This week the GeForce RTX 4090 went on sale, and much to nobody’s surprise they sold out instantly. All eyes now turn to the upcoming RTX 4080. It’ll be available in both 16GB and 12GB varietals, but its performance has so far been a mystery, aside from a vague “2 -4X faster than 3080 Ti” that was teased at the unveiling. Plus, that uplift number is in ray tracing, we assume.
Still, it’d be nice to see some actual numbers, and Nvidia has delivered. We know, they’re from Nvidia, but this is all we have right now. The numbers are designed to show the benefits of DLSS 3, which is a 40-series exclusive. However, the company only shared numbers for four games, highlighting the reality that not many games support Nvidia’s latest technology yet. You can still enjoy its incredible rasterization performance while you wait, though.
The numbers paint a good picture of the gains afforded by DLSS 3. However, the 4080 benchmarks are more interesting than generic performance claims. As you might recall, both 4080s use a different die than the RTX 4090. This is a departure from Ampere, where both the 3080 and 3090 used the “big” GA102 die. For Ada Lovelace, only the 4090 gets the big AD102 chip, with the 16GB 4080 getting a smaller AD103 die and the 12GB receiving an even smaller AD104 die. Both GPUs have radically different specs too, despite sharing the same moniker. The 12GB version has fewer CUDA cores than its big brother and a narrower 192-bit memory bus. This will undoubtedly lead to market confusion.
MSFS DLSS 3 benchmarks, click to expand. (Image: Nvidia)
The benchmarks show that the 12GB version is a bit slower than the 16GB version. This will likely lead to people referring to the 12GB version as the “4070” for the foreseeable future. The 16GB version is also surprisingly close in frame rate to the 4090, despite costing $500 less. Overall, in MSFS the 12GB card is equivalent to the RTX 3090 Ti without DLSS. However, with DLSS 3 enabled it can run over 120fps at 4K, which is impressive. That’s also 19 percent slower than the 16GB version. If you end up getting a 40-series GPU, mark Oct. 7 on your calendar. That’s when the DLSS 3 update arrives, but it will only be offered in a beta version at first. Follow the instructions in the Nvidia blog link above to see how you can join.
A Plague Tale: Requiem RTX 40 series benchmarks, click to expand. (Image: Nvidia)
A Plague Tale: Requiem comes out on Oct. 18, and these are non-ray-traced benchmarks. Like with MSFS, we see an 18 percent delta between the two RTX 4080 cards in DLSS 3. Once again, the RTX 3090 Ti is also faster than the 12GB card in rasterization. We also see a massive gulf between the RTX 4090 and the other GPUs, highlighting the discrepancy in horsepower between all three GPUs.
Marvel’s Spider-Man Remastered DLSS 3 benchmarks, click to expand. (Image: Nvidia)
In Spider-Man, we see similar numbers as the previous titles, as Nvidia has segmented its GPUs quite well. The performance delta between the 12GB and 16GB cards is once again 19 percent. What’s most surprising is the gains for the newer GPUs using DLSS 3, as it does indeed more than double performance for all three of them. The update for this game is available right now with the latest GeForce driver.
F1 22 DLSS 3 benchmarks, click to expand. (Image: Nvidia)
Finally, we come to the racing game, F1 22. According to Nvidia, DLSS 3 makes this game go from unplayable to playable in certain scenarios, at least at 4K. That is the case with the 4080 12GB, which crawls to just 46fps without DLSS 3. With it enabled, it’s able to achieve 113fps according to the company. In this game, we see the biggest gap between the RTX 4080s at 23 percent.
These charts are interesting as a lot of 30-series card owners are curious about the new 4080s’ performance. They seem to be equivalent to the high-end 30 series cards, albeit with a similar or slightly higher price tag. These charts also leave out what the 40-series can do on DLSS 2 games, which is the majority of titles right now. Nvidia certainly makes a strong argument for DLSS 3, with the huge caveat that it only works in specific games. It’s also notable that these are 4K benchmarks, as most gamers are still running 1080p or 1440p. It’s likely Nvidia only released 4K benchmarks because that’s where its new cards truly shine. The RTX 4090 benchmarks showed a lot of games can be bottlenecked by the CPU at lower resolutions, even at 1440p. That includes high-end CPUs too; these GPUs are just that powerful.
We’ll all be waiting for official, third-party benchmarks for the RTX 4080s, which are due in November. That’s also when AMD will unveil its RDNA3 GPUs, so it’s shaping up to be an exciting month in the GPU world. Nvidia has also notably left yawning gaps between its GPUs this time. Those gaps will surely accommodate an RTX 4080 Ti, if necessary. The same goes for the RTX 4090 Ti, which is assuredly waiting in the wings and will use the full AD101 die. The current RTX 4090 is missing over 2,000 CUDA cores from the full spec. Either way, the GPU war is certainly back on, just like the olden days.
Now Read:
- Nvidia Unveils RTX 4090 Ray Tracing Benchmarks
- Nvidia Announces Jetson Orin Nano, Updates Isaac Nova Orin Dev Kit
- Bombshell: EVGA Terminates Relationship with Nvidia
No comments:
Post a Comment