AMD Radeon RX 580 vs NVIDIA GeForce GTX 1070 Ti Comparison

The GTX 1070 Ti and RX 580 are both end-of-life cards from the same era, but the recorded data shows they were never really in the same performance class. The NVIDIA card wins 10 of 11 head-to-head benchmarks, including every gaming and compute test except one Apple-specific API result. The RX 580 counters with a smaller die, a shorter PCB, and a very different ratio of launch positioning. Here is how the data breaks down.

The Verdict

The database makes this a straightforward call for raw performance. The GTX 1070 Ti posts an average benchmark score of 14277 against 12928 for the RX 580, sits in the 56th percentile versus 53rd versus all GPUs, and wins the flagship 3DMark Steel Nomad DX12 test by 46.4% (1471 versus 1005). Passmark G3D, the broadest gaming aggregate in the record, lands at 14674 versus 8813, a 66.5% gap in NVIDIA's favor.

The RX 580's case rests on two points. First, it wins Geekbench Metal outright, 45235 versus 22180, which matters if you run macOS workloads. Second, its physical footprint is shorter at 241 mm versus 267 mm, which helps in compact builds. Both cards share identical power infrastructure: a 180 W versus 185 W TDP, a single 8-pin connector, a dual-slot width, and the same 450 W suggested PSU. If gaming or GPU compute on Windows or Linux is the goal, the 1070 Ti is the pick. If Metal on Apple platforms is the workload, the RX 580 is the only one of the two that the data supports.

Where Each One Wins

The 1070 Ti dominates across effectively the entire gaming spectrum. It wins Passmark DirectX 9 by 66.9%, DirectX 10 by 95.7%, DirectX 11 by 78.3%, and DirectX 12 by 18.6%. Note the pattern: the older the API, the bigger the margin. The RX 580 claws back ground in DX12, where its deficit shrinks to 18.6%, but it never crosses the line in any Passmark DirectX test.

The 1070 Ti also sweeps the modern low-level API benchmarks: Geekbench OpenCL (49219 versus 37453, up 31.4%) and Geekbench Vulkan (61006 versus 45173, up 35%). Compute is the most lopsided category of all, with Passmark GPU Compute at 7165 versus 3488, a 105.4% advantage, literally double the throughput.

The RX 580's single win is Geekbench Metal at 45235 versus 22180, where it leads by 51%. That is not a minor edge, it is a rout, and it is the one workload where the Polis-based card is clearly the right tool. It also edges closer in Geekbench OpenCL and Vulkan than in the Passmark gaming tests, suggesting its architecture holds up better in general-purpose compute than in pure rasterization.

Architecture Differences

These are two fundamentally different designs. The 1070 Ti uses NVIDIA's Pascal architecture on the GP104 die, built on TSMC's 16 nm process, packing 7200 million transistors into a 314 mm² die for a density of 22.9M per mm². The RX 580 uses AMD's GCN 4.0 architecture on the Polaris 20 die, built on GlobalFoundries' 14 nm process, with 5700 million transistors in a 232 mm² die at 24.6M per mm². The AMD part is actually denser despite being the smaller, less complex chip.

On the shader side, the 1070 Ti has 2432 shading units, 152 TMUs, and 64 ROPs, against 2304 shading units, 144 TMUs, and 32 ROPs for the RX 580. The ROP count is the most telling difference: double the ROPs is a major reason the 1070 Ti posts a pixel rate of 107.7 GPixel/s versus 42.88 GPixel/s for the RX 580. Texture rate is 255.8 GTexel/s versus 193.0 GTexel/s. Raw FP32 compute lands at 8.186 TFLOPS versus 6.175 TFLOPS.

The FP16 story is inverted and worth noting. Pascal throttles half-precision to 127.9 GFLOPS at a 1:64 ratio, while GCN 4.0 runs FP16 at full speed, 6.175 TFLOPS at 1:1. Anyone doing half-precision work on AMD's stack gets dramatically more throughput than the NVIDIA card's FP16 path provides.

Memory is nearly identical: both carry 8 GB of GDDR5 on a 256-bit bus, with bandwidth at 256.3 GB/s versus 256.0 GB/s, and both run 8 Gbps effective memory. Clocks differ, with the 1070 Ti at 1607 MHz base and 1683 MHz boost against 1257 MHz base and 1340 MHz boost. Both use PCIe 3.0 x16.

Software support diverges at the API level. The 1070 Ti reports DirectX 12 at feature level 12_1, Vulkan 1.4, and OpenGL 4.6. The RX 580 reports DirectX 12 at 12_0, Vulkan 1.3, and OpenGL 4.6. Neither card has RT cores or tensor cores.

