NVIDIA GeForce GTX 590 vs NVIDIA GeForce GTX 670 Comparison

The GeForce GTX 590 and GeForce GTX 670 represent two distinct generations of NVIDIA flagship and upper-midrange design, and the benchmark data shows a clear shift in performance efficiency between them. While the GTX 590 was the dual-GPU halo product of the Fermi era, the GTX 670 arrived just over a year later as a single-GPU Kepler part. The numbers from the Geekbench OpenCL test reveal that the newer, smaller card is decisively faster in compute workloads, despite the older card's more imposing specifications.

Head-to-Head Benchmarks

The only directly comparable benchmark in the data is the Geekbench OpenCL test, and it is not close. The GTX 670 scores 15,341 points, while the GTX 590 scores 12,830 points. That is a delta of -16.4% from the perspective of the GTX 590, meaning the GTX 670 is approximately 16.4% faster in this specific compute workload. This is a significant margin, especially considering the GTX 590's theoretical advantages in raw memory bandwidth and pixel throughput.

The GTX 590's average benchmark score, which is its only recorded benchmark, sits at 12,830. The GTX 670's average across three different API tests (OpenCL, Metal, and Vulkan) is 12,773, but its OpenCL score alone is far higher at 15,341. The difference between the two cards' average scores is negligible — the GTX 590's average is 12,830 versus the GTX 670's 12,773 — but this is misleading. The average for the GTX 670 is dragged down by its Metal score of 7,424, which is likely an API-specific limitation rather than a reflection of raw hardware capability. In the OpenCL test, which is the only apples-to-apples comparison available, the GTX 670 wins outright.

When placed against their nearest rivals, both cards sit in the same percentile bracket — the 52nd percentile of all GPUs. The GTX 590's closest competitor is the AMD Radeon Pro 455, which scores 12,831, a delta of 0%. The GTX 670's nearest rival is the GTX 590 itself, with a -0.4% delta, confirming that these two cards are considered performance peers in the aggregate database. However, the head-to-head OpenCL result shows the GTX 670 pulling ahead by a wide margin when the same test is run on both.

The texture and pixel rates tell a story of architectural divergence. The GTX 590 has a texture rate of 38.91 GTexel/s and a pixel rate of 19.46 GPixel/s, while the GTX 670 achieves 109.8 GTexel/s and 27.44 GPixel/s respectively. The GTX 670 is roughly 2.8 times faster in texture fill rate and over 40% faster in pixel fill rate, which explains its superior compute performance despite having a narrower memory bus.

The Verdict

The data points to one unambiguous conclusion: the GTX 670 is the better card for modern compute workloads. Its OpenCL score of 15,341 versus the GTX 590's 12,830 represents a 16.4% performance advantage that cannot be ignored. The GTX 670 also does this while drawing 170 W of power compared to the GTX 590's 365 W, needing only a 450 W power supply versus 750 W, and using two 6-pin connectors instead of two 8-pin connectors. It is physically smaller as well: 241 mm long versus 279 mm, and 38 mm wide versus 40 mm.

The GTX 590's only claim to superiority is its memory configuration. It has a 384-bit bus with 164.0 GB/s bandwidth, compared to the GTX 670's 256-bit bus with 192.3 GB/s bandwidth. Interestingly, the GTX 670 actually has higher bandwidth despite the narrower bus, thanks to faster GDDR5 memory running at 6 Gbps effective versus 3.4 Gbps effective. The GTX 590's 1536 MB of memory is also smaller than the GTX 670's 2 GB.

For a buyer choosing between these two end-of-life cards today, the GTX 670 is the rational pick. It wins the only head-to-head benchmark, consumes less than half the power, supports newer PCIe 3.0, and includes Vulkan support (version 1.2.175) that the GTX 590 lacks entirely. The GTX 590's higher launch MSRP of 699 USD versus 399 USD for the GTX 670 further cements the latter as the better engineering achievement.

Architecture Differences

The GTX 590 is built on the GF110 chip using Fermi 2.0 architecture, manufactured on a 40 nm process at TSMC. It packs 3,000 million transistors into a 520 mm² die, yielding a transistor density of 5.8M per mm². The GTX 670 uses the GK104 chip with Kepler architecture, also from TSMC but on a newer 28 nm process. It contains 3,540 million transistors on a much smaller 294 mm² die, achieving 12.0M transistors per mm² — more than double the density.

The compute resources differ dramatically. The GTX 590 has 512 shading units, 64 TMUs, and 48 ROPs. The GTX 670 has 1,344 shading units, 112 TMUs, and only 32 ROPs. This explains why the GTX 670 has such a massive texture rate advantage (109.8 GTexel/s versus 38.91 GTexel/s) while the GTX 590's higher ROP count helps its pixel rate remain competitive (19.46 GPixel/s versus 27.44 GPixel/s — still a win for the GTX 670).

Floating-point performance also favors the GTX 670. The GTX 590 delivers 1,244.2 GFLOPS of FP32 compute, while the GTX 670 delivers 2.634 TFLOPS — more than double. Neither card has RT cores or tensor cores, and both lack FP16 support. The GTX 590's memory clock is 854 MHz (3.4 Gbps effective), while the GTX 670 runs at 1502 MHz (6 Gbps effective). The GTX 670 also has a base clock of 915 MHz and a boost clock of 980 MHz; the GTX 590's base and boost clocks are not specified.

