AMD Radeon Vega 11 vs NVIDIA GeForce GTX 670 Comparison

Head-to-Head Benchmarks

The recorded data presents a split decision between these two end-of-life graphics solutions. The AMD Radeon Vega 11 wins one benchmark decisively, while the NVIDIA GeForce GTX 670 takes the other two by a clear margin.

In the Geekbench Metal test, the AMD Radeon Vega 11 scores 17,392 points against the NVIDIA GeForce GTX 670’s 7,424 points. That is a 134.3% advantage for the AMD part, the largest delta in any of the three recorded tests. This result is striking because Metal is a low-level API that typically favors discrete GPUs with dedicated memory, yet the integrated Vega 11, which relies on system shared memory, outperforms the older Kepler-based card by more than double. The data suggests the Vega 11’s newer architecture and higher API feature level contribute heavily to this outcome, as the GTX 670’s DirectX support is capped at 12 (11_0) while Vega 11 reaches 12 (12_1).

The NVIDIA GeForce GTX 670 counters in the OpenCL test. It posts 15,341 points versus the AMD Radeon Vega 11’s 13,392 points, a 12.7% lead for NVIDIA. This is a meaningful reversal from the Metal result, indicating that the GTX 670’s raw compute resources, 1,344 shading units versus 704, and its dedicated 192.3 GB/s memory bandwidth, give it a structural edge in OpenCL workloads that do not heavily depend on API-level optimizations. The GTX 670’s higher FP32 throughput, 2.634 TFLOPS versus 1.971 TFLOPS, aligns with this outcome.

The Vulkan benchmark widens NVIDIA’s lead further. The GTX 670 scores 15,553 points, while the Vega 11 manages 12,273 points. That is a 21.1% margin for the NVIDIA card. Vulkan is a modern low-overhead API, and the GTX 670’s advantage here suggests that its dedicated memory subsystem and higher texture rate, 109.8 GTexel/s versus 61.60 GTexel/s, matter more than architectural age in this particular test. The Vega 11’s Vulkan support is version 1.3, newer than the GTX 670’s 1.2.175, yet the performance data does not reflect a benefit from that newer API revision in this benchmark.

Looking at average scores across all tests, the AMD Radeon Vega 11 lands at 14,352 points, which places it in the 56th percentile of all GPUs in the database. The NVIDIA GeForce GTX 670 averages 12,773 points, sitting in the 52nd percentile. The delta between their averages is roughly 1,579 points, or about 12.4% in favor of AMD, driven almost entirely by the outsized Metal result. The GTX 670’s nearest rivals include the NVIDIA GeForce GTX 590 and AMD Radeon Pro 455, both within 0.4% of its average score, while the Vega 11’s closest competitor is the NVIDIA GeForce GTX TITAN, which trails it by only 0.1%.

Architecture Differences

The two GPUs come from fundamentally different design philosophies and fabrication eras. The AMD Radeon Vega 11 uses the Picasso chip, built on a 12 nm process at GlobalFoundries, with a die size of 210 mm² housing 4,940 million transistors. That yields a transistor density of 23.5M per mm². The NVIDIA GeForce GTX 670 uses the GK104 chip, fabricated by TSMC on a 28 nm process, with a larger die of 294 mm² but fewer transistors at 3,540 million, resulting in a density of only 12.0M per mm². The process node advantage is clear: AMD packs nearly twice the transistor density into a smaller die.

The Vega 11 is an integrated graphics processor, meaning it shares system memory rather than having its own VRAM. Its memory size, type, bus width, and bandwidth are all listed as “System Shared” or “System Dependent,” which makes its performance highly contingent on the host platform’s RAM configuration. The GTX 670, by contrast, is a discrete card with 2 GB of GDDR5 memory on a 256-bit bus, delivering a fixed 192.3 GB/s bandwidth. This dedicated memory path is a structural advantage in bandwidth-sensitive workloads, as reflected in its OpenCL and Vulkan wins.

Compute resource counts differ substantially. The Vega 11 has 704 shading units, 44 texture mapping units, and 8 raster output pipelines. The GTX 670 has 1,344 shading units, 112 TMUs, and 32 ROPs. These numbers explain the GTX 670’s higher pixel rate of 27.44 GPixel/s versus 11.20 GPixel/s, and its texture rate of 109.8 GTexel/s versus 61.60 GTexel/s. The GTX 670 also leads in FP32 throughput at 2.634 TFLOPS, while the Vega 11 manages 1.971 TFLOPS. The Vega 11 does support FP16 at 3.942 TFLOPS with a 2:1 ratio, a feature the GTX 670 lacks entirely, as its FP16 field is null.

Clock speeds tell another story. The Vega 11 operates at a 300 MHz base clock, boosting to 1400 MHz, which is typical for an integrated part with a 15 W TDP. The GTX 670 runs at a 915 MHz base and 980 MHz boost, with a much higher 170 W TDP, requiring a dual-slot cooler and 2x 6-pin power connectors, plus a suggested PSU of 450 W. The Vega 11 needs no power connectors and consumes a fraction of the power, reflecting its IGP nature.

