NVIDIA GeForce GTX 570 vs NVIDIA GeForce GTX 670 Comparison

NVIDIA’s GeForce GTX 570 and GTX 670 are two end-of-life cards from consecutive generations, but the benchmark data clearly favors the newer Kepler design in the only direct head-to-head test available. The GTX 670 wins the sole Geekbench OpenCL comparison with a score of 15,341 against the GTX 570’s 13,515, an 11.9% margin. However, the story is less one-sided when looking at averages and percentile placement. The GTX 570 posts a higher average benchmark score of 13,515 compared to the GTX 670’s 12,773, and it sits at the 54th percentile versus the GTX 670’s 52nd. This split between raw peak performance and average consistency makes the choice dependent on workload.

Where Each One Wins

The GTX 670 is the clear winner in the only shared benchmark test, Geekbench OpenCL, where it outpaces the GTX 570 by 1,826 points. This indicates that for general-purpose compute workloads that leverage OpenCL, the newer architecture provides a tangible advantage. The GTX 670’s victory is not marginal; an 11.9% lead in this specific test suggests a meaningful architectural edge in parallel processing throughput.

Conversely, the GTX 570 wins in the aggregate. Its average benchmark score of 13,515 exceeds the GTX 670’s 12,773, and it holds a higher percentile rank against all GPUs (54th versus 52nd). This suggests that while the GTX 670 may excel in the specific OpenCL test recorded, the GTX 570 delivers more consistent performance across the broader spectrum of tasks represented by its average score. The data points to a trade-off: the GTX 670 for peak compute in a known API, the GTX 570 for overall balance.

Architecture Differences

The two cards represent distinct architectural philosophies from NVIDIA. The GTX 570 is built on the Fermi 2.0 architecture with the GF110 chip, while the GTX 670 uses the Kepler architecture with the GK104 chip. This generational shift is reflected in the manufacturing process: the GTX 570 uses a 40 nm node from TSMC, whereas the GTX 670 uses a more advanced 28 nm node, also from TSMC. The smaller process allows the GTX 670 to pack 3,540 million transistors into a 294 mm² die, achieving a transistor density of 12.0M per mm². In contrast, the GTX 570 has 3,000 million transistors on a much larger 520 mm² die, yielding a density of just 5.8M per mm².

These architectural differences lead to divergent resource allocations. The GTX 670 features 1,344 shading units, 112 texture mapping units, and 32 ROPs. The GTX 570, meanwhile, has only 480 shading units and 60 TMUs, but a higher ROP count of 40. The GTX 670 compensates for its lower ROP count with higher clock speeds, including a base clock of 915 MHz and a boost clock of 980 MHz, whereas the GTX 570’s clock data is not provided in the fact pack. The memory subsystems also differ, with the GTX 670 using a 256-bit bus and the GTX 570 using a wider 320-bit bus, though the GTX 670’s faster memory clock of 1502 MHz (6 Gbps effective) gives it higher bandwidth at 192.3 GB/s versus 152.0 GB/s for the GTX 570.

Head-to-Head Benchmarks

The only direct benchmark comparison in the data is Geekbench OpenCL, and it is a decisive win for the GTX 670. The GTX 670 scores 15,341, while the GTX 570 scores 13,515, resulting in an 11.9% advantage for the newer card. This is a substantial margin that points to the Kepler architecture’s superior compute efficiency. The GTX 670’s higher texture rate (109.8 GTexel/s versus 43.92 GTexel/s) and pixel rate (27.44 GPixel/s versus 21.96 GPixel/s) further support its dominance in throughput-heavy tasks.

The GTX 570 does not win any benchmark in the head-to-head data, but its higher average score of 13,515 compared to the GTX 670’s 12,773 suggests that the Geekbench OpenCL result may not represent the full performance picture. The GTX 570’s performance in other, unlisted tests likely pulls its average up, while the GTX 670’s average is dragged down by its weaker showing in the Metal and Vulkan tests, where it scores 7,424 and 15,553 respectively. Notably, the GTX 670’s Vulkan score of 15,553 is higher than its OpenCL score, indicating that its performance varies significantly by API.