The GTX 590 supports DirectX 12 (11_0) and OpenGL 4.6, but has no Vulkan support. The GTX 670 also supports DirectX 12 (11_0) and OpenGL 4.6, and adds Vulkan 1.2.175. The GTX 590 uses PCIe 2.0 x16, while the GTX 670 uses PCIe 3.0 x16. Display outputs also differ: the GTX 590 has 3x DVI and 1x mini-DisplayPort, while the GTX 670 has 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2.

FAQ

Q: Which card has better OpenCL performance?

A: The GTX 670 wins decisively with a Geekbench OpenCL score of 15,341 versus the GTX 590's 12,830, a 16.4% advantage.

Q: Do both cards support Vulkan?

A: No. The GTX 670 supports Vulkan 1.2.175, while the GTX 590 has no Vulkan support listed in the data.

Q: What are the power requirements for each card?

A: The GTX 590 has a TDP of 365 W and requires a 750 W power supply, while the GTX 670 has a TDP of 170 W and requires a 450 W power supply.

Q: How do their memory configurations compare?

A: The GTX 590 has 1536 MB of GDDR5 on a 384-bit bus with 164.0 GB/s bandwidth. The GTX 670 has 2 GB of GDDR5 on a 256-bit bus with 192.3 GB/s bandwidth.

Q: Which card has more shading units?

A: The GTX 670 has 1,344 shading units, while the GTX 590 has 512 shading units.

Q: What is the transistor density difference?

A: The GTX 670 has a transistor density of 12.0M per mm² on a 294 mm² die, while the GTX 590 has 5.8M per mm² on a 520 mm² die.

Where Each One Wins

The GTX 670 wins in virtually every measurable performance category from the data. It is faster in OpenCL compute (15,341 versus 12,830), has higher texture fill rate (109.8 GTexel/s versus 38.91 GTexel/s), higher pixel rate (27.44 GPixel/s versus 19.46 GPixel/s), and more than double the FP32 throughput (2.634 TFLOPS versus 1,244.2 GFLOPS). It also has more memory (2 GB versus 1536 MB), higher memory bandwidth (192.3 GB/s versus 164.0 GB/s), and faster effective memory speed (6 Gbps versus 3.4 Gbps).

The GTX 590's advantages are limited to structural attributes. It has more ROPs (48 versus 32), a wider memory bus (384-bit versus 256-bit), and a larger die (520 mm² versus 294 mm²). It also has a higher launch MSRP of 699 USD, though this is not a performance benefit. In terms of physical dimensions, the GTX 590 is longer (279 mm versus 241 mm) and wider (40 mm versus 38 mm), while both cards are the same height at 111 mm.

For specific use cases, the GTX 670 is the clear choice for any OpenCL-based workload, and its Vulkan support makes it more future-proof for modern APIs. The GTX 590's higher ROP count might theoretically benefit certain pixel-bound tasks, but the actual pixel rate data shows the GTX 670 still wins at 27.44 GPixel/s versus 19.46 GPixel/s. The GTX 590's dual-GPU design (implied by its dual 8-pin connectors and 365 W TDP) does not translate into better benchmark results in this dataset.

Specification Differences

| Specification | GTX 590 | GTX 670 |

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

| Chip | GF110 | GK104 |

| Architecture | Fermi 2.0 | Kepler |

| Generation | GeForce 500 | GeForce 600 |

| Process Node | 40 nm | 28 nm |

| Transistors | 3,000 million | 3,540 million |

| Die Size | 520 mm² | 294 mm² |

| Transistor Density | 5.8M / mm² | 12.0M / mm² |

| Base Clock | Not specified | 915 MHz |

| Boost Clock | Not specified | 980 MHz |

| Memory Clock | 854 MHz / 3.4 Gbps effective | 1502 MHz / 6 Gbps effective |

| Memory Size | 1536 MB | 2 GB |

| Memory Bus | 384 bit | 256 bit |

| Memory Bandwidth | 164.0 GB/s | 192.3 GB/s |

| Shading Units | 512 | 1344 |

| TMUs | 64 | 112 |

| ROPs | 48 | 32 |

| Pixel Rate | 19.46 GPixel/s | 27.44 GPixel/s |

| Texture Rate | 38.91 GTexel/s | 109.8 GTexel/s |

| FP32 | 1,244.2 GFLOPS | 2.634 TFLOPS |

| TDP | 365 W | 170 W |

| Power Connectors | 2x 8-pin | 2x 6-pin |

| Suggested PSU | 750 W | 450 W |

| Bus Interface | PCIe 2.0 x16 | PCIe 3.0 x16 |

| Display Outputs | 3x DVI, 1x mini-DisplayPort | 2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2 |

| Vulkan Support | Not specified | 1.2.175 |

| Length | 279 mm (11 inches) | 241 mm (9.5 inches) |

| Width | 40 mm (1.6 inches) | 38 mm (1.5 inches) |

| Release Date | 2011-03-23 | 2012-05-09 |

| Predecessor | GeForce 400 | GeForce 500 |

| Successor | GeForce 600 | GeForce 700 |

| Launch MSRP | 699 USD | 399 USD |