NVIDIA GeForce 920M vs NVIDIA RTX PRO 4000 Blackwell SFF Comparison

NVIDIA’s GeForce 920M and RTX PRO 4000 Blackwell SFF represent two extremes of the GPU spectrum, separated by a decade of architecture and a vast gulf in capability. The data shows a complete mismatch in performance class, with the RTX PRO 4000 delivering massively higher raw throughput, while the 920M remains a legacy part for basic mobile tasks. This analysis breaks down the quantitative differences and what they mean for real-world workloads.

Head-to-Head Benchmarks

Direct head-to-head benchmark data is not available for these two products; the 920M has been tested with Geekbench OpenCL and Vulkan, while the RTX PRO 4000 has a single 3DMark Steel Nomad DX12 result. The performance gap is nonetheless unambiguous when comparing their theoretical peak outputs and the context of their nearest rivals.

The RTX PRO 4000 Blackwell SFF achieves an FP32 throughput of 24.05 TFLOPS, a figure that dwarfs the 920M’s 732.7 GFLOPS. This represents a 32.8x increase in raw single-precision compute power, meaning the RTX PRO 4000 can execute over thirty times more floating-point operations per second. In texture processing, the RTX PRO 4000 delivers 375.8 GTexel/s versus the 920M’s 30.53 GTexel/s, a 12.3x advantage. Pixel fill rates tell a similar story: 128.8 GPixel/s for the RTX PRO 4000 versus 7.632 GPixel/s for the 920M, a 16.9x difference.

In terms of benchmark percentile placement, the 920M sits at the 20th percentile of all GPUs, with an average benchmark score of 3287. Its nearest rivals are the GeForce GT 730M (avg score 3316, -0.9% delta), Intel HD Graphics 530 (avg score 3332, -1.4% delta), GeForce GT 640 (avg score 3210, +2.4% delta), and Intel HD Graphics P4600 (avg score 3389, -3% delta). This places the 920M in the entry-level integrated-graphics class, where its performance is statistically indistinguishable from those parts.

The RTX PRO 4000 holds the 19th percentile, with an average benchmark score of 2910 from the 3DMark Steel Nomad DX12 test. Its nearest rivals are the GeForce RTX 4060 Ti 16 GB (avg score 2907, +0.1% delta), RTX 4060 Ti 8 GB (avg score 2913, -0.1% delta), Quadro P600 (avg score 2923, -0.4% delta), and RTX 4010 (avg score 2893, +0.6% delta). The deltaPct values are all within 0.6%, indicating these five GPUs are functionally tied in this specific workload. This is notable: the RTX PRO 4000’s percentile rank is low despite its massive compute specifications, suggesting the 3DMark Steel Nomad test is not representative of its full capabilities, or that the benchmark results are heavily dependent on driver optimization and specific workload characteristics.

Architecture Differences

The GeForce 920M is built on the Kepler 2.0 architecture using the GK208B chip, manufactured on a 28 nm process at TSMC with 1,020 million transistors on an 87 mm² die. This yields a transistor density of 11.7M per mm². It features 384 shading units, 32 texture mapping units, and 8 raster output units. It has no ray tracing cores and no tensor cores. The memory subsystem is 2 GB of DDR3 on a 64-bit bus, delivering 14.40 GB/s of bandwidth. The base and boost clocks are both 954 MHz, with memory running at 900 MHz (1800 Mbps effective). It supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. The bus interface is PCIe 3.0 x8, and it is an integrated graphics processor (IGP) with portable-device-dependent display outputs.

