AMD Radeon Vega 8 vs NVIDIA GeForce GTX 660 Comparison

# AMD Radeon Vega 8 vs NVIDIA GeForce GTX 660

The AMD Radeon Vega 8 is an integrated graphics processor built on the 14 nm GCN 5.0 architecture, while the NVIDIA GeForce GTX 660 is a dedicated dual-slot card from the Kepler generation on TSMC's 28 nm process. Benchmark averages place them nearly level — the Vega 8 scores 9221 points on average against the GTX 660's 9022, a difference of roughly 2.2%. However, the individual test results tell a more nuanced story: the Vega 8 dominates in Geekbench Metal with a 148.7% advantage, while the GTX 660 wins decisively in OpenCL and Vulkan workloads. Both cards sit at the 45th percentile among all GPUs, meaning they target the same performance tier despite their radically different designs and release dates.

The Verdict

The data presents a clear split based on workload and platform constraints. If you are building a system where power draw and physical space are primary concerns, the AMD Radeon Vega 8 is the obvious choice — it consumes 25 W compared to the GTX 660's 140 W, requires no power connectors, and occupies no expansion slot as an IGP. Its Geekbench Metal score of 10706 is exceptional for its class, and it holds its own in average performance against dedicated cards like the NVIDIA GeForce GTX 960 (9273 average, only 0.6% higher) and the GTX 465 (9294 average, 0.8% higher).

For users who need raw compute performance in OpenCL or Vulkan, the NVIDIA GeForce GTX 660 is the better pick. Its OpenCL score of 11347 beats the Vega 8's 8822 by 22.3%, and its Vulkan score of 11415 surpasses the Vega 8's 8134 by 28.7%. The GTX 660 also offers dedicated 2 GB GDDR5 memory on a 192-bit bus with 144.2 GB/s bandwidth, while the Vega 8 relies on system-shared memory with bandwidth that is system-dependent. If you are upgrading an older desktop and have a 300 W PSU or better, the GTX 660's 241 mm length and dual-slot footprint are manageable trade-offs for its superior compute throughput.

The verdict is straightforward: choose the Vega 8 for integrated efficiency and Metal-focused workflows, or the GTX 660 for dedicated memory, higher raw throughput in OpenCL/Vulkan, and legacy desktop compatibility. Neither card wins outright across all tests — the data shows a 1-2 split in favor of the GTX 660, but the Vega 8's Metal advantage is so large that platform and software priorities will ultimately decide the winner.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon Vega 8 has a higher average benchmark score at 9221, compared to the NVIDIA GeForce GTX 660's 9022. However, the difference is small — approximately 2.2% — and both cards share the same 45th percentile ranking among all GPUs.

Q: How do the two compare in Geekbench Metal performance?

A: The AMD Radeon Vega 8 scores 10706 in Geekbench Metal, while the NVIDIA GeForce GTX 660 scores 4305. This gives the Vega 8 a 148.7% advantage, making it the clear winner for Metal-accelerated applications.

Q: Which GPU performs better in Vulkan workloads?

A: The NVIDIA GeForce GTX 660 wins in Vulkan with a score of 11415, beating the AMD Radeon Vega 8's 8134 by 28.7%. This is a significant margin that favors NVIDIA for Vulkan-based gaming and compute.

Q: What is the power consumption difference?

A: The AMD Radeon Vega 8 has a TDP of 25 W and requires no power connectors, while the NVIDIA GeForce GTX 660 has a TDP of 140 W and needs one 6-pin power connector. The GTX 660 also lists a suggested PSU of 300 W.

Q: Do these GPUs support the same DirectX version?

A: No. The AMD Radeon Vega 8 supports DirectX 12 (12_1), while the NVIDIA GeForce GTX 660 supports DirectX 12 (11_0). Both support OpenGL 4.6, but the Vega 8 supports Vulkan 1.3 while the GTX 660 supports Vulkan 1.2.175.

Q: Which GPU has more shading units?

A: The NVIDIA GeForce GTX 660 has 960 shading units, compared to 512 on the AMD Radeon Vega 8. The GTX 660 also has 80 texture mapping units and 24 ROPs, versus 32 TMUs and 8 ROPs on the Vega 8.

Architecture Differences

The AMD Radeon Vega 8 is built on the GCN 5.0 architecture using the Raven chip, fabricated on a 14 nm process at GlobalFoundries. It integrates 4,940 million transistors on a 210 mm² die, yielding a transistor density of 23.5 million per square millimeter. The architecture supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.3, and features 512 shading units, 32 texture mapping units, and 8 ROPs. The Vega 8 is a Vega IGP from the Raven Ridge generation, with its predecessor being GCN 3.0 IGP and its successor being Vega II IGP. It was released in 2018 and is now end-of-life.

The NVIDIA GeForce GTX 660 uses the Kepler architecture with the GK106 chip, manufactured on a 28 nm process at TSMC. It contains 2,540 million transistors on a 221 mm² die, with a transistor density of 11.5 million per square millimeter — roughly half the density of the Vega 8. The Kepler architecture supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175, which is a slightly older Vulkan version than the Vega 8's 1.3. The GTX 660 is part of the GeForce 600 generation, succeeding the GeForce 500 and preceding the GeForce 700. It was released in 2012 and is also end-of-life.

The most significant architectural difference is the integration approach: the Vega 8 is an IGP with system-shared memory and motherboard-dependent display outputs, while the GTX 660 is a dedicated card with its own memory subsystem and fixed display outputs (2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2). The process node difference (14 nm vs 28 nm) explains the Vega 8's higher transistor density despite having fewer total transistors. The Vega 8 also supports FP16 computation at 2.253 TFLOPS with a 2:1 ratio, while the GTX 660 has no listed FP16 capability.

