NVIDIA GeForce GTX 1070 Ti vs NVIDIA GeForce GTX 580 Comparison

FAQ

Q: Which GPU has the higher average benchmark score?

A: The GTX 1070 Ti leads with 14,277 average points, while the GTX 580 scores 15,283. However, the 1070 Ti’s score is dragged down by its inclusion of older DirectX 9/10/11 tests, where it scores 90–207, whereas the 580's only score (Geekbench OpenCL) is 15,283.

Q: How does the GTX 580 compare to its nearest rivals in raw compute?

A: In Geekbench OpenCL, the GTX 580 scores 15,283, placing it within 0.1% of the AMD Radeon 680M (15,270) and 0.6% of the RTX 3050 OEM (15,199). It trails the RTX 2060 (15,290) by just 7 points, effectively a tie.

Q: What is the biggest performance gap in the head-to-head data?

A: The GTX 1070 Ti wins the only shared benchmark, Geekbench OpenCL, by 68.9%. It scores 49,219 versus the GTX 580’s 15,283 — a 3.2x advantage in raw compute throughput.

Q: Which card has higher memory bandwidth, and by how much?

A: The GTX 1070 Ti offers 256.3 GB/s bandwidth versus 192.4 GB/s on the GTX 580 — a 33% advantage. The 1070 Ti also has more memory (8 GB vs 1.5 GB) and a faster effective speed (8 Gbps vs 4 Gbps).

Q: How do their pixel and texture rates compare?

A: The GTX 1070 Ti delivers 107.7 GPixel/s and 255.8 GTexel/s, while the GTX 580 manages 24.70 GPixel/s and 49.41 GTexel/s. That’s a 4.4x pixel-rate and 5.2x texture-rate advantage for the newer card.

Q: What is the percentile ranking difference between the two?

A: The GTX 580 sits at the 58th percentile of all GPUs, slightly above the GTX 1070 Ti’s 56th percentile. This reflects the 580’s single strong OpenCL score versus the 1070 Ti’s broader but mixed test suite.

Where Each One Wins

The data paints a lopsided picture, but each card has a distinct niche. The GTX 580’s only benchmark result — Geekbench OpenCL at 15,283 — places it just 0.1% behind the AMD Radeon 680M and 0.6% ahead of the RTX 3050 OEM. That suggests the Fermi card remains competitive in legacy OpenCL compute workloads, likely due to its 512 shading units and 384-bit bus. In percentile terms, the 580 ranks 58th overall, which is higher than the 1070 Ti’s 56th percentile. For users running OpenCL-centric tasks on older platforms, the 580 still holds its own against modern integrated and entry-level discrete GPUs.

The GTX 1070 Ti wins everywhere else. Its Geekbench OpenCL score of 49,219 is 3.2x higher than the 580’s. It also posts strong results in Vulkan (61,006) and Metal (22,180), which the 580 cannot run at all — the older card lacks Vulkan support entirely. In DirectX 12, the 1070 Ti scores 1,471 in 3DMark Steel Nomad, and its PassMark G3D score of 14,674 shows broad rasterization strength. The 1070 Ti’s win count is 1-0 in head-to-head tests, but its real advantage lies in API coverage: it supports DirectX 12_1, Vulkan 1.4, and modern display outputs, making it viable for current games and compute APIs. The 580 is capped at DirectX 11_0 with no Vulkan, limiting it to older software.

Architecture Differences

The GTX 580 uses the GF110 chip built on TSMC’s 40 nm process, packing 3,000 million transistors into a 520 mm² die. That yields a transistor density of just 5.8M per mm². The architecture is Fermi 2.0, which supports DirectX 12 (11_0) and OpenGL 4.6, but no Vulkan. Its memory subsystem pairs 1,536 MB of GDDR5 on a 384-bit bus with 192.4 GB/s bandwidth. The card has 512 shading units, 64 TMUs, and 48 ROPs. Compute rates are 1.581 TFLOPS FP32, 24.70 GPixel/s pixel fill, and 49.41 GTexel/s texture fill.

The GTX 1070 Ti uses the GP104 chip on TSMC’s 16 nm process, with 7,200 million transistors on a 314 mm² die — a 22.9M per mm² density, nearly 4x higher than the 580. Pascal architecture brings full DirectX 12_1, Vulkan 1.4, and OpenGL 4.6. The memory setup is 8 GB GDDR5 on a 256-bit bus, delivering 256.3 GB/s. The 1070 Ti has 2,432 shading units, 152 TMUs, and 64 ROPs. Its compute specs are 8.186 TFLOPS FP32, 127.9 GFLOPS FP16 (1:64 ratio), 107.7 GPixel/s, and 255.8 GTexel/s. The 16 nm node also enables higher clocks: 1,607 MHz base and 1,683 MHz boost, versus the 580’s unspecified base/boost.

