NVIDIA GeForce RTX 4070 AD103 vs NVIDIA GeForce RTX 5090 D Comparison

Head-to-Head Benchmarks

The benchmark database contains no recorded head-to-head results for the NVIDIA GeForce RTX 4070 AD103, as its benchmark list is empty. The NVIDIA GeForce RTX 5090 D has a full set of recorded scores across multiple test suites. This means the comparison relies entirely on the RTX 5090 D's measured performance and the RTX 4070 AD103's specification-derived capabilities.

The RTX 5090 D posts an average benchmark score of 77,712 across its recorded tests. Its percentile versus all GPUs is 92, placing it in the upper tier of the database. The nearest rival data gives context to this figure. The AMD Radeon RX 6650M XT scores 76,904, which is 1.1% lower than the RTX 5090 D. The AMD Radeon RX 6850M XT scores 78,940, which is 1.6% higher. The NVIDIA Tesla P100 PCIe 12 GB scores 79,396, which is 2.1% higher, and the NVIDIA Tesla P100 PCIe 16 GB scores 79,605, which is 2.4% higher. The RTX 5090 D therefore sits within a narrow band of competing hardware, slightly below four other recorded cards in average score but above one.

Within the RTX 5090 D's own benchmark suite, the compute-oriented tests show the strongest results. The Geekbench Vulkan score is 376,915, while the Geekbench OpenCL score is 310,674. The PassMark G3D score is 44,065, and the PassMark GPU Compute score is 28,396. These figures indicate that the card's raw compute throughput is its dominant strength. The DirectX tests are comparatively modest: PassMark DirectX 9 scores 434, DirectX 11 scores 371, DirectX 10 scores 231, and DirectX 12 scores 219. The PassMark G2D score is 1,487, which reflects 2D desktop performance rather than 3D rendering.

The 3DMark Steel Nomad DX12 test returns a score of 14,326. This is a single modern DirectX 12 workload, and it confirms the card's ability to handle current-generation rendering. The spread between the Vulkan and OpenCL scores suggests that the architecture favors Vulkan's lower-overhead path, which aligns with the Blackwell 2.0 design. The database shows no wins for either card in head-to-head benchmarks, as the head-to-head array is empty. All performance conclusions for the RTX 4070 AD103 must therefore be drawn from its hardware configuration, not from measured results.

Architecture Differences

The two cards come from different generations and architectures. The RTX 4070 AD103 uses the AD103 chip built on the Ada Lovelace architecture, while the RTX 5090 D uses the GB202 chip built on Blackwell 2.0. Both are fabricated by TSMC on a 5 nm process node. The transistor counts differ substantially: the AD103 contains 45,900 million transistors on a 379 mm² die, giving a density of 121.1 million transistors per square millimeter. The GB202 contains 92,200 million transistors on a 750 mm² die, giving a density of 122.9 million per square millimeter. The density is nearly identical, but the GB202 die is almost twice the physical size and holds just over twice the transistor count.

The execution resources scale accordingly. The RTX 4070 AD103 has 5,888 shading units, 184 texture mapping units, and 64 raster operation units. The RTX 5090 D has 21,760 shading units, 680 texture mapping units, and 176 raster operation units. That is roughly 3.7 times the shading units, 3.7 times the TMUs, and 2.75 times the ROPs. The RT core count is 46 on the AD103 versus 170 on the GB202, and the tensor core count is 184 versus 680. These ratios indicate that the Blackwell 2.0 chip dedicates a larger share of its die to ray tracing and tensor operations relative to the Ada chip.

Clock speeds are close but not identical. The RTX 4070 AD103 has a base clock of 1920 MHz and a boost clock of 2475 MHz. The RTX 5090 D has a base clock of 2017 MHz and a boost clock of 2407 MHz. The Ada card boosts higher, but the Blackwell card starts from a higher base. The memory subsystem differs even more sharply. The RTX 4070 AD103 uses 12 GB of GDDR6X on a 192-bit bus, with memory clocked at 1313 MHz (21 Gbps effective) and a bandwidth of 504.2 GB/s. The RTX 5090 D uses 32 GB of GDDR7 on a 512-bit bus, with memory clocked at 1750 MHz (28 Gbps effective) and a bandwidth of 1.79 TB/s. The bandwidth difference is roughly 3.5 times in favor of the RTX 5090 D, which is consistent with the larger bus and faster memory type.

The pixel and texture rates reflect these differences. The RTX 4070 AD103 delivers 158.4 GPixel/s and 455.4 GTexel/s. The RTX 5090 D delivers 423.6 GPixel/s and 1,636.8 GTexel/s. The FP32 and FP16 throughput figures are identical within each card, indicating a 1:1 ratio. The RTX 4070 AD103 computes 29.15 TFLOPS for both, while the RTX 5090 D computes 104.8 TFLOPS for both. The Blackwell card is about 3.6 times faster in raw floating-point work.

The interface and connectivity also differ. The RTX 4070 AD103 uses PCIe 4.0 x16, while the RTX 5090 D uses PCIe 5.0 x16. Display outputs are similar in count but different in version: the RTX 4070 AD103 has 1x HDMI 2.1 and 3x DisplayPort 1.4a, while the RTX 5090 D has 1x HDMI 2.1b and 3x DisplayPort 2.1b. Both cards support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The power requirements are far apart. The RTX 4070 AD103 has a TDP of 200 W with a suggested PSU of 550 W. The RTX 5090 D has a TDP of 575 W with a suggested PSU of 950 W. Both use a single 16-pin power connector and are dual-slot designs.

FAQ

Q: Which card has the higher memory bandwidth?