Specification Differences

The two GPUs differ across nearly every specification category. Clock speeds show the GTX 660 running at 980 MHz base and 1032 MHz boost, while the Vega 8 runs at 300 MHz base and 1100 MHz boost — a much wider range with a lower floor but a slightly higher ceiling. Memory configurations are fundamentally different: the Vega 8 uses system-shared memory with system-dependent bandwidth, while the GTX 660 has 2 GB of GDDR5 on a 192-bit bus with 144.2 GB/s bandwidth and a 1502 MHz memory clock (6 Gbps effective).

Compute resources favor the GTX 660 in raw counts: 960 shading units versus 512, 80 TMUs versus 32, and 24 ROPs versus 8. This translates to higher pixel and texture rates for the GTX 660 — 20.64 GPixel/s and 82.56 GTexel/s, compared to the Vega 8's 8.8 GPixel/s and 35.2 GTexel/s. FP32 performance also favors the GTX 660 at 1.981 TFLOPS versus 1,126.4 GFLOPS for the Vega 8. However, the Vega 8's TDP of 25 W is dramatically lower than the GTX 660's 140 W, and the Vega 8 needs no power connectors while the GTX 660 requires one 6-pin connector.

Physical dimensions and interface also differ: the Vega 8 is an IGP with no slot width, while the GTX 660 is dual-slot with a 241 mm (9.5 inches) length. The GTX 660 uses a PCIe 3.0 x16 bus interface, while the Vega 8's bus interface is simply "IGP." The Vega 8's display outputs are motherboard-dependent, whereas the GTX 660 offers fixed DVI, HDMI, and DisplayPort outputs. The GTX 660 has a launch MSRP of 229 USD, while the Vega 8 has no listed launch MSRP.

Head-to-Head Benchmarks

The head-to-head results show a split between APIs. In Geekbench Metal, the AMD Radeon Vega 8 delivers a score of 10706, obliterating the GTX 660's 4305. This is a 148.7% delta — the largest margin in any test — and suggests the Vega 8 is exceptionally well-optimized for Metal workloads, possibly due to its newer architecture and driver support. This single result propels the Vega 8's average score above that of the GTX 660 despite losing the other two tests.

In Geekbench OpenCL, the NVIDIA GeForce GTX 660 scores 11347, beating the Vega 8's 8822 by 22.3%. This is a substantial win that aligns with the GTX 660's higher shading unit count and dedicated memory bandwidth. The GTX 660's 960 shading units and 144.2 GB/s memory bandwidth likely contribute to its OpenCL advantage, though the exact factors are not isolated in the data.

In Geekbench Vulkan, the GTX 660 again takes the win with 11415 versus 8134, a 28.7% margin. This is the GTX 660's largest victory and indicates strong Vulkan driver maturity for the Kepler architecture, despite the Vega 8 supporting a newer Vulkan version (1.3 vs 1.2.175). The overall head-to-head tally is 1 win for the Vega 8 and 2 wins for the GTX 660.

Average scores across all benchmarks place the Vega 8 at 9221 and the GTX 660 at 9022. The nearest rivals for the Vega 8 include the AMD Radeon 890M (9210 average, 0.1% lower), the NVIDIA GeForce GTX 960 (9273 average, 0.6% higher), and the GTX 465 (9294 average, 0.8% higher). The GTX 660's nearest rivals include the NVIDIA TITAN V CEO Edition (9037 average, 0.2% higher), the GTX 560 (9058 average, 0.4% higher), and the AMD Radeon 550X (8918 average, 1.2% lower).

Where Each One Wins

The AMD Radeon Vega 8 wins decisively in Metal-based applications, as evidenced by its 148.7% benchmark advantage. This makes it the preferred choice for macOS or iOS development environments, Metal-rendering software, or any workload that leverages Apple's graphics API. The Vega 8 also wins on efficiency — its 25 W TDP is 115 W lower than the GTX 660's 140 W, making it suitable for low-power builds, thin-and-light systems, or any scenario where heat and power draw are critical constraints. Its IGP form factor means no extra slot usage, no power connectors, and motherboard-dependent outputs, which simplifies system integration. The Vega 8's newer Vulkan 1.3 support also gives it a feature-level advantage over the GTX 660's Vulkan 1.2.175.

The NVIDIA GeForce GTX 660 wins in OpenCL and Vulkan, with margins of 22.3% and 28.7% respectively. This makes it the stronger choice for general-purpose GPU compute on those APIs, including many gaming workloads that use Vulkan, OpenCL-based rendering, or scientific computing. The GTX 660's dedicated 2 GB GDDR5 memory with 144.2 GB/s bandwidth provides predictable performance that does not depend on system memory configuration, unlike the Vega 8's system-dependent bandwidth. Its higher pixel rate (20.64 GPixel/s) and texture rate (82.56 GTexel/s) suggest better raw rasterization throughput, which benefits traditional 3D rendering. The GTX 660 also has a fixed set of display outputs (2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2), which may be preferable for multi-monitor setups where motherboard outputs are limited or absent.

For a builder choosing between the two, the deciding factor is workload: Metal-centric workflows and power-sensitive builds point to the Vega 8, while OpenCL/Vulkan compute and dedicated-memory requirements point to the GTX 660. Both cards are end-of-life, so availability and driver support should be verified before purchase, but the benchmark data clearly defines their respective strengths.