The architectural leap is stark. Pascal’s 16 nm FinFET process allows 2.4x more transistors on a 40% smaller die. The 1070 Ti has 4.75x more shading units, 2.38x more TMUs, and 1.33x more ROPs. Fermi’s 40 nm planar design hits thermal and power walls quickly; the 1070 Ti’s smaller process enables a 244 W to 180 W TDP reduction despite far more compute hardware.

Specification Differences

| Specification | GTX 580 | GTX 1070 Ti |

|---|---|---|

| Chip | GF110 | GP104 |

| Process node | 40 nm | 16 nm |

| Transistors | 3,000 million | 7,200 million |

| Die size | 520 mm² | 314 mm² |

| Transistor density | 5.8M / mm² | 22.9M / mm² |

| Base clock | N/A | 1,607 MHz |

| Boost clock | N/A | 1,683 MHz |

| Memory speed | 4 Gbps effective | 8 Gbps effective |

| Memory size | 1,536 MB | 8 GB |

| Memory bus | 384 bit | 256 bit |

| Bandwidth | 192.4 GB/s | 256.3 GB/s |

| Shading units | 512 | 2,432 |

| TMUs | 64 | 152 |

| ROPs | 48 | 64 |

| Pixel rate | 24.70 GPixel/s | 107.7 GPixel/s |

| Texture rate | 49.41 GTexel/s | 255.8 GTexel/s |

| FP32 | 1.581 TFLOPS | 8.186 TFLOPS |

| FP16 | N/A | 127.9 GFLOPS |

| TDP | 244 W | 180 W |

| Power connectors | 1x 6-pin + 1x 8-pin | 1x 8-pin |

| Suggested PSU | 550 W | 450 W |

| Bus interface | PCIe 2.0 x16 | PCIe 3.0 x16 |

| Display outputs | 2x DVI, 1x mini-HDMI 1.3a | 1x DVI, 1x HDMI 2.0, 3x DisplayPort 1.4a |

| DirectX | 12 (11_0) | 12 (12_1) |

| Vulkan | N/A | 1.4 |

| Dimensions | 267 mm L, 111 mm H | 267 mm L, 112 mm H, 40 mm W |

| Launch MSRP | 499 USD | 399 USD |

The 1070 Ti is longer in height by 1 mm and adds a width dimension (40 mm) that the 580 doesn’t specify. Both are dual-slot cards with identical 267 mm length. The 1070 Ti uses a single 8-pin connector versus the 580’s 6+8-pin combo, and its suggested PSU drops from 550 W to 450 W.

Head-to-Head Benchmarks

The only direct comparison available is Geekbench OpenCL, and it is a rout. The GTX 1070 Ti scores 49,219 against the GTX 580’s 15,283 — a 68.9% delta in favor of the Pascal card. That 3.2x gap is consistent with the raw compute specs: 8.186 TFLOPS versus 1.581 TFLOPS FP32, and 255.8 GTexel/s versus 49.41 GTexel/s. In this workload, the 1070 Ti’s 2,432 shading units simply overwhelm the 580’s 512.

Looking at broader benchmarks, the 1070 Ti’s Vulkan score of 61,006 and Metal score of 22,180 show strong modern API performance, but the 580 has no comparable results — it lacks Vulkan entirely and was never tested in Metal. The 1070 Ti also posts 1,471 in 3DMark Steel Nomad DX12, a test the Fermi card cannot run due to its 11_0 feature level. PassMark scores further illustrate the gap: the 1070 Ti hits 14,674 in G3D and 7,165 in GPU compute, while the 580 has no PassMark data.

The percentile comparison is the one place the 580 looks better. Its 58th percentile versus the 1070 Ti’s 56th suggests that the 580’s single OpenCL score punches above its weight class, landing it near the RTX 2060 (15,290) and ahead of the RX 7600 (15,171). The 1070 Ti’s percentile is dragged down by its PassMark DirectX 9/10/11 scores (90–207), which are legacy tests that heavily penalize modern architectures. In real-world modern workloads, the 1070 Ti is categorically faster, but the niche legacy OpenCL performance of the 580 remains notable for such an old card.