AMD Radeon RX 580 vs NVIDIA GeForce GTX 670 Comparison

The AMD Radeon RX 580 and NVIDIA GeForce GTX 670 represent two distinct eras of GPU design, separated by five years of architectural evolution. The benchmark data shows a decisive victory for the newer AMD card, yet the GTX 670's legacy as a once-flagship part remains visible in its specifications. This analysis compares the two across compute workloads, architectural philosophies, and practical use cases, drawing exclusively from the provided facts.

Head-to-Head Benchmarks

The head-to-head results are unambiguous, with the AMD Radeon RX 580 winning all three benchmark tests. The most dramatic margin appears in Geekbench Metal, where the RX 580 scores 45,235 against the GTX 670's 7,424. This represents a 509.3% delta, meaning the AMD card delivers over six times the performance in Apple's Metal API. Such a gap suggests not just raw speed differences but also fundamental compatibility advantages—the GTX 670 launched years before Metal became a standard, and its Kepler architecture lacks the optimized compute paths that Polaris enjoys.

In Geekbench OpenCL, the RX 580 again dominates with a score of 37,453 versus 15,341, a 144.1% delta. The margin here is narrower than Metal but still substantial, indicating that the RX 580's 5,700 million transistors on a 14 nm process translates into roughly 2.4 times the OpenCL throughput. The Vulkan test shows a similar story: 45,173 for the RX 580 against 15,553 for the GTX 670, a 190.4% delta. Vulkan is a modern API that both cards support, but the RX 580's newer GCN 4.0 architecture with explicit support for Vulkan 1.3 clearly outperforms the GTX 670's older Kepler design, which tops out at Vulkan 1.2.175.

The overall average benchmark scores reflect this dominance, though the gap narrows considerably. The RX 580 averages 12,928 points across all tests, while the GTX 670 averages 12,773—a difference of only 155 points or roughly 1.2%. This near-parity in average scores is curious given the head-to-head blowouts. The explanation lies in the benchmark sets: the GTX 670 only has three Geekbench scores in its profile, while the RX 580 includes additional PassMark tests (DirectX 9/10/11/12, G2D, G3D, GPU Compute) and a 3DMark Steel Nomad result. Those extra tests, particularly PassMark G3D at 8,813 and GPU Compute at 3,488, pull the RX 580's average up, but the GTX 670's three strong Geekbench numbers keep it competitive in the aggregate.

Architecture Differences

The architectural chasm between these two GPUs is vast. The RX 580 uses the Polaris 20 chip built on GCN 4.0 architecture, manufactured by GlobalFoundries on a 14 nm process. It packs 5,700 million transistors into a 232 mm² die, yielding a transistor density of 24.6 million per square millimeter. In contrast, the GTX 670 uses the GK104 chip on NVIDIA's Kepler architecture, built by TSMC on a 28 nm process. It contains 3,540 million transistors spread across a larger 294 mm² die, giving a density of just 12.0 million per square millimeter. The RX 580's density advantage is nearly twofold, a direct consequence of the newer fabrication node.

Core counts tell a similar story. The RX 580 fields 2,304 shading units, 144 texture mapping units, and 32 ROPs. The GTX 670 counters with 1,344 shading units, 112 TMUs, and 32 ROPs. The RX 580 has 71% more shaders and 28% more TMUs, while both match at 32 ROPs. Clock speeds amplify this advantage: the RX 580 runs at 1257 MHz base and 1340 MHz boost, versus the GTX 670's 915 MHz base and 980 MHz boost. Higher clocks on more cores produce a multiplicative performance gap that the benchmark scores confirm.

Memory configurations diverge sharply. The RX 580 ships with 8 GB of GDDR5 on a 256-bit bus, delivering 256.0 GB/s of bandwidth at 2000 MHz (8 Gbps effective). The GTX 670 offers just 2 GB of GDDR5 on the same 256-bit bus, but at 1502 MHz (6 Gbps effective), bandwidth drops to 192.3 GB/s. The RX 580 provides four times the capacity and 33% more bandwidth. Pixel and texture rates follow suit: the RX 580 hits 42.88 GPixel/s and 193.0 GTexel/s, while the GTX 670 manages 27.44 GPixel/s and 109.8 GTexel/s. Floating-point performance shows the starkest contrast—6.175 TFLOPS FP32 for the RX 580 versus 2.634 TFLOPS for the GTX 670, a 2.34x advantage. The RX 580 even offers FP16 at 6.175 TFLOPS (1:1), while the GTX 670 has no FP16 capability listed.

The Verdict

The data leaves no room for ambiguity: the AMD Radeon RX 580 is the superior GPU in every measurable category. It wins all head-to-head benchmarks by margins ranging from 144.1% to 509.3%, offers more than double the FP32 compute throughput, and provides four times the memory capacity. The GTX 670's only comparative strengths are its lower power draw—170 W TDP versus 185 W—and its dual 6-pin power connectors versus the RX 580's single 8-pin. For a user choosing between these two today, the RX 580 is the clear pick unless power constraints are absolute.