FAQ

Q: Which card is faster overall?

A: The GTX 1070 Ti, decisively. It wins 10 of 11 head-to-head benchmarks and holds an average score of 14277 versus 12928.

Q: Does the RX 580 win anything?

A: Yes, Geekbench Metal, where it scores 45235 versus 22180, a 51% lead. This matters only for Metal-based workloads.

Q: How close do they get in modern APIs?

A: Passmark DirectX 12 is the tightest gaming test, with the 1070 Ti ahead by 18.6%. Geekbench Vulkan is a 35% gap in NVIDIA's favor.

Q: Do they need different power supplies?

A: No. Both are dual-slot cards with one 8-pin connector and a 450 W suggested PSU. TDP is 180 W versus 185 W.

Q: Which has more memory bandwidth?

A: Effectively a tie: 256.3 GB/s versus 256.0 GB/s, both from 8 GB of GDDR5 on a 256-bit bus.

Q: Which supports the newer Vulkan version?

A: The 1070 Ti reports Vulkan 1.4, while the RX 580 reports Vulkan 1.3. The 1070 Ti also reports the higher DirectX feature level, 12_1 versus 12_0.

Head-to-Head Benchmarks

Start with the marquee numbers. In 3DMark Steel Nomad DX12, the 1070 Ti scores 1471 against 1005, a 46.4% win that frames every gaming comparison here. Passmark G3D reinforces it at 14674 versus 8813, up 66.5%.

The generational gaming pattern is stark. Passmark DirectX 9: 207 versus 124, up 66.9%. DirectX 10: 90 versus 46, up 95.7%. DirectX 11: 107 versus 60, up 78.3%. Then DirectX 12 tightens to 51 versus 43, up 18.6%. The 1070 Ti wins every tier, but the gap compresses as APIs get more modern, which tells you the Polaris architecture was designed with low-level APIs in mind even if it could not match Pascal's rasterization throughput.

Compute is where it gets ugly for the RX 580. Passmark GPU Compute reads 7165 versus 3488, a 105.4% advantage for the 1070 Ti. Geekbench OpenCL goes to NVIDIA 31.4% (49219 versus 37453), and Geekbench Vulkan goes to NVIDIA 35% (61006 versus 45173). Even the 2D test favors the 1070 Ti, 876 versus 769, a 13.9% edge.

Then there is Metal. Geekbench Metal is the sole RX 580 victory and it is enormous: 45235 versus 22180. The delta is 51% in AMD's favor, the single largest gap in either direction across the entire head-to-head record. For Apple-platform compute, the ranking of these two cards completely flips.

Specification Differences

• Chip and architecture: GP104 (Pascal, GeForce 10) versus Polaris 20 (GCN 4.0, RX 500 generation)
• Process and foundry: 16 nm at TSMC versus 14 nm at GlobalFoundries
• Die and transistors: 314 mm² with 7200 million transistors versus 232 mm² with 5700 million; density is 22.9M/mm² versus 24.6M/mm²
• Clocks: 1607 MHz base and 1683 MHz boost versus 1257 MHz base and 1340 MHz boost; memory runs at 2002 MHz versus 2000 MHz (both 8 Gbps effective)
• Shading units: 2432 versus 2304
• TMUs: 152 versus 144
• ROPs: 64 versus 32
• Pixel rate: 107.7 GPixel/s versus 42.88 GPixel/s
• Texture rate: 255.8 GTexel/s versus 193.0 GTexel/s
• FP32: 8.186 TFLOPS versus 6.175 TFLOPS
• FP16: 127.9 GFLOPS at 1:64 versus 6.175 TFLOPS at 1:1, the one spec where the RX 580 is categorically stronger
• TDP: 180 W versus 185 W
• Outputs: DVI plus HDMI 2.0 plus three DisplayPort 1.4a versus HDMI 2.0b plus three DisplayPort 1.4a (no DVI)
• API support: DirectX 12_1, Vulkan 1.4 versus DirectX 12_0, Vulkan 1.3 (both OpenGL 4.6)
• Length: 267 mm versus 241 mm
• Release date: November 2017 versus April 2017
• Launch MSRP: 399 USD versus 229 USD
• Succession: the 1070 Ti sits between GeForce 900 and GeForce 20; the RX 580 succeeded Arctic Islands and was followed by Vega

Both cards are end-of-life, both carry 8 GB of GDDR5 on a 256-bit bus at effectively identical bandwidth, both are dual-slot with one 8-pin connector and a 450 W suggested PSU, and both use PCIe 3.0 x16. The performance gap, however, is one-sided everywhere except Metal.