We had been talking about the new generation of NVIDIA graphics cards for months, and today they are finally here. In this analysis we are going to show you what is currently the most powerful gaming graphics card in the world, the NVIDIA GeForce RTX 4090 which carries the new Ada Lovelace architecture, loaded with new features and new technologies such as DLSS 3.0.
Although yesterday we were able to see some analysis of the NVIDIA Founders Edition model, it is not until today that we can finally show you the custom models of this new graphics card, and in this case, the manufacturer has not sent us a graphic own but a model customized by Gigabyte, specifically the Gigabyte GeForce RTX 4090 Gaming OC 24G, which has an optimized heatsink and slightly higher operating frequencies.
NVIDIA RTX 4090, technical characteristics
As we just mentioned, in our possession we have a customized version of Gigabyte, but even so, except for the operating speed (which we will see the concrete value later), the hardware is identical to the reference model. For this reason, in the following table, you can find the complete technical specifications of both the RTX 4090 and the RTX 3090 Ti, which is the model that comes to replace the NVIDIA GPU family.
Comparison:
Tests | GeForce RTX 3090 Ti | GeForce RTX 4090 |
---|---|---|
Fabrication process | Samsung 8nm | TSMC 4nm |
transistors | 28.200 million | 76.300 million |
form factor | Triple slot | Triple slot |
video outputs | 3x DisplayPort 1x HDMI |
3x DisplayPort 1x HDMI |
power connectors | 1×16 pin (3×8 pin adapter) | 1 x 16 pin (4 x 8 pin adapter) |
Graphics clusters/texture processor | 7 / 42 | 11 / 64 |
YE | 84 | 128 |
CUDA cores | 10,752 | 16,384 |
Tensor Cores | 336 (3rd Gen) | 512 (4th Gen) |
RT Cores | 84 (2nd Gen) | 128 (3rd Gen) |
Texture Units | 336 | 512 |
ROPs | 112 | 176 |
Speed (Boost) | 1860MHz | 2520MHz |
memory speed | 10,500MHz | 10.501MHz |
Memory Data Rate | 21Gbps | 21Gbps |
L2 cache | 6144KB | 73728KB |
total video memory | 24GB GDDR6X | 24GB GDDR6X |
memory interface | 384bit | 384bit |
Memory Bandwidth | 1,008 GB/s | 1,008 GB/s |
Texture Rate (bilinear) | 625 GT/s | 1290.2 GT/s |
Recommended power supply | 850W | 850W |
Total Graphics Power (TGP) | 450W | 450W |
Maximum temperature | 93ºC | 90ºC |
PCIe interface | 4.0 | 4.0 |
Compute Scores:
We will go into detail soon to talk about the new Ada Lovelace architecture, but to begin with, here is also the block diagram that illustrates the figures that we have put in the table. We can see that the new generation AD102 GPU incorporates, in its complete form (since the RTX 4080 has a “layered” form), a total of 12 Graphics Processing Clusters (GPCs), 72 Texture Processing Clusters (TPCs), 144 Streaming Multiprocessors (SMs), and a 384-bit memory interface with 12 32-bit memory controllers.
Opinions:
One of the most debated topics about this new generation of graphics cards is their size, and that is that they are huge. The custom Gigabyte model that NVIDIA has sent us for this review measures no more and no less than 35 centimetres long, 13.6 centimetres deep at its widest part, and 7.7 centimetres wide, completely occupying 3 slots.
Although Gigabyte does not specify the exact weight of this graphics card on the packaging, we have weighed it and we are dealing with 2,170 grams of graphics card. Of course, the manufacturer includes a support bracket to support the weight, a type of support that does not use the bottom of the box to support the weight, but we must screw it into one of the base plate supports (later we will see it in detail), which will alleviate the pressure on the PCIe socket of the motherboard and prevent the graphics PCB itself from bending.
Controversy:
Another controversial issue is food; NVIDIA recommends a power supply of at least 850W to power this graphics card, even though its TGP is 450 watts. NVIDIA has used the new 16-pin connector that some sources already incorporate, but in any case, the curious thing is the adapter included in the graphics card, which goes from being 3 8-pin connectors to 4, thus increasing the potential power consumed.
Finally, the other widely debated topic about this new generation is the price, and that is that this RTX 4090 has an NVIDIA RRP of no less than 1,959 Euros, although it is true that these custom models are somewhat more expensive, and this specific model from Gigabyte is already listed in stores for 2,049.90 Euros. Indeed, we are talking about a graphics card that alone costs more than a mid-high-end gaming PC, so let’s see if this is worth it or not.
What do you want 24 GB of memory for?
Of course, 24 GB of dedicated VRAM seems like an outrage for a gaming-oriented graphics card, as much as NVIDIA says that it is designed to play at 8K resolution… and indeed it is. However, for many workloads in creative environments, the size of the graphics memory directly influences the data sets that can be viewed and analyzed at the same time, so the 24GB of GDDR6X memory in the RTX 4090 will not they are exaggerated in this case (for gaming, it certainly seems so).
This massive amount of VRAM allows content creators to work with large data sets using a large number of materials and textures, as well as workflows using multiple GPU-intensive applications without worrying about running out of memory or experiencing slowdowns.
Why so Needy?
For GPU rendering, memory is especially important, as 3D scenes that are larger than available memory need to reduce their complexity by using lower-resolution texture data or reducing the fidelity of the meshes. Another possibility is to split a scene into separate elements that are also rendered separately and then recomposed, which requires additional work and affects productivity.
Therefore, a larger video memory size reduces the need to make these types of compromises, as it allows users to work with and render larger scenes in real-time. Of course, as we mentioned before, for gaming it does not make much sense.