RTX 4090 vs. RTX 3090: In terms of performance, the old graphics card can no longer hold a candle to its successor. The comparison reveals whether the RTX 3090 is still worthwhile.
On October 12, 2022, Nvidia will have its RTX 4090 brought on the market. The graphics card manufacturer’s recommended retail price for its current top model was 1949 euros, and three months later the card is around 100 euros cheaper. The predecessor of the current top model, which is now two years old, is almost 450 euros cheaper: the RTX 3090. It is clear that the new card logically calculates faster than its predecessor. How fast and whether this is worth the extra charge remains questionable.
RTX 4090 vs. RTX 3090: The technical specifications
So let’s compare the two graphics cards with each other based on the following technical specifications:
- NVIDIA CUDA Compute Units: CUDA is short for Compute Unified Device Architecture and is the programming interface through which programs interact with the graphics processor. The higher the number of processing units, the more calculation the processor can do and the faster the card.
- Boost clock in gigahertz (GHz): The higher, the faster the graphics card calculates its tasks
- Base clock (GHz): see boost timing
- Storage: This is where the graphics processor stores tasks that it needs to access quickly. Logical: the larger the memory, the more tasks the processor can store in memory
- Memory interface width: The highway between GPU and memory. The wider it is, the more tasks (bit) can be transferred in parallel
- Raytracing Cores: Ray tracing is a technology designed to realistically calculate light rays in the game. That eats up a lot of performance, which is why Nvidia gives its cards extra cores that take on this task
- Tensor calculation units: Are special calculation cores that accelerate the calculation of matrix multiplications. This is important for machine learning (artificial intelligence, or AI for short)
- NVIDIA DLSS: Stands for Deep Learning Super Sampling and is an artificial intelligence that calculates images. DLSS uses the tensor processing units mentioned above for this
- power consumption
RTX 4090 | RTX 3090 | |
CUDA compute units | 16384 | 10496 |
Boost Clock (GHz) | 2.52 | 1.70 |
Base Clock (GHz) | 2.23 | 1.40 |
Storage | 24GB GDDR6X | 24GB GDDR6X |
Memory interface width | 384 bits | 384 bits |
Ray Tracing Cores | 128 of the 3rd generation | 82 of the 2nd generation |
Tensor calculation units | 512 of the 4th generation | 328 of the 3rd generation |
NVIDIA DLSS | 3 | 2.3 |
PCIe | Gen 4 | Gen 4 |
DisplayPort | 1.4a | 1.4a |
power consumption | 450 watts | 350 watts |
RTX 4090 vs. RTX 3090: More cores, more performance
The table clearly shows what Nvidia has been working on for the last two years: more processing units that clock higher. The RTX 4090 offers almost exactly 56 percent more CUDA cores compared to the RTX 3090. This is mainly due to the production of the chips: Nvidia provided its own RTX 3090 still using the 8 nanometer process, the manufacturer has halved it to 4 nanometers in the new generation.
Simply means that Nvidia can theoretically build twice the number of computing units on the same chip size. However, that would be too much of a good thing, because twice the number of computing units would also result in more power consumption and waste heat, which a heat sink has to dissipate.
Nevertheless, the leap between the generations is considerable, because the 56 percent more computing units on the RTX 4090 According to the manufacturer, they only consume around 28 percent more power than the RTX 3090. In addition to the number of cores, their clock rate is also noticeable: In comparison, it increases by 48 percent in boost mode and by 59 percent in the base clock.
We can state: Within two years, Nvidia has managed to implement 56 percent more computing cores that calculate 48 percent faster under load. Unsurprisingly, it is already clear that the RTX 4090 is at least one step ahead of its predecessor.
Improved ray tracing and artificial learning
The second nose ahead is somewhat hidden in the details and probably also explains the advertised almost fourfold performance advantage. On the one hand, Nvidia has improved its “Deep Learning Super Sampling”, i.e. the learning artificial intelligence for calculating images. Its third generation is reserved for the RTX 4000 series, the RTX 3090 still works with the second generation of DLSS.