The RTX PRO 4000 Blackwell SFF uses the Blackwell 2.0 architecture with the GB203 chip, manufactured on a 5 nm process at TSMC with 45,600 million transistors on a 378 mm² die. This gives a transistor density of 120.6M per mm², a 10.3x improvement in density over the 920M. It has 8,960 shading units, 280 TMUs, and 96 ROPs, plus 70 dedicated ray tracing cores and 280 tensor cores. Memory is 24 GB of GDDR7 on a 192-bit bus, yielding 432.0 GB/s of bandwidth. The base clock is a low 405 MHz, but the boost clock reaches 1342 MHz. The memory clock is 1125 MHz (18 Gbps effective). It supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The bus interface is PCIe 5.0 x8, and it is a dual-slot card with four mini-DisplayPort 2.1b outputs measuring 167 mm in length, 69 mm in height, and 40 mm in width. The RTX PRO 4000’s FP16 throughput is 24.05 TFLOPS (1:1 ratio with FP32), while the 920M has no listed FP16 capability.

The architectural gap is generational. Kepler 2.0 lacks hardware ray tracing and tensor acceleration entirely, while Blackwell 2.0 is built around them. The 70 RT cores and 280 tensor cores on the RTX PRO 4000 enable workloads that the 920M cannot attempt, including hardware-accelerated ray tracing and AI inference. The memory difference is similarly stark: 24 GB GDDR7 versus 2 GB DDR3, with a 30x bandwidth advantage (432.0 GB/s vs 14.40 GB/s). The RTX PRO 4000 also supports the latest DirectX 12 Ultimate feature set, whereas the 920M is limited to the baseline DirectX 12 (11_0) feature level.

The Verdict

The data leads to an unambiguous verdict: the RTX PRO 4000 Blackwell SFF is in a completely different performance class than the GeForce 920M. Any user requiring modern 3D rendering, ray tracing, AI acceleration, or large memory footprints must choose the RTX PRO 4000. The 920M’s 2 GB DDR3 memory and 14.40 GB/s bandwidth are insufficient for any contemporary professional workload, and its lack of RT and tensor cores eliminates it from consideration for ray-traced or AI-accelerated tasks.

Conversely, the 920M may be the only option in its specific niche: ultra-low-power integrated graphics for legacy laptops. Its 33 W TDP and IGP form factor mean it can be embedded in portable devices without discrete power connectors or cooling solutions. For basic 2D desktop rendering, video playback, and very light GPU-accelerated tasks on a 2015-era mobile platform, the 920M is functional. The RTX PRO 4000, despite its relatively modest 70 W TDP, still requires a dual-slot footprint and a 250 W suggested PSU, making it unsuitable for the 920M’s intended environment.

Benchmark data shows the 920M performs within 3% of its nearest rivals (GT 730M, HD Graphics 530, GT 640, HD Graphics P4600), confirming it is a baseline entry-level part. The RTX PRO 4000 performs within 0.6% of its nearest rivals (RTX 4060 Ti 16 GB, RTX 4060 Ti 8 GB, Quadro P600, RTX 4010) in the Steel Nomad test, indicating it is competitive with mid-range desktop GPUs. The choice is clear: for any workload beyond the most basic, the RTX PRO 4000 is the only viable option. For a strictly legacy integrated mobile GPU, the 920M remains the only one of the two that fits that form factor.