The GTX 670's place in history is as a former high-end part, evidenced by its 399 USD launch MSRP versus the RX 580's 229 USD launch MSRP. Yet the RX 580, despite launching five years later at a lower price, outperforms it across the board. This reflects the rapid pace of GPU evolution: the RX 580's 14 nm process and GCN 4.0 architecture render the GTX 670's 28 nm Kepler design obsolete. The percentile rankings reinforce this—the RX 580 sits at the 53rd percentile of all GPUs, while the GTX 670 sits at the 52nd, a marginal difference that belies the benchmark gulf but aligns with their similar average scores.

FAQ

Q: Which GPU has a higher average benchmark score?

A: The AMD Radeon RX 580 averages 12,928 points, while the NVIDIA GeForce GTX 670 averages 12,773 points. The RX 580 leads by 155 points, or approximately 1.2%.

Q: How much faster is the RX 580 in Geekbench Metal?

A: The RX 580 scores 45,235 versus the GTX 670's 7,424, a delta of 509.3%. This is the largest performance gap between the two cards in any shared benchmark.

Q: Do both GPUs support the same DirectX version?

A: No. The RX 580 supports DirectX 12 (12_0), while the GTX 670 supports DirectX 12 (11_0). The RX 580 also supports Vulkan 1.3, whereas the GTX 670 tops out at Vulkan 1.2.175.

Q: What are the memory capacities of each card?

A: The RX 580 has 8 GB of GDDR5, while the GTX 670 has 2 GB of GDDR5. Both use a 256-bit bus, but the RX 580's bandwidth is 256.0 GB/s versus 192.3 GB/s for the GTX 670.

Q: Which card has a smaller die size?

A: The RX 580 has a 232 mm² die, while the GTX 670 has a 294 mm² die. Despite being smaller, the RX 580 packs 5,700 million transistors versus the GTX 670's 3,540 million.

Q: What is the FP32 performance difference?

A: The RX 580 delivers 6.175 TFLOPS FP32, while the GTX 670 delivers 2.634 TFLOPS. The RX 580 is 2.34 times faster in raw single-precision compute.

Where Each One Wins

The AMD Radeon RX 580 wins in nearly every scenario the data covers. For modern compute workloads—OpenCL, Vulkan, and Metal—it is the only viable choice, with deltas ranging from 144.1% to 509.3%. Its 8 GB memory capacity makes it suitable for texture-heavy applications and larger datasets, a category where the GTX 670's 2 GB would be a hard bottleneck. The RX 580's higher pixel rate (42.88 GPixel/s versus 27.44 GPixel/s) and texture rate (193.0 GTexel/s versus 109.8 GTexel/s) give it clear advantages in rasterization and filtering tasks. Its FP16 support at 1:1 ratio opens up compute paths unavailable on the GTX 670.

The NVIDIA GeForce GTX 670's wins are narrower and more circumstantial. It draws less power at 170 W versus 185 W, making it slightly more efficient in absolute terms, though the RX 580 delivers vastly more performance per watt given its scores. The GTX 670's dual 6-pin power connectors may fit older power supplies without an 8-pin cable, a minor practical advantage. Its 2x DVI outputs are legacy-friendly for older monitors, whereas the RX 580 offers 1x HDMI 2.0b and 3x DisplayPort 1.4a, which are modern but may require adapters for DVI-only displays. The GTX 670 also has a smaller physical profile at 111 mm height and 38 mm width, versus the RX 580's unspecified height and width, which could matter in tight chassis.

Specification Differences

The two cards diverge on nearly every specification. Process node: 14 nm (GlobalFoundries) for the RX 580 versus 28 nm (TSMC) for the GTX 670. Transistors: 5,700 million versus 3,540 million. Die size: 232 mm² versus 294 mm². Transistor density: 24.6M / mm² versus 12.0M / mm². Base clock: 1257 MHz versus 915 MHz. Boost clock: 1340 MHz versus 980 MHz. Memory clock: 2000 MHz (8 Gbps effective) versus 1502 MHz (6 Gbps effective). Memory size: 8 GB versus 2 GB. Memory bandwidth: 256.0 GB/s versus 192.3 GB/s. Shading units: 2304 versus 1344. TMUs: 144 versus 112. ROPs are equal at 32. Pixel rate: 42.88 GPixel/s versus 27.44 GPixel/s. Texture rate: 193.0 GTexel/s versus 109.8 GTexel/s. FP32: 6.175 TFLOPS versus 2.634 TFLOPS. FP16: 6.175 TFLOPS (1:1) versus none. TDP: 185 W versus 170 W. Power connectors: 1x 8-pin versus 2x 6-pin. Display outputs: 1x HDMI 2.0b, 3x DisplayPort 1.4a versus 2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2. DirectX: 12 (12_0) versus 12 (11_0). Vulkan: 1.3 versus 1.2.175. Release date: 2017-04-17 versus 2012-05-09. Launch MSRP: 229 USD versus 399 USD. The RX 580's length matches the GTX 670 at 241 mm (9.5 inches), but the GTX 670 adds height (111 mm) and width (38 mm) dimensions the RX 580 does not list.