The new graphics that were presented Nvidia They have managed to arouse great enthusiasm among many people, while for others it is a type of hardware that has simply made some updates compared to previous versions, without too significant changes. But the real way to know what has changed between generations is to compare everything they offer, and that is why today we present to you a summary of the technical specifications of the main models so you can see what the difference between the RTX40 and the RTX50.
Many of the changes that Nvidia did for the new graphics that it has prepared are related to the ability to offer superior performance both for gaming and for AI, in this aspect the brand has tried to create hardware that serves general use in both cases . But in addition to this, one of the things that surprised us the most is that in addition to arriving with an increase in performance, contrary to what we expected, the base price indicated by the company is lower than we expected, even reducing the release price compared to the previous generation.
RTX 4070 vs. RTX 5070
Starting with the comparison we are going to make, the first thing we need to do is that the RTX5070 It presents a series of improvements at the cost of considerably higher consumption than the previous one, which is repeated in practically all graphics of the new generation. The new model also comes with a fairly notable update in all aspects, also featuring one of the most important changes, the move to GDDR7 which offers much greater memory bandwidth:
| RTX4070 | RTX5070 | |
|---|---|---|
| Clock | 1920 MHz 2475 MHz |
2160MHz 2510 MHz |
| Shading units | 5888 | 6144 |
| TMU | 184 | 192 |
| ROP | 64 | 96 |
| Calculation units | 46 | 48 |
| Tensor kernels | 184 | 192 |
| Ray tracing cores | 46 | 48 |
| L1 cache | 128 KB per cluster | 128 KB per cluster |
| L2 cache | 36 MB shared | 40 MB shared |
| TFLOPS (FP32) | 29.2 | 30.8 |
| Interface | PCIe 4.0×16 | PCIe 5.0×16 |
| TGP | 200W | 250W |
| Process | 5nm | 4nm |
| Memory | 12 GB GDDR6X 1313 MHz | 12 GB GDDR7 1750 MHz |
| Memory bandwidth | 504.2 GB/s | 672.2 GB/s |
| Basic introductory price | 599$ | 549$ |
RTX 4080 vs. RTX 5080
Browsing the high-end range that NVIDIA offers us, we can see that the first tests carried out in 3D Mark show quite good data, since the RTX5080 It has 13% more score reaching 34201 points compared to the 30163 of the RTX 4080. In addition, the price reduction is the most notable, since in the previous model it was practically 50 $, while this model has a price 200$ less than the previous generation when it was released:
| RTX4080 | RTX5080 | |
|---|---|---|
| Clock | 2205 MHz 2505 MHz |
2300 MHz 2620 MHz |
| Shading units | 9728 | 10752 |
| TMU | 304 | 336 |
| ROP | 112 | 168 |
| Calculation units | 76 | 84 |
| Tensor kernels | 304 | 336 |
| Ray tracing cores | 76 | 84 |
| L1 cache | 128 KB per cluster | 128 KB per cluster |
| L2 cache | 64 MB shared | 64 MB shared |
| TFLOPS (FP32) | 48.7 | 56.3 |
| Interface | PCIe 4.0×16 | PCIe 5.0×16 |
| TGP | 320W | 360W |
| Process | 5nm | 4nm |
| Memory | 16GB GDDR6X 1400MHz | 16 GB GDDR7 2400 MHz |
| Memory bandwidth | 716.8 GB/s | 960 GB/s |
| Basic introductory price | 1199$ | 999$ |
RTX 4090 vs. RTX 5090
Finally we have the two flagships of each generation, in the first tests which were carried out on a RTX5090 They show that it offers a score 16% higher than the RTX 4090 in 3D Mark by offering 49122 points compared to 42169 for the previous generation. But that's not the only thing to point out about this model, as it doesn't improve in as many aspects as one might imagine.
It offers a much lower clock frequency with a much higher price, practically 500 $ difference more, but which can be influenced by the change of memory, since by offering 32 GB of GDDR7, the bandwidth is 77 % higher than the previous generation, the biggest improvement of this generation compared to the two models we saw previously which were around 33-34%:
| RTX4090 | RTX5090 | |
|---|---|---|
| Clock | 2235 MHz 2520 MHz |
2010 MHz 2400 MHz |
| Shading units | 16384 | 21760 |
| TMU | 512 | 680 |
| ROP | 176 | 340 |
| Calculation units | 128 | 170 |
| Tensor kernels | 512 | 680 |
| Ray tracing cores | 128 | 170 |
| L1 cache | 128 KB per cluster | 128 KB per cluster |
| L2 cache | 72 MB shared | 88 MB shared |
| TFLOPS (FP32) | 82.6 | 104.5 |
| Interface | PCIe 4.0×16 | PCIe 5.0×16 |
| TGP | 450W | 575W |
| Process | 5nm | 4nm |
| Memory | 24 GB GDDR6X 2625 MHz | 32 GB GDDR7 |
| Memory bandwidth | 1010 GB/s | 1792 GB/s |
| Basic introductory price | 1599$ | 1999$ |
Which graph to choose in 2025?
After seeing all the changes with which the new NVIDIA graphics arrive, it is clear that the new generation offers a series of advantages compared to the previous one, the most notable being the price reduction compared to the launch of the RTX 40 in the middle of the year. range and high-end versions, since the RTX 5090 is $500 more expensive in base. In this regard, the recommendations that we can give are quite simple, if we have to choose between choosing an RTX 40 around 50-100 $ cheaper because it is a previous generation or an RTX 50, it would be better to choose the latter, especially if we are talking about the RTX5080since it is the one which should have the smallest price difference compared to the RTX4080.
Table of Contents
