NVIDIA GeForce RTX 5090 D vs NVIDIA RTX 6000 Ada Generation Comparison

Head-to-Head Benchmarks

The recorded data shows a split decision in the direct comparisons, with each card taking one victory. In Geekbench OpenCL, the NVIDIA RTX 6000 Ada Generation edges out the RTX 5090 D by a razor-thin margin, scoring 311,629 versus 310,674. That 0.3% difference is negligible in real-world terms, effectively a statistical tie in raw compute throughput. The more dramatic separation appears in Geekbench Vulkan, where the RTX 5090 D delivers 376,915 points against the RTX 6000 Ada's 262,845. That represents a 30.3% advantage for the GeForce card, a commanding lead in API-specific performance that points to architectural efficiency rather than simple brute force.

Looking at the broader benchmark context, the RTX 6000 Ada Generation sits in the 99th percentile of all GPUs in the database, with an average benchmark score of 287,237 across its recorded tests. The RTX 5090 D, by contrast, holds a 92nd percentile ranking with an average score of 77,712. That average is heavily skewed by the inclusion of Passmark legacy DirectX tests, which produce low scores (231 for DirectX 10, 371 for DirectX 11, 219 for DirectX 12, 434 for DirectX 9), as well as a 2D score of 1,487 and a compute score of 28,396. The GeForce card's Passmark G3D score of 44,065 shows strong rasterization performance, but the average benchmark figure is not directly comparable to the workstation card's Geekbench-only results.

The RTX 6000 Ada's nearest rivals in the database include the NVIDIA L40 at 284,111 (1.1% behind), the NVIDIA L40S at 295,763 (2.9% ahead), the AMD Instinct MI300X at 317,994 (9.7% ahead), and the NVIDIA L20 at 251,147 (14.4% behind). These deltas place the RTX 6000 Ada firmly in the professional compute tier, trading blows with dedicated data center accelerators. The RTX 5090 D's nearest rivals are a different class entirely: the AMD Radeon RX 6650M XT at 76,904 (1.1% behind), the AMD Radeon RX 6850M XT at 78,940 (1.6% ahead), the NVIDIA Tesla P100 PCIe 12 GB at 79,396 (2.1% ahead), and the NVIDIA Tesla P100 PCIe 16 GB at 79,605 (2.4% ahead). This comparison set reflects the RTX 5090 D's mixed benchmark portfolio rather than its true compute ceiling, as the Vulkan result demonstrates.

Architecture Differences

The two cards come from different NVIDIA generations and are built on different architectures. The RTX 6000 Ada Generation uses the AD102 chip with Ada Lovelace architecture, part of the Workstation Ada generation, while the RTX 5090 D uses the GB202 chip with Blackwell 2.0 architecture, part of the GeForce 50 generation. Both are manufactured by TSMC on a 5 nm process node, but the similarities end there. The AD102 die measures 609 mm² and contains 76,300 million transistors, yielding a density of 125.3 million transistors per mm². The GB202 die is larger at 750 mm², packs 92,200 million transistors, and has a slightly lower density of 122.9 million per mm².

Clock behavior differs substantially. The RTX 6000 Ada has a base clock of 915 MHz and a boost clock of 2505 MHz, while the RTX 5090 D starts at 2017 MHz base and boosts to 2407 MHz. The workstation card runs a lower base clock but achieves a higher boost, while the GeForce card operates at a much higher floor. Memory configurations diverge in capacity, type, and speed. The RTX 6000 Ada features 48 GB of GDDR6 on a 384-bit bus, delivering 960.0 GB/s of bandwidth. The RTX 5090 D offers 32 GB of GDDR7 on a wider 512-bit bus, reaching 1.79 TB/s. The memory clock rates reflect the generational shift: 2500 MHz with 20 Gbps effective for the Ada card, versus 1750 MHz with 28 Gbps effective for the Blackwell card.

Compute resources scale differently across the two dies. The RTX 6000 Ada has 18,176 shading units, 568 texture mapping units, 192 ROPs, 142 RT cores, and 568 tensor cores. The RTX 5090 D increases shading units to 21,760, TMUs to 680, and RT cores to 170, but reduces ROPs to 176 while keeping tensor cores at 680. The resulting throughput figures favor the GeForce card in raw FP32 and FP16, both rated at 104.8 TFLOPS versus 91.06 TFLOPS for the Ada card. Pixel rate tells a different story: the RTX 6000 Ada delivers 481.0 GPixel/s versus 423.6 GPixel/s for the RTX 5090 D, reflecting the higher ROP count. Texture rate favors the GeForce card at 1,636.8 GTexel/s against 1,422.8 GTexel/s.

Power and interface requirements are not comparable in magnitude. The RTX 6000 Ada has a 300 W TDP with a 700 W suggested PSU, while the RTX 5090 D draws 575 W and calls for a 950 W PSU. Both use a single 16-pin connector and dual-slot cooling. The bus interface also advances: PCIe 4.0 x16 on the Ada card, PCIe 5.0 x16 on the Blackwell card. Display outputs differ, with the RTX 6000 Ada offering four DisplayPort 1.4a connections, while the RTX 5090 D provides one HDMI 2.1b and three DisplayPort 2.1b outputs.

The Verdict

The data points to a clear use-case separation. The RTX 6000 Ada Generation is the professional compute card, evidenced by its 99th percentile ranking, its 48 GB memory capacity, and its standing among workstation rivals like the L40S and Instinct MI300X. The RTX 5090 D is the consumer gaming and general-purpose card, with a 92nd percentile ranking and a benchmark suite that includes 3DMark Steel Nomad DX12 (14,326) alongside multiple Passmark legacy tests. The Vulkan result is decisive for the GeForce card, but the OpenCL result is essentially a draw.

