NVIDIA RTX 5080/RTX 5090 ~1.5x Faster than the RTX 4080/RTX 4090 at Same Settings (W/O DLSS 4)

NVIDIA has officially announced its GeForce RTX 50 series graphics cards, including the RTX 5070, RTX 5070 Ti, RTX 5080, and RTX 5090. There’s a ton of content to sift through, and in this one, we’ll present the sheer hardware performance gains. In other words, the framerates achieved without DLSS 4 or “Multi-Frame Generation.” Before we get to that, let’s go through the specifications and the pricing.

NVIDIA RTX 5090|RTX 5080|RTX 5070 Specs

The Blackwell gaming GPUs leverage the TSMC N4P process technology, a refinement of the 5nm node used by the preceding Ada Lovelace family. A first important points to note here:

• Barring the RTX 5090, the shader counts and in-turn the single-precision compute throughput (FP32) is largely similar across the two generations:
  • The RTX 5090 packs 21,760 shaders across 170 SMs, up from 16,384 cores and 128 SMs on the RTX 4090. The SP throughput is only 27% higher on the former versus the latter.
  • The RTX 5080 integrates 10,752 FP32 cores across 80 SMs versus 10,240 cores and 66 SMs on the RTX 4080 Super. We get 56.3 TFLOPs for the former, and 52.22 TFLOPs for the latter. A mere 8% bump.
  • The RTX 5070 Ti incorporates 8,960 shaders, up from 8,448 on the RTX 4070 Ti. Consequently, we get 43.9 TFLOPs, up from 44.10 TFLOPs on the former. Hardly substantial.
  • The RTX 5070 features fewer shader cores than its predecessor: 6,144 vs. 7,168 on the RTX 4070 Super. A compute throughput of 30.9 TFLOPs, lower than it’s predecessor’s 35.48.

Of course, these are merely theortical performance estimates, but the message is clear. NVIDIA is heavily relying on software-side innovations (DLSS 4) for much of its generational gains.

Interestingly, the boost clocks have stayed the same or slightly regressed, despite the increased power budget. I reckon the thermals are to blame for that. We might see AIB variants with higher freqencies. The power envelope has expanded on account of the higher core count and memory on the top-end SKUs, with the RTX 5090 approaching the 600W mark and the RTX 5080 hitting 360W under load.

As far as the hardware is concerned, the switch to GDDR7 memory is the primary upgrade. This results in substantial bandwidth uplifts for the higher-end SKUs, most notably the RTX 5090: 1008 GB/s->1792 GB/s. The massive 512-bit bus of the flagship is of note (a first in consumer GPUs), while the rest of the lineup retains the existing width.

NVIDIA RTX 5090 vs 4090|RTX 5080 vs 4080 S: Without DLSS 4

NVIDIA’s official marketing material contains only two titles running DLSS 3 (or no DLSS) on both the RTX 5090/5080 and the RTX 4090/4080. These include Far Cry 6 (native) and Plague Tale: Requiem (DLSS3), where the RTX 50 GPUs are 40-50% faster than their 40-series counterparts.

Of course, these are rough estimates, but don’t expect gains of over 50% in most like-to-like scenarios without DLSS 4.

NVIDIA DLSS 4 Supported Games: 75 at Debut

Fortunately, the GeForce RTX 5080 and 5090 are launching with Day 0 support for DLSS 4 across 75 games and apps. These are as follows:

Gamers won’t have to wait for developers to add support for DLSS 4 “Multi-Frame Generation” in these titles. It can be enabled via the NVIDIA app, along with the enhnaced frame generation algorithm that reduces memory usage and improves performance.

DLSS Transformer Model for RTX 40|RTX 30|RTX 20

For existing RTX GPU owners, the update brings enhanced upscaling and ray reconstruction using the DLSS Transformer Model which improves visual fidelity with better temporal stability, less ghosting, and increased detail in motion. This model leverages the same architecture powering AI models like ChatGPT, Gemini, and Flux.

This update should be available by the month’s end in beta, and graducally rolled out across more titles.