The next graphics card will be the GeForce RTX 5090 top of the range from NVIDIA for the general consumer market, and all indications are that it will represent a significant leap over the GeForce RTX 4090, both in terms of raw power in rasterization and ray tracing, and in terms of efficiency, thanks to the improvements that the Blackwell architecture and the use of GDDR7 memory will bring.

Over the past few months, various leaks have been produced that have left us with interesting data about the GeForce RTX 5090, since last year we published a special where we told you everything we knew so far about the said graphics card. Time has passed and I have new information which allows me to outline the possible specs of the GeForce RTX 5090 much better, as well as other important things like its possible performance and release date.

In this article, I’ll share with you everything I know so far about the GeForce RTX 5090. Keep in mind that Some of the dates you’ll see below are confirmedand that the others still have rumor status, so they may not come true. However, you can be sure that all this information is reliable and makes a lot of sense.

The information is current with the latest rumors and leaks that have surfaced so far that contain important changes that would affect its performance and release date. Nothing is confirmed yet, but you know how it is, eventually a lot of these leaks and rumors turn out to be true.

GeForce RTX 5090: possible specifications

AI generated image.

• Graphics core GB202 made on TSMC’s 4nm node.
• 170 SM units.
• 21,760 shaders at speeds between 2.9 GHz and 3.2 GHz.
• 680 texturing units.
• 176 raster units.
• 170 fourth generation RT cores.
• 680 fifth generation tensor cores.
• 384-bit or 512-bit bus.
• 24 GB or 32 GB of GDDR7 memory at 32 Gbps (bandwidth 1536 GB/s 384-bit bus, 2048 GB/s 512-bit bus).
• 96 MB L2 cache.

The difference in specifications compared to the GeForce RTX 4090 is so big that it can be seen with the naked eye and in a very simple way. If these specifications are confirmed, the number of shaders will increase from 16,384 to 21,760and of course we will also have more tensor cores and RT cores in GeForce RTX 5090 as well as more texture and raster units.

The increase in bandwidth compared to the GeForce RTX 4090 will be gigantic because it reaches 1001 GB/s. Higher bandwidth will improve the performance of GeForce RTX 5090 at 4K resolution and higher and also very positive impact when working with ray tracingas this technology relies heavily on memory bandwidth.

The increase in L2 cache also deserves special mention will change from 72 MB to 96 MB. This will further reduce the dependence of the GeForce RTX 5090 on the memory subsystem and increase the number of hits when calling said cache even at 4K resolution, which will lead to a significant increase in performance below this resolution.

As for consumption, I think the most likely is the GeForce RTX 5090 be placed in the 450 watt rangewhich is exactly the same TGP that we have in the GeForce RTX 4090. This would translate into higher power per watt consumed, and therefore higher energy efficiency, thanks to the increase in performance of the new NVIDIA.

However, the latest information suggests that this graphics card could have consumption up to 600 watts, which is a significant increase, since we are talking about an extra 150 watts. We’ll have to wait and see what happens, but the source that reported this new information is one of the most reliable and also says that this graphics card will have 32 GB of GDDR7 memory.

The GeForce RTX 5090 will use the Blackwell architecture, so it will have with all the improvements it will introduce at the level of shaders, tensor cores and RT cores. We can expect a significant increase in terms of raw performance, new features and optimizations applied to ray tracing and AI acceleration, as well as a new optical flow accelerator that could improve performance and image generation quality.

As for the production node, it is most likely GeForce RTX 5090 it is manufactured by TSMC’s 4nm process. This means that the jump will be less pronounced than with the previous generation, since the GeForce RTX 40 is manufactured in a 5 nm node. However, it is clear that this jump of nodes, even if it is not so significant, will make it possible to reduce the size of transistors, increase their density and improve consumption values.

GeForce RTX 5090 benchmark and performance

In absolute numbers, if everything I just told you is true, the GeForce RTX 5090 will have a raw performance of 130.56 TFLOPs in FP32 if the GPU is running at a stable average of 3 GHz in turbo mode, a value that is well beyond the 82 ,58. TFLOPs that the GeForce RTX 4090 achieves at FP32. That means that The GeForce RTX 5090 would be up to 58% more powerful than the GeForce RTX 4090 at FP32.

