AMD Radeon RX Vega 64 vs NVIDIA GeForce RTX 5090 D Comparison

The NVIDIA GeForce RTX 5090 D and the AMD Radeon RX Vega 64 represent two very different eras of GPU design, and the recorded benchmark data makes the gap between them unmistakable. The RTX 5090 D, built on the Blackwell 2.0 architecture and released in January 2025, sweeps all three head-to-head benchmarks against the Vega 64, a 2017-era GCN 5.0 part. The magnitude of those wins, however, varies by test, and the database contains some interesting placement quirks for both cards that are worth unpacking before drawing conclusions.

Head-to-Head Benchmarks

Across the three tests where both GPUs have recorded scores, the RTX 5090 D wins all of them, and by enormous margins.

The most dramatic result is 3DMark Steel Nomad (DirectX 12), where the RTX 5090 D scores 14326 against the Vega 64's 1669. That is roughly 8.6 times the score, a gap of about 758 percent in favor of the NVIDIA card. Steel Nomad is a modern, demanding DirectX 12 workload, and this result captures how far the rendering pipeline has evolved between GCN 5.0 and Blackwell 2.0. The Vega 64 was a flagship in its day, but this test simply overwhelms its 4096 shading units and 12.66 TFLOPS of FP32 throughput.

Geekbench Vulkan shows a similar story: 376915 for the RTX 5090 D versus 67032 for the Vega 64, a lead of approximately 462 percent. Geekbench OpenCL is the closest contest of the three, with the RTX 5090 D scoring 310674 against 62552, still a lead of nearly 397 percent. Even in the test where the Vega 64 fares relatively best, the Blackwell card delivers almost five times the compute result, which tracks with the raw hardware disparity: 104.8 TFLOPS of FP32 against 12.66 TFLOPS.

One caveat for database users: the Vega 64 also has a Geekbench Metal score of 68750 recorded, but the RTX 5090 D has no Metal result in the database, so no comparison is possible on that API. Similarly, the RTX 5090 D carries a full set of Passmark results (Passmark G3D at 44065, GPU Compute at 28396, plus DirectX 9 through 12 subtests ranging from 219 to 434 and a G2D score of 1487) for which the Vega 64 has no recorded entries. The head-to-head record is therefore limited to the three shared tests, all won by NVIDIA.

Context from each card's rival list adds useful perspective. The RTX 5090 D's average benchmark score of 77712 places it within 1.1 percent of the AMD Radeon RX 6650M XT (76904), 1.6 percent behind the RX 6850M XT (78940), and 2.1 to 2.4 percent behind the NVIDIA Tesla P100 PCIe variants (79396 and 79605). The Vega 64's average of 50001 sits almost exactly level with the GeForce RTX 5070 Ti (49957, a 0.1 percent gap) and the Intel Arc A550M (49737, 0.5 percent), while trailing the RX 6900 XT (50951) by 1.9 percent and leading the RX 6800 XT (48477) by 3.1 percent. In percentile terms, the RTX 5090 D ranks in the 92nd percentile against all GPUs in the database and the Vega 64 in the 86th, a six-point spread that is smaller than the raw benchmark gaps might suggest, reflecting how the percentile distribution compresses near the top.

The Verdict

For any workload represented in the head-to-head data, the RTX 5090 D is the clear choice. A gamer or compute user choosing between these two cards specifically is choosing between a card that wins every shared benchmark by roughly four to eight times and one that wins none. The Vega 64's case rests entirely on secondary factors visible in the spec sheet rather than the benchmark record.

Power is the most obvious differentiator. The RTX 5090 D has a TDP of 575 W with a suggested PSU of 950 W and a single 16-pin connector, while the Vega 64 draws 295 W, uses two 8-pin connectors, and pairs with a suggested 600 W PSU. The Vega 64 is also physically smaller: 280 mm long, 111 mm tall, and 40 mm wide, versus 304 mm by 137 mm by 48 mm for the RTX 5090 D. Both are dual-slot cards, but the NVIDIA part is longer, taller, and thicker, and demands substantially more from the host system.

Availability matters too. The Vega 64 is listed as end-of-life, while the RTX 5090 D is in active production. For a new build, only one of these can realistically be purchased as current inventory. The database lists a launch MSRP of 2,299 USD for the RTX 5090 D and 499 USD for the Vega 64; each figure is stated here once as recorded and no further pricing analysis is offered.

In short: pick the RTX 5090 D for performance in every measured category. Pick the Vega 64 only if the constraints that matter are power draw, physical dimensions, or legacy platform compatibility such as PCIe 3.0, none of which the benchmarks capture.