For workloads that rely on Vulkan, the RTX 5090 D holds a 30.3% advantage. For OpenCL compute, the RTX 6000 Ada is marginally ahead, though the 0.3% delta is within noise. The RTX 6000 Ada's production status is end-of-life, while the RTX 5090 D is active, which matters for availability. The RTX 6000 Ada launched on December 2, 2022, with a launch MSRP of 6,799 USD. The RTX 5090 D launched on January 29, 2025, with a launch MSRP of 2,299 USD.

The successor chain reinforces the positioning: the RTX 6000 Ada follows Workstation Ampere and leads to Blackwell PRO W, while the RTX 5090 D follows GeForce 40 and leads to GeForce 60. Users needing maximum memory capacity and sustained compute with lower power draw should favor the Ada card. Users needing higher raw throughput, better Vulkan performance, and newer display outputs should choose the 5090 D.

FAQ

Q: Which card wins in Geekbench OpenCL?

A: The NVIDIA RTX 6000 Ada Generation scores 311,629 against the RTX 5090 D's 310,674, a 0.3% difference that makes the result effectively a tie.

Q: How large is the Vulkan performance gap?

A: The RTX 5090 D scores 376,915 in Geekbench Vulkan, which is 30.3% higher than the RTX 6000 Ada's 262,845.

Q: What memory configuration does each card use?

A: The RTX 6000 Ada has 48 GB of GDDR6 on a 384-bit bus with 960.0 GB/s bandwidth. The RTX 5090 D has 32 GB of GDDR7 on a 512-bit bus with 1.79 TB/s bandwidth.

Q: How do the TDP requirements compare?

A: The RTX 6000 Ada has a 300 W TDP and suggests a 700 W PSU, while the RTX 5090 D has a 575 W TDP and suggests a 950 W PSU.

Q: What is the production status of each card?

A: The RTX 6000 Ada Generation is end-of-life, while the RTX 5090 D is active in production.

Q: Which card has higher FP32 throughput?

A: The RTX 5090 D achieves 104.8 TFLOPS in FP32, while the RTX 6000 Ada delivers 91.06 TFLOPS.

Where Each One Wins

The RTX 6000 Ada Generation wins in memory capacity with 48 GB versus 32 GB, in pixel fill rate with 481.0 GPixel/s versus 423.6 GPixel/s, and in ROP count with 192 versus 176. Its lower TDP of 300 W versus 575 W makes it the more power-efficient choice for sustained compute workloads, and its PCIe 4.0 interface is sufficient for its bandwidth needs. The OpenCL benchmark result, though marginal, also favors this card.

The RTX 5090 D wins decisively in Vulkan performance with a 30.3% margin, in FP32 and FP16 throughput with 104.8 TFLOPS versus 91.06 TFLOPS, in texture rate with 1,636.8 GTexel/s versus 1,422.8 GTexel/s, and in memory bandwidth with 1.79 TB/s versus 960.0 GB/s. It also offers more shading units (21,760 versus 18,176), more TMUs (680 versus 568), more RT cores (170 versus 142), and more tensor cores (680 versus 568). The newer PCIe 5.0 bus interface and the inclusion of HDMI 2.1b and DisplayPort 2.1b outputs give it an edge in modern display connectivity.

For gaming and DirectX workloads, the RTX 5090 D's Passmark scores provide evidence of strong performance: 44,065 in G3D and 28,396 in compute, alongside a 3DMark Steel Nomad DX12 score of 14,326. The RTX 6000 Ada has no recorded DirectX or 3DMark scores in the database, so its gaming credentials rest solely on its architecture and compute results.

Specification Differences

The two cards differ in nearly every measured specification. The chip is AD102 for the RTX 6000 Ada and GB202 for the RTX 5090 D, with architectures Ada Lovelace and Blackwell 2.0 respectively. Transistor counts are 76,300 million versus 92,200 million, and die sizes are 609 mm² versus 750 mm². Transistor density is 125.3M per mm² for the Ada card and 122.9M per mm² for the GeForce card.

Clock speeds differ: base clocks are 915 MHz versus 2017 MHz, boost clocks are 2505 MHz versus 2407 MHz, and memory clocks are 2500 MHz (20 Gbps effective) versus 1750 MHz (28 Gbps effective). Memory size is 48 GB GDDR6 versus 32 GB GDDR7, with bus widths of 384-bit versus 512-bit and bandwidths of 960.0 GB/s versus 1.79 TB/s.

Compute units vary: shading units are 18,176 versus 21,760, TMUs are 568 versus 680, ROPs are 192 versus 176, RT cores are 142 versus 170, and tensor cores are 568 versus 680. Pixel rates are 481.0 GPixel/s versus 423.6 GPixel/s, texture rates are 1,422.8 GTexel/s versus 1,636.8 GTexel/s, and FP32/FP16 figures are 91.06 TFLOPS versus 104.8 TFLOPS.

Power and physical specs show the RTX 5090 D as the larger, hungrier card: TDP is 300 W versus 575 W, suggested PSU is 700 W versus 950 W, and dimensions are 267 mm by 112 mm versus 304 mm by 137 mm by 48 mm. Both use dual-slot cooling and a single 16-pin power connector. Bus interfaces are PCIe 4.0 x16 versus PCIe 5.0 x16. Display outputs are 4x DisplayPort 1.4a versus 1x HDMI 2.1b and 3x DisplayPort 2.1b. DirectX, OpenGL, and Vulkan API support are identical at 12 Ultimate (12_2), 4.6, and 1.4 respectively.