Specification Differences

| Specification | NVIDIA GeForce 920M | NVIDIA RTX PRO 4000 Blackwell SFF |

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

| Architecture | Kepler 2.0 | Blackwell 2.0 |

| Chip | GK208B | GB203 |

| Process Node | 28 nm | 5 nm |

| Transistors | 1,020 million | 45,600 million |

| Die Size | 87 mm² | 378 mm² |

| Base Clock | 954 MHz | 405 MHz |

| Boost Clock | 954 MHz | 1342 MHz |

| Memory Size | 2 GB | 24 GB |

| Memory Type | DDR3 | GDDR7 |

| Memory Bus Width | 64 bit | 192 bit |

| Memory Bandwidth | 14.40 GB/s | 432.0 GB/s |

| Shading Units | 384 | 8,960 |

| TMUs | 32 | 280 |

| ROPs | 8 | 96 |

| RT Cores | None | 70 |

| Tensor Cores | None | 280 |

| Pixel Rate | 7.632 GPixel/s | 128.8 GPixel/s |

| Texture Rate | 30.53 GTexel/s | 375.8 GTexel/s |

| FP32 | 732.7 GFLOPS | 24.05 TFLOPS |

| FP16 | None | 24.05 TFLOPS (1:1) |

| TDP | 33 W | 70 W |

| Slot Width | IGP | Dual-slot |

| Power Connectors | None | None |

| Suggested PSU | None | 250 W |

| Bus Interface | PCIe 3.0 x8 | PCIe 5.0 x8 |

| Display Outputs | Portable Device Dependent | 4x mini-DisplayPort 2.1b |

| DirectX | 12 (11_0) | 12 Ultimate (12_2) |

| Vulkan | 1.2.175 | 1.4 |

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

| Release Date | 2015-03-12 | 2025-08-10 |

| Predecessor | GeForce 800M | Workstation Ada |

| Successor | GeForce 10 Mobile | None |

FAQ

Q: Which GPU has higher raw compute performance?

A: The RTX PRO 4000 Blackwell SFF has an FP32 throughput of 24.05 TFLOPS, which is 32.8x higher than the GeForce 920M’s 732.7 GFLOPS.

Q: What are the memory capacity and bandwidth differences?

A: The RTX PRO 4000 has 24 GB of GDDR7 memory with 432.0 GB/s bandwidth on a 192-bit bus. The 920M has 2 GB of DDR3 with 14.40 GB/s bandwidth on a 64-bit bus.

Q: Does the GeForce 920M support ray tracing or tensor cores?

A: No. The 920M has no RT cores and no tensor cores. The RTX PRO 4000 has 70 RT cores and 280 tensor cores.

Q: How do the two GPUs compare in benchmark percentile ranking?

A: The 920M is at the 20th percentile with an average score of 3287, while the RTX PRO 4000 is at the 19th percentile with an average score of 2910. These percentiles are based on different benchmark suites, so they are not directly comparable.

Q: What is the power consumption difference?

A: The 920M has a TDP of 33 W, while the RTX PRO 4000 has a TDP of 70 W. The RTX PRO 4000 also recommends a 250 W system PSU.

Q: Which GPU is currently in production?

A: The RTX PRO 4000 is marked as Active, while the GeForce 920M is End-of-life. The 920M was released on 2015-03-12, and the RTX PRO 4000 on 2025-08-10.

Where Each One Wins

GeForce 920M: The 920M wins exclusively in the ultra-mobile, low-power integrated segment. Its 33 W TDP and IGP form factor with no power connectors make it suitable for portable devices where discrete GPU power delivery is impossible. It outperforms its closest rivals (GT 730M, HD Graphics 530, GT 640, HD Graphics P4600) by up to 2.4% in average benchmark score, placing it at the top of its immediate peer group. For legacy systems requiring basic 2D acceleration or video output on a 28 nm Kepler platform, the 920M is the only one of the two that fits. It also supports PCIe 3.0 x8, which is compatible with older motherboards, whereas the RTX PRO 4000 requires PCIe 5.0 x8.

RTX PRO 4000 Blackwell SFF: The RTX PRO 4000 wins in every measurable performance category. It has 23.3x more shading units (8,960 vs 384), 8.75x more TMUs (280 vs 32), and 12x more ROPs (96 vs 8). Its memory bandwidth is 30x higher, and its FP32 compute is 32.8x higher. The 70 RT cores and 280 tensor cores enable ray-traced rendering and AI acceleration, which the 920M cannot perform. The 24 GB GDDR7 frame buffer supports large models and high-resolution textures that would exhaust the 920M’s 2 GB DDR3. The RTX PRO 4000’s four mini-DisplayPort 2.1b outputs support modern multi-monitor professional setups, versus the 920M’s portable-device-dependent outputs. The RTX PRO 4000 is competitive with mid-range desktop GPUs, as shown by its deltaPct values within 0.6% of the RTX 4060 Ti 16 GB and RTX 4060 Ti 8 GB in 3DMark Steel Nomad DX12. For any professional 3D, AI, or high-bandwidth workload, the RTX PRO 4000 is the definitive winner.