Specification Differences

The two cards diverge on nearly every major specification. The GTX 570 uses the GF110 chip on a 40 nm process, while the GTX 670 uses the GK104 chip on a 28 nm process. The GTX 670 has a smaller die (294 mm² versus 520 mm²) but more transistors (3,540 million versus 3,000 million), leading to a much higher transistor density (12.0M / mm² versus 5.8M / mm²). The GTX 670 also has a higher memory clock (1502 MHz versus 950 MHz) and more memory (2 GB versus 1280 MB), though the GTX 570 has a wider memory bus (320-bit versus 256-bit).

In terms of compute resources, the GTX 670 leads with 1,344 shading units and 112 TMUs, while the GTX 570 has 480 shading units and 60 TMUs. The GTX 570 has more ROPs (40 versus 32). The GTX 670 achieves higher pixel and texture rates (27.44 GPixel/s and 109.8 GTexel/s versus 21.96 GPixel/s and 43.92 GTexel/s) and higher FP32 performance (2.634 TFLOPS versus 1,405.4 GFLOPS). Power consumption favors the GTX 670, with a TDP of 170 W versus 219 W for the GTX 570, and a lower suggested PSU of 450 W versus 550 W.

The GTX 670 also offers more modern connectivity and API support. It uses PCIe 3.0 x16 versus PCIe 2.0 x16, and it provides a richer display output set: 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2, compared to the GTX 570’s 2x DVI and 1x mini-HDMI 1.3a. The GTX 670 supports Vulkan 1.2.175, while the GTX 570 has no Vulkan support listed. Both cards support DirectX 12 (11_0) and OpenGL 4.6. Physical dimensions also differ, with the GTX 670 being shorter at 241 mm versus 267 mm for the GTX 570.

FAQ

Q: Which card is faster in Geekbench OpenCL?

A: The NVIDIA GeForce GTX 670 wins the Geekbench OpenCL test with a score of 15,341, compared to the GTX 570’s 13,515, giving the GTX 670 an 11.9% lead.

Q: Does the GTX 570 have a higher average benchmark score?

A: Yes, the GTX 570 has an average benchmark score of 13,515, while the GTX 670’s average is 12,773, making the GTX 570 the higher-scoring card on average.

Q: What are the power requirements for each card?

A: The GTX 670 has a TDP of 170 W and a suggested PSU of 450 W, while the GTX 570 has a TDP of 219 W and a suggested PSU of 550 W.

Q: Which card has more shading units?

A: The GTX 670 has 1,344 shading units, while the GTX 570 has 480, a significant difference in compute core count.

Q: Does the GTX 670 support Vulkan?

A: Yes, the GTX 670 supports Vulkan 1.2.175, whereas the GTX 570 does not have a Vulkan version listed in the data.

Q: How do their memory bandwidths compare?

A: The GTX 670 has higher memory bandwidth at 192.3 GB/s, despite a narrower 256-bit bus, while the GTX 570 offers 152.0 GB/s over a 320-bit bus.

The Verdict

The data supports the GTX 670 for users prioritizing raw compute performance in OpenCL, as it delivers an 11.9% higher score in that specific test. Its advantages in shading units, texture rate, pixel rate, and FP32 throughput (2.634 TFLOPS versus 1,405.4 GFLOPS) make it the superior choice for parallel processing workloads. Additionally, the GTX 670 is more power-efficient, drawing 170 W versus 219 W, and it supports modern features like PCIe 3.0 and Vulkan.

However, the GTX 570 holds its own in overall consistency. Its higher average benchmark score of 13,515 and better percentile ranking (54th versus 52nd) suggest that it may perform better in tasks not covered by the single head-to-head test. The GTX 570 also has more ROPs (40 versus 32) and a wider memory bus (320-bit versus 256-bit), which could benefit certain memory-intensive scenarios despite its lower bandwidth.

For a user who knows their workload relies on OpenCL compute, the GTX 670 is the clear pick. For someone seeking a more balanced card with a proven track record across a broader range of applications, the GTX 570’s higher average score makes it the safer choice. The GTX 670 is the technological leap forward, but the GTX 570’s aggregate numbers show it remains a formidable performer even against its successor.