Intel Arc A770M vs NVIDIA GeForce RTX 3070 Mobile Comparison

Where Each One Wins

The benchmark data paints a clear split between these two mobile GPUs. The NVIDIA GeForce RTX 3070 Mobile takes 7 of the 10 head-to-head tests, while the Intel Arc A770M wins 3. However, the distribution of those wins tells a more interesting story than the raw count.

The NVIDIA GPU dominates in modern, compute-heavy workloads. Its largest victories come in PassMark G3D (30% ahead), PassMark GPU Compute (42.9% ahead), and the DirectX 10 and 11 legacy tests (101.8% and 100% ahead, respectively). It also wins in the 3DMark Steel Nomad DX12 test (4.5% ahead), Geekbench OpenCL (3.8% ahead), and Geekbench Vulkan (16.6% ahead). This pattern suggests the RTX 3070 Mobile is the stronger choice for general 3D rendering, compute tasks, and Vulkan-based applications.

The Intel Arc A770M counters in three specific areas: PassMark DirectX 12 (8.6% ahead), PassMark DirectX 9 (10.1% ahead), and PassMark G2D (9.8% ahead). The G2D win is notable because it measures 2D graphics performance, a category where Intel's architecture shows unexpected strength. The DirectX 12 and 9 wins indicate that Intel's driver and hardware implementation handles those specific API paths better than NVIDIA's, even though NVIDIA wins in most other categories.

The average benchmark scores confirm the NVIDIA GPU's overall edge: 20,534 versus 18,383 for Intel, a difference of roughly 11.7%. The percentile rankings also favor NVIDIA, sitting at the 65th percentile of all GPUs compared to Intel's 62nd. This is not a landslide victory, but it is a consistent one across most test categories.

Architecture Differences

The architectural divide between these two GPUs is substantial. NVIDIA's GA104 chip uses the Ampere architecture, fabricated on Samsung's 8 nm process. The die measures 392 mm² and contains 17,400 million transistors, resulting in a transistor density of 44.4 million per square millimeter. Intel's DG2-512 chip uses the Xe-HPG architecture, built on TSMC's 6 nm process. Its die is slightly larger at 406 mm², but it packs significantly more transistors: 21,700 million, for a density of 53.4 million per square millimeter.

The core configurations differ fundamentally. The RTX 3070 Mobile has 5,120 shading units, 160 texture mapping units, and 80 raster operation units. It also includes 40 ray tracing cores and 160 tensor cores, the latter being essential for AI-accelerated features like DLSS. The Arc A770M has fewer shading units at 4,096, but compensates with more TMUs (256) and more ROPs (128). It has 32 ray tracing cores and, notably, no tensor cores listed in the database. This means Intel relies on other methods for AI acceleration, while NVIDIA has dedicated hardware.

Clock speeds favor Intel dramatically. The Arc A770M runs at a base clock of 1650 MHz and boosts to 2050 MHz. The RTX 3070 Mobile starts at 1110 MHz and boosts to 1560 MHz. This explains why Intel can achieve competitive or superior performance in certain tests despite having fewer shading units. The higher clocks also drive Intel's pixel rate to 262.4 GPixel/s and texture rate to 524.8 GTexel/s, both far exceeding NVIDIA's 124.8 GPixel/s and 249.6 GTexel/s.

Memory configurations also diverge. Both use GDDR6 on a 256-bit bus, but Intel offers 16 GB versus NVIDIA's 8 GB. Intel also has higher memory bandwidth at 512.0 GB/s compared to 448.0 GB/s, driven by a 2000 MHz memory clock versus 1750 MHz. For texture-heavy or large-dataset workloads, the extra capacity and bandwidth could matter.

The process node advantage is clear: TSMC's 6 nm is more advanced than Samsung's 8 nm, allowing Intel to pack more transistors into a similar die area. Yet NVIDIA achieves comparable or better overall performance with an older process, which speaks to architectural efficiency. Power draw is close: 115 W for NVIDIA versus 120 W for Intel, a negligible difference in mobile terms.

Head-to-Head Benchmarks

The most decisive NVIDIA victory comes in PassMark DirectX 10, where the RTX 3070 Mobile scores 113 against Intel's 56, a 101.8% advantage. The DirectX 11 test shows a similar pattern: 138 versus 69, a 100% lead. These are legacy APIs, but the margin is so large that it suggests fundamental differences in how each GPU handles older instruction paths.

PassMark G3D delivers the next biggest win for NVIDIA: 15,309 versus 11,774, a 30% gap. This is the aggregate 3D performance metric and arguably the most representative of general gaming capability. PassMark GPU Compute follows with 6,827 versus 4,778, a 42.9% lead, indicating NVIDIA's advantage in non-graphics compute workloads.

Geekbench Vulkan shows a 16.6% NVIDIA win (86,768 versus 74,422), while Geekbench OpenCL is closer at 3.8% (92,939 versus 89,494). The 3DMark Steel Nomad DX12 test is tightest of all NVIDIA wins: 2,380 versus 2,278, just 4.5% apart. This suggests that in modern, DirectX 12-based workloads, the two GPUs are nearly equivalent.

