AMD Radeon Vega 3 vs NVIDIA GeForce 920MX Comparison

Head-to-Head Benchmarks

The recorded data shows a split decision between the AMD Radeon Vega 3 and the NVIDIA GeForce 920MX. Each part claims one benchmark win, and the margins are telling. In Geekbench OpenCL, the NVIDIA GeForce 920MX posts a score of 4068 against 3963 for the AMD Radeon Vega 3. That works out to a 2.6% lead for NVIDIA. It is a narrow margin, well within the kind of run-to-run variation that typifies mobile graphics testing.

The Geekbench Vulkan result flips the script decisively. The AMD Radeon Vega 3 scores 3961, while the NVIDIA GeForce 920MX manages only 2987. That is a 32.6% advantage for the Vega 3. Where the two parts were nearly inseparable in OpenCL, the Vulkan gap is enormous. The data indicates that the AMD IGP is far more comfortable with modern low-level graphics APIs, while the older Maxwell part loses significant ground.

Looking at the broader benchmark averages, the AMD Radeon Vega 3 holds an average score of 4268 across its three recorded tests (Geekbench Metal, OpenCL, and Vulkan). The NVIDIA GeForce 920MX averages 3528 across just its two tests. The difference in average score is about 740 points, roughly 21% in favor of the AMD part. However, this average is flattered by the fact that the NVIDIA chip lacks a Metal result, and its Vulkan number drags its mean down considerably.

The nearest rival data for each part places them in different performance tiers. The AMD Radeon Vega 3 sits at the 25th percentile of all GPUs, with its closest competitor being the NVIDIA GeForce GTX 460M at 4282 average score, a delta of only -0.3%. The NVIDIA GeForce 920MX sits lower at the 21st percentile, with the NVIDIA GeForce GT 545 as its nearest match at 3594 average score, a delta of -1.8%. The percentile gap of four points reinforces that the AMD part occupies a slightly higher standing in the overall performance distribution.

Where Each One Wins

The NVIDIA GeForce 920MX wins the OpenCL compute workload, but only by a hair. Its 4068 score versus 3963 represents a 2.6% advantage. For applications that rely on OpenCL acceleration, the two parts are effectively interchangeable in practice. The margin is so small that driver version, thermal state, or system memory configuration could easily flip the result.

The AMD Radeon Vega 3 wins Vulkan by a landslide. Its 3961 score against 2987 is the single largest head-to-head delta in the entire comparison. Vulkan is becoming the standard for modern game engines and compute workloads, so this win carries outsized weight. The 32.6% advantage suggests that the GCN 5.0 architecture handles Vulkan's explicit command buffers and multi-threaded rendering far better than Maxwell.

In terms of raw throughput metrics from the specification data, the NVIDIA part holds leads in several theoretical peak rates. The 920MX offers 508.4 GFLOPS of FP32 compute against 422.4 GFLOPS for the Vega 3, a 20% advantage in raw floating-point throughput. Its pixel rate is 7.944 GPixel/s versus 4.400 GPixel/s, a 80% advantage in fillrate. Texture rate also favors NVIDIA at 23.83 GTexel/s versus 13.20 GTexel/s, a 80% margin. Yet these theoretical advantages did not translate into a decisive OpenCL win, and they were completely inverted in Vulkan.

The AMD part wins on memory flexibility. Its memory is system shared, meaning it can draw on whatever system RAM is available, with bandwidth listed as system dependent. The NVIDIA part is locked to 2 GB of DDR3 on a 64-bit bus with 14.40 GB/s of fixed bandwidth. For workloads that exceed 2 GB of video memory, the Vega 3 has a structural advantage, since it can allocate far more memory if the host system has enough RAM.

Architecture Differences

The two parts come from entirely different design lineages. The AMD Radeon Vega 3 uses the Picasso chip built on GCN 5.0 architecture, manufactured on a 12 nm process at GlobalFoundries. The NVIDIA GeForce 920MX uses the GM108S chip built on the Maxwell architecture, manufactured on a 28 nm process at TSMC. The process node difference is substantial: 12 nm versus 28 nm. This explains why the AMD chip packs 4,940 million transistors into a 210 mm² die, while the NVIDIA chip fits 1,020 million transistors into 77 mm². Transistor density tells the story: AMD achieves 23.5M transistors per mm², NVIDIA only 13.2M.

The execution resources differ in count and ratio. The AMD part has 192 shading units, 12 texture mapping units, and 4 render output units. The NVIDIA part has 256 shading units, 24 TMUs, and 8 ROPs. NVIDIA doubles the TMU and ROP counts and offers 33% more shaders. Clock speeds partially compensate: the AMD part runs at a 300 MHz base and 1100 MHz boost, while the NVIDIA part runs at 965 MHz base and 993 MHz boost. The AMD boost clock is about 11% higher, but its base clock is far lower, indicating a power-adaptive design that can scale down aggressively when idle.

Memory architecture is fundamentally different. The AMD Vega 3 has no dedicated video memory; it shares system memory entirely, with bus width and bandwidth listed as system dependent. The NVIDIA 920MX has 2 GB of dedicated DDR3 on a 64-bit bus delivering 14.40 GB/s. Dedicated memory provides predictable latency and bandwidth, while shared memory provides flexibility and capacity. The NVIDIA memory clock is listed at 900 MHz with 1800 Mbps effective data rate.

API support also differs. The AMD part supports DirectX 12 (12_1), OpenGL 4.6, and Vulkan 1.3. The NVIDIA part supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.4. The higher DirectX feature level on the AMD side (12_1 versus 11_0) explains some of its Vulkan advantage. The NVIDIA part has a newer Vulkan runtime version (1.4 versus 1.3), but that did not help its Vulkan score.

