AMD Radeon R9 M390X vs NVIDIA GeForce GTX 780 Comparison

AMD Radeon R9 M390X and NVIDIA GeForce GTX 780 are two end-of-life graphics cards from different eras, built for very different purposes. The M390X is a mobile-first MXM module from 2015, while the GTX 780 is a 2013 desktop powerhouse. Benchmark data shows a split decision: the AMD card dominates in Metal workloads, while the NVIDIA card takes a clear lead in Vulkan. Their average scores are close, but the way they achieve them tells a story about architectural priorities and software optimization.

Where Each One Wins

The AMD Radeon R9 M390X is the clear winner in Apple-centric Metal workloads. Its Geekbench Metal score of 24,373 completely dwarfs the GTX 780's 10,114, representing a 141% advantage. This is not a marginal lead; it is a decisive stomping. For anyone running Metal-based applications, the M390X is the only rational choice between these two.

The NVIDIA GeForce GTX 780 takes the crown in Vulkan performance. It scores 24,514 versus the M390X's 16,951, a 30.9% lead. This shows NVIDIA's older Kepler architecture still holds its own in modern cross-platform graphics APIs, likely due to better driver maturity for Vulkan. If the workload is Vulkan-centric, the GTX 780 is the better performer.

Looking at average benchmark scores across all tests, the M390X edges ahead with 20,662 versus the GTX 780's 19,164. That is a 7.8% difference in average performance. However, the head-to-head wins are split evenly at one each. The practical takeaway: the M390X is for Metal, the GTX 780 is for Vulkan, and for anything else, the M390X has a slight overall edge.

Architecture Differences

The two cards come from fundamentally different design philosophies. The AMD Radeon R9 M390X uses the Amethyst chip based on GCN 3.0 architecture, built on a 28 nm process at TSMC. It packs 5,000 million transistors on a 366 mm² die, resulting in a transistor density of 13.7M per mm². The GTX 780 uses the GK110 chip with Kepler architecture, also on TSMC's 28 nm node, but with 7,080 million transistors on a much larger 561 mm² die, giving a lower density of 12.6M per mm².

Memory configurations differ significantly. The M390X has 4 GB of GDDR5 on a 256-bit bus, delivering 160.0 GB/s of bandwidth. The GTX 780 has 3 GB of GDDR5 on a wider 384-bit bus, achieving 288.4 GB/s. That is an 80% bandwidth advantage for NVIDIA, which matters in high-resolution textures and compute-heavy scenes.

Compute resources are also distributed differently. The M390X has 2,048 shading units, 128 TMUs, and 32 ROPs. The GTX 780 has more of everything: 2,304 shading units, 192 TMUs, and 48 ROPs. This explains the GTX 780's higher peak rates of 43.30 GPixel/s and 173.2 GTexel/s, versus the M390X's 23.14 GPixel/s and 92.54 GTexel/s. In raw throughput, the GTX 780 is roughly 87% faster in pixel fill and 87% faster in texture fill.

Power and physical specs are polar opposites. The M390X draws just 75 W and needs no power connectors, fitting an MXM Module form factor. The GTX 780 demands 250 W, requires 1x 6-pin plus 1x 8-pin power connectors, and is a dual-slot card measuring 267 mm in length, 111 mm in height, and 38 mm in width. The suggested PSU for the GTX 780 is 600 W, while the M390X has no such requirement. API support shows the M390X supporting DirectX 12 (12_0), while the GTX 780 only reaches DirectX 12 (11_0). Both support OpenGL 4.6, but the GTX 780 has a slightly newer Vulkan version at 1.2.175 versus 1.2.170.

Head-to-Head Benchmarks

The Geekbench Metal test is the single biggest differentiator. The M390X scores 24,373, which is 141% higher than the GTX 780's 10,114. This is a massive gap that cannot be explained by clock speeds alone. The M390X's memory runs at 1250 MHz (5 Gbps effective), while the GTX 780's memory runs at 1502 MHz (6 Gbps effective). The M390X has no base or boost clock listed, while the GTX 780 operates at 863 MHz base and 902 MHz boost. Despite the GTX 780's higher memory clock and wider bus, the M390X absolutely crushes it in Metal. This strongly suggests that AMD's GCN architecture has much better Metal optimization, likely due to its presence in Apple systems.

In Vulkan, the tables turn completely. The GTX 780 scores 24,514, which is 30.9% ahead of the M390X's 16,951. This is a substantial victory for NVIDIA, and it shows that Kepler can still be competitive in modern APIs when drivers are well-tuned. The GTX 780's higher compute throughput (4.156 TFLOPS FP32 versus 2.961 TFLOPS for the M390X) likely contributes to this win, as Vulkan tends to scale with raw compute resources.

