benchmark recorded. It leads the 3DMark Steel Nomad DX12 test by 29.7 percent, every PassMark DirectX test from 9 through 12, and overall G3D by 13.5 percent. If gaming is the mission and the PSU can handle it, the 1080 Ti is the stronger card.

The RTX 2070 argues back with a much lower 175 W TDP against 250 W, a single 8-pin connector instead of a 6-pin plus 8-pin, a 450 W suggested PSU instead of 600 W, and a shorter, narrower card: 229 mm long and 35 mm wide versus 267 mm and 40 mm. It also wins the Vulkan compute test by 103.7 percent and OpenCL by 17.7 percent, and it carries ray tracing and tensor hardware the 1080 Ti cannot access. Its percentile ranking against all GPUs in the database is 63, versus 58 for the 1080 Ti.

Practical read: builders with a modest power supply, a compact case, or an interest in ray tracing and modern compute APIs should take the RTX 2070. Builders chasing maximum frame rates with headroom in the power budget should take the GTX 1080 Ti.

Specification Differences

• Architecture and chip: Turing TU106 versus Pascal GP102
• Process node: 12 nm versus 16 nm, both at TSMC
• Transistors: 10,800 million versus 11,800 million
• Die size: 445 mm² versus 471 mm²
• Clocks: 1410 MHz base and 1620 MHz boost for the 2070 versus 1481 MHz base and 1582 MHz boost for the 1080 Ti
• Memory: 8 GB GDDR6 at 448.0 GB/s on a 256-bit bus versus 11 GB GDDR5X at 484.4 GB/s on a 352-bit bus
• Shading units, TMUs, ROPs: 2304 / 144 / 64 versus 3584 / 224 / 88
• RT and tensor cores: 36 and 288 on the 2070; neither exists on the 1080 Ti
• FP16: 14.93 TFLOPS (2:1) versus 177.2 GFLOPS (1:64)
• TDP: 175 W versus 250 W
• Power connectors: 1x 8-pin versus 1x 6-pin plus 1x 8-pin
• Suggested PSU: 450 W versus 600 W
• Display outputs: the 2070 adds DVI and USB Type-C; the 1080 Ti carries only HDMI 2.0 and three DisplayPort 1.4a
• DirectX feature level: 12 Ultimate (12_2) versus 12 (12_1)
• Dimensions: 229 x 113 x 35 mm versus 267 x 112 x 40 mm
• Release: October 2018 versus March 2017
• Launch MSRP: 499 USD versus 699 USD

Head-to-Head Benchmarks

The GTX 1080 Ti's biggest win is PassMark GPU Compute at 9632 versus 6411, a 33.4 percent lead. Close behind is 3DMark Steel Nomad DX12, 2231 versus 1569, a 29.7 percent gap, which is the clearest single indicator of modern gaming performance in this set. From there the Pascal card's margins tighten but never flip in gaming: PassMark DirectX 11 goes to it 151 to 129 (14.6 percent), G3D 18600 to 16094 (13.5 percent), 2D graphics 939 to 819 (12.8 percent), DirectX 12 66 to 60 (9.1 percent), DirectX 9 231 to 211 (8.7 percent), and DirectX 10 118 to 111 (5.9 percent).

The RTX 2070's two wins are both in Geekbench compute, and one of them is enormous. Geekbench Vulkan lands at 82521 for the Turing card against 40511 for Pascal, a 103.7 percent lead, more than double the older card's score. That result reflects the 2070's 14.93 TFLOPS FP16 throughput against the 1080 Ti's 177.2 GFLOPS, a gap of nearly three orders of magnitude in mixed precision. Geekbench OpenCL is closer but still favors the 2070 at 79966 versus 67929, a 17.7 percent edge. Note that the 1080 Ti also posts a Geekbench Metal score of 30624, a test the 2070 does not have a recorded result for.

Context from the database: the RTX 2070's nearest rivals by average score include the NVIDIA Tesla K80 at 18866, the AMD Radeon Pro 5700 XT at 18685, the AMD Radeon RX 560X at 18626, and the NVIDIA RTX 2000 Ada Generation at 18954, all within a percent or so. The 1080 Ti clusters with the AMD Radeon R9 M380 at 15521, the AMD Radeon Pro W5500 at 15679, the NVIDIA GeForce RTX 2060 at 15290, and the NVIDIA GeForce GTX 580 at 15283.

Where Each One Wins

GTX 1080 Ti territory: every DirectX gaming test recorded, from legacy DirectX 9 through DirectX 12, plus the 3DMark Steel Nomad stress test and overall PassMark G3D. It also wins general GPU Compute on PassMark by 33.4 percent and 2D graphics by 12.8 percent. Add the 11 GB frame buffer and higher 484.4 GB/s bandwidth, and it is the pick for high-resolution gaming, longevity in memory-hungry titles, and older API workloads. The cost is a 250 W TDP, dual power connectors, a 600 W PSU recommendation, and a physically longer card.

RTX 2070 territory: Vulkan and OpenCL compute, where it leads by 103.7 and 17.7 percent respectively, plus any workload that benefits from its 36 RT cores, 288 tensor cores, or strong 2:1 FP16 throughput. Its DirectX 12 Ultimate (12_2) support covers modern feature levels the 1080 Ti cannot reach. It is also the efficient build choice: 175 W, one 8-pin connector, a 450 W PSU suggestion, a 229 mm length, and a USB Type-C output among its display options. For compact systems, power-limited builds, and compute or ray tracing workloads, the Turing card is the clear data-backed choice.