Power draw is nearly identical: 15 W for the AMD IGP, 16 W for the NVIDIA MXM module. Neither requires external power connectors. The AMD part is an IGP with no slot width, integrated directly into the processor package, while the NVIDIA part is an MXM module using PCIe 3.0 x8. Display outputs are motherboard dependent for AMD and portable device dependent for NVIDIA. Both parts are end-of-life products. The AMD part released on 2019-11-19, while the NVIDIA part released on 2016-03-24, a gap of roughly three and a half years.

The Verdict

The data points to a clear but nuanced conclusion. For modern Vulkan-based workloads, the AMD Radeon Vega 3 is the superior part by a 32.6% margin. That is a decisive advantage, not a marginal one. Any application that leverages Vulkan for rendering or compute will run substantially better on the AMD IGP. The DirectX 12 (12_1) support further positions the AMD part for contemporary games that use feature level 12_1 features.

For OpenCL compute, the NVIDIA GeForce 920MX holds a marginal 2.6% lead. This is statistically significant in the sense that it is the recorded result, but practically negligible. Users running OpenCL workloads should expect near-identical performance from either part. The NVIDIA part's theoretical advantages in FP32 throughput, pixel rate, and texture rate did not translate into a commanding OpenCL win, suggesting that memory bandwidth or driver overhead limited its real-world performance.

The overall percentile rankings support the AMD part. It sits at the 25th percentile of all GPUs, four points above the NVIDIA part's 21st percentile. Its average benchmark score of 4268 is about 21% higher than the 920MX's 3528. However, the average comparison is skewed by the missing Metal result for NVIDIA and the very low Vulkan score. If the NVIDIA part had a Metal score comparable to its OpenCL result, its average would be higher, but it would still trail the AMD part.

The AMD part wins on process technology, transistor density, API feature level, and Vulkan performance. The NVIDIA part wins on raw shader count, TMU/ROP counts, theoretical fillrates, and dedicated memory. In practical benchmark terms, the AMD part wins the more important test (Vulkan) by a wide margin and loses the other test (OpenCL) by a negligible amount. The verdict favors the AMD Radeon Vega 3 for gaming and modern API workloads, while the NVIDIA GeForce 920MX remains competitive for legacy OpenCL compute tasks.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon Vega 3 averages 4268 across its three tests, while the NVIDIA GeForce 920MX averages 3528 across its two tests. The AMD part leads by roughly 740 points.

Q: How large is the Vulkan performance gap?

A: The AMD Radeon Vega 3 scores 3961 in Geekbench Vulkan, while the NVIDIA GeForce 920MX scores 2987. That is a 32.6% advantage for AMD.

Q: Does the NVIDIA GeForce 920MX win any benchmarks?

A: Yes, it wins Geekbench OpenCL with a score of 4068 against 3963 for the AMD Radeon Vega 3, a 2.6% margin.

Q: What memory configurations do the two parts use?

A: The AMD Vega 3 uses system shared memory with bandwidth dependent on the host system. The NVIDIA 920MX has 2 GB of DDR3 on a 64-bit bus with 14.40 GB/s bandwidth.

Q: Which part has more shading units?

A: The NVIDIA GeForce 920MX has 256 shading units, while the AMD Radeon Vega 3 has 192. NVIDIA also has double the TMUs (24 versus 12) and double the ROPs (8 versus 4).

Q: What are the process node differences?

A: The AMD Radeon Vega 3 is built on a 12 nm process at GlobalFoundries with 4,940 million transistors on a 210 mm² die. The NVIDIA GeForce 920MX uses a 28 nm process at TSMC with 1,020 million transistors on a 77 mm² die.

Specification Differences

| Specification | AMD Radeon Vega 3 | NVIDIA GeForce 920MX |

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

| Chip | Picasso | GM108S |

| Architecture | GCN 5.0 | Maxwell |

| Process Node | 12 nm | 28 nm |

| Foundry | GlobalFoundries | TSMC |

| Transistors | 4,940 million | 1,020 million |

| Die Size | 210 mm² | 77 mm² |

| Transistor Density | 23.5M / mm² | 13.2M / mm² |

| Base Clock | 300 MHz | 965 MHz |

| Boost Clock | 1100 MHz | 993 MHz |

| Memory Size | System Shared | 2 GB |

| Memory Type | System Shared | DDR3 |

| Memory Bus Width | System Shared | 64 bit |

| Memory Bandwidth | System Dependent | 14.40 GB/s |

| Shading Units | 192 | 256 |

| TMUs | 12 | 24 |

| ROPs | 4 | 8 |

| Pixel Rate | 4.400 GPixel/s | 7.944 GPixel/s |

| Texture Rate | 13.20 GTexel/s | 23.83 GTexel/s |

| FP32 Performance | 422.4 GFLOPS | 508.4 GFLOPS |

| FP16 Performance | 844.8 GFLOPS (2:1) | None |

| TDP | 15 W | 16 W |

| Slot Width | IGP | MXM Module |

| Bus Interface | IGP | PCIe 3.0 x8 |

| DirectX Support | 12 (12_1) | 12 (11_0) |

| OpenGL Support | 4.6 | 4.6 |

| Vulkan Support | 1.3 | 1.4 |

| Release Date | 2019-11-19 | 2016-03-24 |

| Predecessor | GCN 3.0 IGP | GeForce 800M |

| Successor | Vega II IGP | GeForce 10 Mobile |

| Production Status | End-of-life | End-of-life |