The average benchmark scores put the M390X at 20,662, which is 7.8% higher than the GTX 780's 19,164. This places the M390X in the 66th percentile of all GPUs, while the GTX 780 sits at the 64th percentile. Both cards are surrounded by similar performers: the M390X is 0.2% behind the AMD Radeon RX 5600 XT and 0.4% ahead of the Intel Arc A750. The GTX 780 is 0.1% behind the NVIDIA TITAN Xp and 0.4% ahead of the NVIDIA Tesla K20m. These are tight groupings, showing that both cards are firmly in the mid-range of historical GPU performance.

FAQ

Q: Which card is better for Metal-based applications?

A: The AMD Radeon R9 M390X is overwhelmingly better, scoring 24,373 in Geekbench Metal versus the GTX 780's 10,114. That is a 141% advantage, making the M390X the only sensible option for Metal workloads.

Q: Which card performs better in Vulkan?

A: The NVIDIA GeForce GTX 780 wins in Vulkan, scoring 24,514 versus the M390X's 16,951. This is a 30.9% lead for NVIDIA, showing Kepler's strength in this API.

Q: How do their average benchmark scores compare?

A: The M390X has a higher average benchmark score of 20,662, which is 7.8% above the GTX 780's 19,164. The M390X also ranks slightly higher in the percentile at 66 versus 64.

Q: What are the memory bandwidth differences?

A: The GTX 780 has significantly more bandwidth at 288.4 GB/s, thanks to a 384-bit bus, while the M390X offers 160.0 GB/s on a 256-bit bus. The GTX 780's bandwidth is 80% higher.

Q: Which card has more compute units?

A: The GTX 780 has more shading units (2,304 versus 2,048), more TMUs (192 versus 128), and more ROPs (48 versus 32). It also has higher FP32 performance at 4.156 TFLOPS versus 2.961 TFLOPS.

Q: What are the power requirements for each?

A: The M390X is a low-power MXM module drawing 75 W with no power connectors. The GTX 780 is a dual-slot desktop card drawing 250 W, requiring 1x 6-pin and 1x 8-pin connectors and a 600 W suggested PSU.

The Verdict

The data presents a clear split based on use case. If you are working in a Metal environment, the AMD Radeon R9 M390X is the definitive choice. Its 141% lead in Metal benchmarks is so large that no other metric matters for that scenario. It also has a higher average benchmark score and better percentile ranking, making it the more balanced overall performer.

If your workloads are Vulkan-based, the NVIDIA GeForce GTX 780 is the better card. Its 30.9% advantage in Vulkan is substantial, and its superior raw compute throughput (4.156 TFLOPS versus 2.961 TFLOPS) and memory bandwidth (288.4 GB/s versus 160.0 GB/s) give it an edge in compute-heavy tasks. However, its 250 W power draw and dual-slot size make it a much more demanding piece of hardware.

The M390X is the practical choice for mobile or compact systems, given its 75 W power draw and MXM form factor. The GTX 780 is a desktop-only card that needs a 600 W PSU and significant case space. For mixed workloads, the M390X's higher average score and better Metal support tip the balance, but the GTX 780 remains a viable option for Vulkan enthusiasts who do not mind the power footprint.

Specification Differences

| Specification | AMD Radeon R9 M390X | NVIDIA GeForce GTX 780 |

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

| Chip | Amethyst | GK110 |

| Architecture | GCN 3.0 | Kepler |

| Process Node | 28 nm | 28 nm |

| Transistors | 5,000 million | 7,080 million |

| Die Size | 366 mm² | 561 mm² |

| Transistor Density | 13.7M / mm² | 12.6M / mm² |

| Base Clock | Not specified | 863 MHz |

| Boost Clock | Not specified | 902 MHz |

| Memory Clock | 1250 MHz (5 Gbps effective) | 1502 MHz (6 Gbps effective) |

| Memory Size | 4 GB | 3 GB |

| Memory Type | GDDR5 | GDDR5 |

| Memory Bus Width | 256 bit | 384 bit |

| Memory Bandwidth | 160.0 GB/s | 288.4 GB/s |

| Shading Units | 2048 | 2304 |

| TMUs | 128 | 192 |

| ROPs | 32 | 48 |

| Pixel Rate | 23.14 GPixel/s | 43.30 GPixel/s |

| Texture Rate | 92.54 GTexel/s | 173.2 GTexel/s |

| FP32 Performance | 2.961 TFLOPS | 4.156 TFLOPS |

| FP16 Performance | 2.961 TFLOPS (1:1) | Not specified |

| TDP | 75 W | 250 W |

| Slot Width | MXM Module | Dual-slot |

| Power Connectors | None | 1x 6-pin + 1x 8-pin |

| Suggested PSU | Not specified | 600 W |

| Display Outputs | Portable Device Dependent | 2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2 |

| DirectX Support | 12 (12_0) | 12 (11_0) |

| OpenGL Support | 4.6 | 4.6 |

| Vulkan Support | 1.2.170 | 1.2.175 |

| Dimensions | Not specified | 267 mm x 111 mm x 38 mm |

| Release Date | 2015-05-04 | 2013-05-22 |

| Launch MSRP | Not specified | 649 USD |