Architecture Differences

The architectural distance between these two GPUs spans nearly a decade of design philosophy. The RTX 5090 D uses the GB202 chip on TSMC's 5 nm process, packing 92,200 million transistors into a 750 mm² die for a density of 122.9 million transistors per square millimeter. The Vega 64's Vega 10 chip, fabricated by GlobalFoundries on 14 nm, holds 12,500 million transistors across 495 mm² at 25.3 million per square millimeter. That is roughly 7.4 times the transistor count and close to five times the density for the Blackwell part.

Compute resources scale accordingly. The RTX 5090 D has 21760 shading units, 680 texture mapping units, 176 render output units, 170 RT cores, and 680 tensor cores. The Vega 64 has 4096 shading units, 256 TMUs, and 64 ROPs, and no dedicated RT or tensor cores at all, since GCN 5.0 predates both hardware ray tracing and accelerated matrix operations. Resulting throughput: 423.6 GPixel/s and 1636.8 GTexel/s for NVIDIA against 98.94 GPixel/s and 395.8 GTexel/s for AMD. FP16 behavior differs too: the RTX 5090 D runs FP16 at 1:1 with FP32 (104.8 TFLOPS each), while the Vega 64 doubles FP16 throughput to 25.33 TFLOPS, a 2:1 ratio that was a Vega-specific feature.

Memory subsystems reflect different eras of bandwidth strategy. The Vega 64 uses 8 GB of HBM2 on a 2048-bit bus at 945 MHz (1890 Mbps effective), yielding 483.8 GB/s. The RTX 5090 D uses 32 GB of GDDR7 on a 512-bit bus at 1750 MHz (28 Gbps effective) for 1.79 TB/s. The capacity gap, four times, is as significant as the bandwidth gap for modern workloads.

Clocking and platform support diverge as well: 2017 MHz base and 2407 MHz boost for the RTX 5090 D against 1247 MHz base and 1546 MHz boost for the Vega 64. The NVIDIA card uses PCIe 5.0 x16 with DirectX 12 Ultimate (12_2), Vulkan 1.4, and OpenGL 4.6, plus one HDMI 2.1b and three DisplayPort 2.1b outputs. The AMD card uses PCIe 3.0 x16, DirectX 12 (12_1), Vulkan 1.3, OpenGL 4.6, and one HDMI 2.0b with three DisplayPort 1.4a outputs.

FAQ

Q: How large is the performance gap in the shared benchmarks?

A: The RTX 5090 D leads by about 758 percent in 3DMark Steel Nomad (14326 vs 1669), about 462 percent in Geekbench Vulkan (376915 vs 67032), and about 397 percent in Geekbench OpenCL (310674 vs 62552).

Q: Which card has the higher database percentile ranking?

A: The RTX 5090 D sits in the 92nd percentile against all GPUs in the database; the Vega 64 sits in the 86th.

Q: How much power does each card require?

A: The RTX 5090 D has a 575 W TDP, one 16-pin connector, and a suggested 950 W PSU. The Vega 64 has a 295 W TDP, two 8-pin connectors, and a suggested 600 W PSU.

Q: Does the Vega 64 support ray tracing?

A: No. It has no RT cores and no tensor cores, while the RTX 5090 D includes 170 RT cores and 680 tensor cores.

Q: Which cards are the closest rivals to each in the database?

A: The RTX 5090 D is within about 1 to 2.4 percent of the RX 6650M XT, RX 6850M XT, and both Tesla P100 PCIe variants on average score. The Vega 64 is within about 0.1 to 3.1 percent of the RTX 5070 Ti, Arc A550M, RX 6900 XT, and RX 6800 XT.

Q: Are both cards still in production?

A: No. The Vega 64 is listed as end-of-life; the RTX 5090 D is listed as active.

Where Each One Wins

The RTX 5090 D wins everywhere the database measures both cards: DirectX 12 gaming simulation, Vulkan compute, and OpenCL compute. Add its unmatched-in-this-pairing Passmark suite, including a G3D score of 44065 and GPU Compute score of 28396, and it is the choice for high-refresh gaming, ray-traced titles, heavy compute workloads, and anything needing 32 GB of fast GDDR7 with 1.79 TB/s of bandwidth.

The Vega 64's wins are infrastructural rather than performance-based. It wins on power envelope, with a TDP roughly half that of the Blackwell card and a PSU recommendation 350 W lower. It wins on physical size, being shorter, shorter in height, and thinner. Its HBM2 memory on a 2048-bit bus was a bandwidth-forward design for its era, and its 2:1 FP16 ratio at 25.33 TFLOPS still represents meaningful half-precision throughput. For legacy PCIe 3.0 systems or older display setups, its connectivity is sufficient. But on every measured benchmark in this comparison, the data shows a clean sweep for the RTX 5090 D, and by margins that leave no room for interpretation.