A: The RTX 5090 D has a recorded bandwidth of 1.79 TB/s, while the RTX 4070 AD103 has 504.2 GB/s. The RTX 5090 D's bandwidth is approximately 3.5 times higher.

Q: How do the transistor densities compare?

A: The RTX 4070 AD103 has a density of 121.1 million transistors per square millimeter, and the RTX 5090 D has 122.9 million per square millimeter. The densities are nearly identical, despite the RTX 5090 D having a larger die.

Q: What are the FP32 compute figures for each card?

A: The RTX 4070 AD103 delivers 29.15 TFLOPS, and the RTX 5090 D delivers 104.8 TFLOPS. The RTX 5090 D is roughly 3.6 times faster in FP32 and FP16, as both cards use a 1:1 ratio.

Q: Which card has more ray tracing cores?

A: The RTX 4070 AD103 has 46 RT cores, and the RTX 5090 D has 170 RT cores. The RTX 5090 D has about 3.7 times as many RT cores.

Q: What is the production status of each card?

A: The RTX 4070 AD103 is listed as end-of-life, while the RTX 5090 D is listed as active. The RTX 4070 AD103 was released on 2024-02-29, and the RTX 5090 D was released on 2025-01-29.

Q: How does the RTX 5090 D compare to its nearest rivals in average score?

A: The RTX 5090 D has an average benchmark score of 77,712. The AMD Radeon RX 6650M XT is 1.1% lower, the AMD Radeon RX 6850M XT is 1.6% higher, the NVIDIA Tesla P100 PCIe 12 GB is 2.1% higher, and the NVIDIA Tesla P100 PCIe 16 GB is 2.4% higher.

The Verdict

The data shows a clear generational and performance gulf between these two cards. The RTX 5090 D is the stronger part by every measured metric in the database. Its FP32 throughput is 104.8 TFLOPS versus 29.15 TFLOPS, its memory bandwidth is 1.79 TB/s versus 504.2 GB/s, and its memory capacity is 32 GB versus 12 GB. The RTX 5090 D also has more shading units, TMUs, ROPs, RT cores, and tensor cores. Its pixel rate is 423.6 GPixel/s versus 158.4 GPixel/s, and its texture rate is 1,636.8 GTexel/s versus 455.4 GTexel/s.

The RTX 5090 D's benchmark results confirm its position. Its average score of 77,712 and its 92nd percentile ranking place it among the top cards in the database. Its nearest rivals are all within 2.4% of its score, which suggests that the RTX 5090 D is competitive with, but not dominant over, the AMD Radeon RX 6850M XT and the Tesla P100 variants in aggregate scoring. The RTX 5090 D's Geekbench Vulkan score of 376,915 and OpenCL score of 310,674 show strong compute performance, while its PassMark G3D score of 44,065 indicates solid 3D rendering.

The RTX 4070 AD103 has no recorded benchmark scores, so its performance cannot be directly compared from measured data. Its specifications place it well below the RTX 5090 D in raw throughput. It is also end-of-life, while the RTX 5090 D is active. The RTX 5090 D is the logical choice for workloads that demand maximum compute, memory bandwidth, and ray tracing resources. The RTX 4070 AD103, with its lower TDP of 200 W and smaller physical footprint, fits scenarios where power and size constraints matter more than absolute performance. The launch MSRP of the RTX 4070 AD103 was 599 USD, and the RTX 5090 D was 2,299 USD. The RTX 5090 D uses a newer PCIe 5.0 interface, newer DisplayPort 2.1b outputs, and a newer memory type in GDDR7.

Specification Differences

The following fields differ between the two cards:

• Chip: AD103 versus GB202
• Architecture: Ada Lovelace versus Blackwell 2.0
• Generation: GeForce 40 versus GeForce 50
• Transistors: 45,900 million versus 92,200 million
• Die size: 379 mm² versus 750 mm²
• Transistor density: 121.1M / mm² versus 122.9M / mm²
• Base clock: 1920 MHz versus 2017 MHz
• Boost clock: 2475 MHz versus 2407 MHz
• Memory clock: 1313 MHz (21 Gbps effective) versus 1750 MHz (28 Gbps effective)
• Memory size: 12 GB versus 32 GB
• Memory type: GDDR6X versus GDDR7
• Memory bus width: 192 bit versus 512 bit
• Memory bandwidth: 504.2 GB/s versus 1.79 TB/s
• Shading units: 5,888 versus 21,760
• TMUs: 184 versus 680
• ROPs: 64 versus 176
• RT cores: 46 versus 170
• Tensor cores: 184 versus 680
• Pixel rate: 158.4 GPixel/s versus 423.6 GPixel/s
• Texture rate: 455.4 GTexel/s versus 1,636.8 GTexel/s
• FP32: 29.15 TFLOPS versus 104.8 TFLOPS
• FP16: 29.15 TFLOPS (1:1) versus 104.8 TFLOPS (1:1)
• TDP: 200 W versus 575 W
• Suggested PSU: 550 W versus 950 W
• Bus interface: PCIe 4.0 x16 versus PCIe 5.0 x16
• Display outputs: 1x HDMI 2.1, 3x DisplayPort 1.4a versus 1x HDMI 2.1b, 3x DisplayPort 2.1b
• Dimensions: 240 mm x 110 mm x 40 mm versus 304 mm x 137 mm x 48 mm
• Production status: End-of-life versus Active
• Release date: 2024-02-29 versus 2025-01-29
• Predecessor: GeForce 30 versus GeForce 40
• Successor: GeForce 50 versus GeForce 60
• Launch MSRP: 599 USD versus 2,299 USD
• Percentile versus all GPUs: 50 versus 92
• Average benchmark score: 0 versus 77,712
• Benchmark list: empty versus full set of 10 recorded tests