Intel's wins are narrower but consistent. PassMark DirectX 12 shows 70 versus 64, an 8.6% Intel lead. PassMark DirectX 9 shows 178 versus 160, a 10.1% lead. The G2D test is closest: 711 versus 641, a 9.8% lead. These wins indicate Intel's strengths in specific API paths and 2D operations, but they do not offset NVIDIA's larger margins in more demanding tests.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The NVIDIA GeForce RTX 3070 Mobile has an average benchmark score of 20,534, while the Intel Arc A770M scores 18,383. This places NVIDIA at the 65th percentile of all GPUs and Intel at the 62nd.

Q: How do the two GPUs compare in DirectX 12 performance?

A: The Intel Arc A770M wins the PassMark DirectX 12 test with a score of 70 versus NVIDIA's 64, an 8.6% advantage. However, in the 3DMark Steel Nomad DX12 test, NVIDIA wins with 2,380 versus 2,278, a 4.5% lead. The results depend on the specific benchmark.

Q: Which GPU has more memory and bandwidth?

A: The Intel Arc A770M has 16 GB of GDDR6 memory with 512.0 GB/s bandwidth. The NVIDIA GeForce RTX 3070 Mobile has 8 GB with 448.0 GB/s. Both use a 256-bit bus.

Q: What are the key architectural differences in ray tracing?

A: The RTX 3070 Mobile has 40 ray tracing cores and 160 tensor cores. The Arc A770M has 32 ray tracing cores and no tensor cores listed in the database.

Q: Does the Intel GPU have higher clock speeds?

A: Yes. The Arc A770M has a base clock of 1650 MHz and boost of 2050 MHz. The RTX 3070 Mobile operates at 1110 MHz base and 1560 MHz boost.

Q: Which GPU wins in compute workloads?

A: The NVIDIA GPU wins PassMark GPU Compute with 6,827 versus 4,778, a 42.9% lead. It also wins Geekbench OpenCL with 92,939 versus 89,494, a 3.8% margin.

The Verdict

The data points to a clear but nuanced conclusion. For most gaming and compute scenarios, the NVIDIA GeForce RTX 3070 Mobile is the stronger GPU. Its 30% lead in PassMark G3D and 42.9% lead in GPU compute are decisive. The 16.6% Vulkan advantage and 4.5% 3DMark Steel Nomad win reinforce this, showing NVIDIA's edge extends to modern APIs.

Users who prioritize legacy DirectX 10 or 11 games should strongly favor the RTX 3070 Mobile, as its performance there is roughly double that of the Intel Arc A770M. The same applies to AI-accelerated workloads, given NVIDIA's 160 tensor cores versus Intel's lack of dedicated tensor hardware.

The Intel Arc A770M is the better choice for specific scenarios. Its 16 GB memory capacity and 512.0 GB/s bandwidth make it more suitable for memory-hungry applications. The 8.6% DirectX 12 win and 10.1% DirectX 9 win show Intel's strength in those API paths. The 9.8% G2D lead suggests better 2D performance, which could matter for desktop or productivity tasks.

The overall average score difference of 11.7% in NVIDIA's favor is significant but not overwhelming. The RTX 3070 Mobile sits at the 65th percentile versus Intel's 62nd, a modest gap. For balanced workloads, NVIDIA is the safer pick. For specific API preferences or memory-intensive tasks, Intel's offering deserves consideration.

Specification Differences

| Specification | NVIDIA GeForce RTX 3070 Mobile | Intel Arc A770M |

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

| Architecture | Ampere | Xe-HPG |

| Process node | 8 nm (Samsung) | 6 nm (TSMC) |

| Transistors | 17,400 million | 21,700 million |

| Die size | 392 mm² | 406 mm² |

| Transistor density | 44.4M / mm² | 53.4M / mm² |

| Base clock | 1110 MHz | 1650 MHz |

| Boost clock | 1560 MHz | 2050 MHz |

| Memory clock | 1750 MHz (14 Gbps effective) | 2000 MHz (16 Gbps effective) |

| Memory size | 8 GB GDDR6 | 16 GB GDDR6 |

| Memory bandwidth | 448.0 GB/s | 512.0 GB/s |

| Shading units | 5120 | 4096 |

| TMUs | 160 | 256 |

| ROPs | 80 | 128 |

| Ray tracing cores | 40 | 32 |

| Tensor cores | 160 | None |

| Pixel rate | 124.8 GPixel/s | 262.4 GPixel/s |

| Texture rate | 249.6 GTexel/s | 524.8 GTexel/s |

| FP32 | 15.97 TFLOPS | 16.79 TFLOPS |

| FP16 | 15.97 TFLOPS (1:1) | 33.59 TFLOPS (2:1) |

| TDP | 115 W | 120 W |

| Slot width | Not specified | IGP |

| Release date | 2021-01-11 | Not specified |