API support favors the newer AMD chip. The Vega 11 lists DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.3. The GTX 670 lists DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. The DirectX feature level difference is notable: 12_1 on Vega 11 versus 11_0 on GTX 670, which can affect certain modern rendering features. The GTX 670’s release date is May 2012, while the Vega 11 launched in September 2019, a gap of over seven years that explains the API and process node differences.

The Verdict

The benchmark data does not crown a single winner across all tests. The AMD Radeon Vega 11 wins the Metal benchmark by a massive 134.3% margin, making it the clear choice for applications that rely heavily on Apple’s Metal API, such as macOS-based creative suites. Its average score of 14,352 also sits higher in the percentile ranking, 56th versus the GTX 670’s 52nd, indicating better overall performance across the recorded tests.

The NVIDIA GeForce GTX 670 wins the OpenCL and Vulkan tests, with leads of 12.7% and 21.1% respectively. For users running Vulkan-based games or OpenCL compute workloads, the GTX 670 is the stronger option, despite its older architecture. Its dedicated 2 GB GDDR5 memory and 192.3 GB/s bandwidth provide a stable foundation that the Vega 11’s system-shared memory cannot match in these specific tests.

The GTX 670 also carries a launch MSRP of 399 USD, which the database records, but no such figure exists for the Vega 11. The Vega 11’s integrated nature means it adds no discrete card cost, but its performance is system-dependent, so real-world results can vary based on the host platform’s memory speed and capacity. The GTX 670 offers predictable, consistent performance thanks to its fixed memory configuration.

For a user prioritizing modern API support and Metal-centric workloads, the AMD Radeon Vega 11 is the data-backed choice. For Vulkan gaming and OpenCL compute, the NVIDIA GeForce GTX 670 holds the advantage. Neither card is currently in production, as both are marked end-of-life, so availability is limited to used markets or existing systems.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon Vega 11 averages 14,352 points across all recorded tests, while the NVIDIA GeForce GTX 670 averages 12,773 points. The Vega 11 also ranks higher in the database percentile, at 56th versus 52nd for the GTX 670.

Q: Why does the AMD Radeon Vega 11 win the Metal benchmark by such a large margin?

A: The Vega 11 scores 17,392 in Geekbench Metal versus the GTX 670’s 7,424, a 134.3% difference. The Vega 11’s newer GCN 5.0 architecture and DirectX 12 (12_1) support likely contribute to better API-level performance, as its raw compute resources are lower than the GTX 670’s.

Q: Does the NVIDIA GeForce GTX 670 have more memory bandwidth?

A: Yes. The GTX 670 has 2 GB of GDDR5 memory on a 256-bit bus, providing 192.3 GB/s bandwidth. The Vega 11 uses system shared memory, with bandwidth listed as system dependent, so it has no fixed memory bandwidth figure.

Q: Which GPU is more power-efficient according to the data?

A: The AMD Radeon Vega 11 has a 15 W TDP and no power connectors, while the NVIDIA GeForce GTX 670 has a 170 W TDP and requires 2x 6-pin power connectors plus a 450 W suggested PSU. The Vega 11 is an integrated part, so it draws far less power.

Q: What are the Vulkan performance differences?

A: The GTX 670 scores 15,553 in Geekbench Vulkan, beating the Vega 11’s 12,273 by 21.1%. The GTX 670 also has higher FP32 throughput at 2.634 TFLOPS versus 1.971 TFLOPS, which aligns with its Vulkan lead.

Q: Do both GPUs support modern APIs?

A: Both support OpenGL 4.6. The Vega 11 supports DirectX 12 (12_1) and Vulkan 1.3, while the GTX 670 supports DirectX 12 (11_0) and Vulkan 1.2.175. The Vega 11 has a newer feature level in DirectX, but the GTX 670 still performs better in the recorded Vulkan test.

Where Each One Wins

AMD Radeon Vega 11: The Metal benchmark is its territory, with a 134.3% lead over the GTX 670. This makes it the better choice for any workload that leverages Apple’s Metal API, which is common in macOS-based video editing, 3D rendering, and creative applications. Its higher average score of 14,352 and 56th percentile ranking also suggest it is the more balanced performer overall across the tested workloads. The Vega 11’s FP16 support at 3.942 TFLOPS, which the GTX 670 lacks entirely, could benefit compute tasks that use half-precision arithmetic. Its 12 nm process and 23.5M transistor density per mm² indicate a more modern design that is better suited to API-feature-heavy environments.

NVIDIA GeForce GTX 670: Vulkan and OpenCL are its strong suits. It wins Vulkan by 21.1% and OpenCL by 12.7%, making it the preferred option for Vulkan-based games and OpenCL compute workloads such as physics simulations or scientific calculations. Its dedicated 192.3 GB/s bandwidth and 2 GB GDDR5 memory provide consistent, predictable performance, unlike the Vega 11’s system-dependent memory. The GTX 670’s higher shading unit count (1,344 versus 704), texture rate (109.8 GTexel/s versus 61.60 GTexel/s), and pixel rate (27.44 GPixel/s versus 11.20 GPixel/s) give it a raw throughput advantage that pays off in these lower-level API tests. Its 52nd percentile ranking, while lower than the Vega 11’s, still places it in the mid-range of all GPUs in the database.