Nvidia RTX 50 Series Index
A sourced specification index of every cataloged Nvidia RTX 50-series variant, including memory, throughput, and power.
A generation is a range, not one GPU
The RTX 50 series is easiest to misunderstand when its family name is treated as a single specification. A generation stretches across products with different enabled resources, memory capacities, thermal targets, and intended systems. Desktop and mobile names may sound close while their operating envelopes differ sharply. This index keeps the source variants visible so the family can be examined as a product range rather than reduced to the fastest launch announcement.
The catalog contains 10 variants RTX 50-series variants under the source section rule, spanning 2025 through 2025. The unit is variants, not board-partner models, retail listings, or unique marketing stems. Nvidia contributes 10 variants by definition; AMD and Intel counts remain 0 variants and 0 variants because this is a vendor generation index, not a cross-vendor threshold. Counts and dates are loaded from the catalog at render time rather than embedded in the essay.
Reading the specification spread
Memory size is immediately visible, but it should be read alongside memory technology, theoretical FP32 throughput, and thermal design power. Those values describe different constraints. Capacity affects residency, throughput describes a theoretical arithmetic ceiling, and TDP indicates a design power class rather than wall-socket consumption in every workload. A higher value in one column does not automatically make a variant preferable for every use.
Generation pages are also snapshots of segmentation. Vendors use one architecture across price and power bands by changing active resources, clocks, memory configurations, and package targets. The resulting gaps and clusters can say more about the intended market than the shared generation label. This table is a sourced inventory, not a recommendation list and not a benchmark chart. It preserves unknowns as absent evidence, does not invent values from neighboring products, and links the rendered result to the underlying Wikipedia revision. That makes the page useful for the narrower question it can answer honestly: which variants the catalog places in this source-defined series, and how their published specifications differ.
Catalog variants
| Vendor | Variant | Year | VRAM | Memory | FP32 GFLOPS | TDP | Connector evidence | Node | Transistors M |
|---|---|---|---|---|---|---|---|---|---|
| nvidia | GeForce RTX 5080 | 2025 | 16,384 MB | GDDR7 | 49,352 | 360 W | not recorded by source | — | 45,600 |
| nvidia | GeForce RTX 5090 | 2025 | 32,768 MB | GDDR7 | 87,780 | 575 W | not recorded by source | — | 92,200 |
| nvidia | GeForce RTX 5090 D | 2025 | 32,768 MB | GDDR7 | 87,780 | 575 W | not recorded by source | — | 92,200 |
| nvidia | GeForce RTX 5070 Ti | 2025 | 16,384 MB | GDDR7 | 41,126 | 300 W | not recorded by source | — | 45,600 |
| nvidia | GeForce RTX 5070 | 2025 | 12,288 MB | GDDR7 | 28,430 | 250 W | not recorded by source | — | 31,100 |
| nvidia | GeForce RTX 5060 Ti | 2025 | 8,192 MB | GDDR7 | 22,183 | 180 W | not recorded by source | — | 21,900 |
| nvidia | GeForce RTX 5060 Ti 16 GB | 2025 | 16,384 MB | GDDR7 | 22,183 | 180 W | not recorded by source | — | 21,900 |
| nvidia | GeForce RTX 5060 | 2025 | 8,192 MB | GDDR7 | 17,510 | 145 W | not recorded by source | — | 21,900 |
| nvidia | GeForce RTX 5050 | 2025 | 8,192 MB | GDDR6 | 11,863 | 130 W | not recorded by source | — | 16,900 |
| nvidia | GeForce RTX 5090 D V2 | 2025 | 24,576 MB | GDDR7 | 87,780 | 575 W | not recorded by source | — | 92,200 |