Compared to DLSS 2, the third-generation artificial intelligence calculates every other frame, like notebookcheck.com reports, which doubles the frame rate of the AI. The RTX 4090 also has 512 fourth-generation Tensor cores. That’s 52 percent more cores than the Tensor cores RTX 3090.
Different technology, same game: The manufacturer also made significant improvements to the ray tracing performance. In the RTX 4090 128 third-generation ray tracing cores calculate the light rays. In the RTX 3090 there are only 82 of the second generation. The RTX 4090 was given its creators 52 percent more computing units for ray tracing.
But all of this only makes up for the advertised fourfold increase in performance when combined. So the software has to be able to enjoy all the innovations at all and so far only one game can do that: “Cyberpunk 2077”. It’s the only game on the market to date that has implemented Nvidia’s Raytracing Overdrive technology. how PCGamesHardware reports, it is about the efficient use of several features of the RTX 4090 such as DLSS 3 and shader execution reordering.
The weakness of the RTX 4090
It remains inexplicable why Nvidia is denying its new graphics card PCIe 5.0 and DisplayPort 2.0. As a reminder, the motherboard’s PCIe lanes are the computer’s information superhighway. The newer the generation, the higher the data throughput of the individual PCIe lanes. This is important for people who like to use a lot of NVM memory. A PCIe 5.0 graphics card only occupies eight PCIe lanes on a PCIe 5.0 mainboard for 31.52 gigabytes (GB) of data throughput.
This leaves more lanes for storage. PCIe 4.0 cards, on the other hand, need 16 lanes for the same data throughput. Nor may that be irrelevant to most consumers. But if you want to use the card for the next five to six years, you could be annoyed with the future upgrade of the mainboard and processor.
The same applies to DisplayPort 2.0. According to Nvidia, DisplayPort 1.4a is sufficient for most consumers because the interface enables 4K at 120 Hertz. That’s also true, but there are already 4K monitors that work with a 240 Hertz refresh rate. They are still extremely expensive, but here too it is true that more and more gamers will be able to afford them in the next one to two years. Competitor AMD is already using DisplayPort 2.0 for its RDNA 3 graphics cards. And that’s what a buyer can expect from a graphics card for almost 2000 euros.
RTX 4090 vs RTX 3090 in games
Linus TechTips took a closer look at the performance of the RTX 4090 in games. Unsurprisingly, the RTX 4090 significantly outperforms its predecessor. The calculates without DLLS and raytracing RTX 4090 62 percent more frames per second in Cyberpunk 2077, 82 percent more in Forza Horizon 5, 75 percent more in Assassin’s Creed: Valhalla, and 80 percent more in Shadows of the Tomb Raider. Only “Counter Strike – Global Offensive” is an exception. Here the RTX 4090 is only 6.5 percent above its predecessor.
However, this is probably not due to the card, but to the fact that the processor simply cannot calculate more frames per second. If gamers turn on ray tracing in the games, the RTX 4090 trumps them RTX 3090 in “Cyberpunk” even by 91 percent. The situation is different with ray tracing and DLLS: Here the lead in the game melts to “only” 77 percent.
Conclusion: RTX 4090 vs. RTX 3090
The gap between the generations is huge. Gamers can get one from Amazon for 1399 euros RTX 3090. The cheapest one costs 1838 euros RTX 4090 at the shipping giant. This results in a price difference of around 32 percent. With the compared technical specifications and performance data, the price of 1399 euros for the RTX 3090 in 2023 makes no sense. Here you should definitely reach for the RTX 4090.
If you don’t want to spend more than 1400 euros, you shouldn’t use the RTX 3090 either. Here is one RTX 4080 on, which also has more power than the RTX 3090 and costs around 1429 euros at Amazon. The RTX 4080 is even a spark faster than the RTX 3090 Ti for 1599 euros. If Nvidia’s RTX 3000 series does not drop significantly in price, you should definitely grab the new generation (as of January 31, 2023).
Sources: nvidia.com, youtube.com/LinusTechtips, PCGameshardware.de
You might also be interested in:
This article contains so-called affiliate links. Further information are available here.