The difference in games could be less because, as I told you on many occasions, the raw performance values ​​in FP32 of the graphics card It doesn’t directly equate to gaming performance. Personally, I think we’ll see an improvement of between 30% and 40% in rasterization compared to the GeForce RTX 4090, and that in ray tracing the improvement will be much larger, probably between 80% and 100%, depending on each specific game. .

Configuration with 384-bit bus and 24 GB of GDDR7 memory at 32 Gbps would offer worse performance on another with a 512-bit bus and 32 GB of GDDR7 memory, not so much because of the greater amount of graphics memory, which would not have any type of impact in games, unless we move in 8K with ray tracing and maximum qualities, but rather due to more available bandwidth.

This graphics card will use the new Blackwell architecture, which means we can expect major improvements which will affect the performance and its efficiency values. With the launch of the GeForce RTX 40, NVIDIA introduced the third generation of RT cores, DLSS Frame Generation and Ray Reconstruction technologies, and we also saw the debut of SER, which changed the order of processing threads to improve performance at the shader level.

With this new generation of NVIDIA, we can expect a performance fourth generation RT cores, enhancements and optimizations focused on ray tracing, and NVIDIA may also take the opportunity to enhance the DLSS platform with new technology or updates that improve existing ones, either in terms of performance, image quality, or both.

GeForce RTX 5090 would have no direct equivalent in this generation and would represent a generational leap similar to what we saw with the GeForce RTX 4090 compared to the GeForce RTX 3090 Ti in rasterization, but the improvement would be much greater in ray tracing.

Recommended processor and power supply

The question of power does not raise many doubts. If the GeForce RTX 5090 has a TGP of 450 watts, the power requirement will be the same as the GeForce RTX 4090, which means that the power supply will be enough. 850 watts if we use a low-power processor, e.g. Ryzen 7 7800X3D, a 1000 watts if we use a processor with higher consumption, e.g. Intel Core i9-14900K.

However, if TGP rises to 600 watts We’re going to need a much more powerful resource, that’s for sure. According to the previous example, it would be convenient to calculate with a 1000 watt source If we are going to accompany the GeForce RTX 5090 with a processor with moderate consumption and higher we would need if we are going to use a processor like Intel Core i9-14900K.

The power values ​​I gave you are above the minimum recommended value, that is, we could reduce it a bit, but its power value would already be too tight and it would mean that it would end up working with certain values. too much almost 100% capacity.

Consider that a computer configured with a Core i9-14900K, 32GB of DDR5 memory, two M.2 SSDs, six 120mm fans, an AIO liquid cooling kit, and some attached gaming peripherals, including a keyboard and mouse, already It can achieve a load consumption of 840 watts.

The source must of course be of good quality and must have the necessary power connectors. In this case, it is best to choose a model that has a 16-pin connector so you don’t have to use adapterswhich end up not looking good at all aesthetically and would be cumbersome because in this case we would have to connect four heads.

As for the processor, it’s a complicated matter. The GeForce RTX 5090 will be an extremely powerful graphics card, and that means will likely suffer from bottlenecks even at 4K, something that has already happened in rasterization with the GeForce RTX 4090 in certain games. In order to take full advantage of it, we will need a very powerful processor.

I think it won’t make sense to accompany the GeForce RTX 5090 nothing worse than Ryzen 7 7800X3D or Intel Core i9-13900Keven though we use it to play in 4K. We will have bottlenecks with lower CPUs, and if we enable DLSS Super Resolution, this bottleneck will only get worse as the base rendering resolution will be reduced and thus the CPU’s impact on performance will increase.

With ray tracing on and 4K resolution the problem of bottlenecks is changing because the load that the GPU has to support increases enormously, and this greatly reduces the real impact that the processor has on performance. In this case, more modest CPUs can handle the job without a problem, but for an optimal experience in general, it will be ideal to start from the level I indicated above.

Possible release date and price of GeForce RTX 5090

AI generated image.

We believed that the launch of the GeForce RTX 5090 would happen at the end of this year, specifically between October and Novemberjust a little bit after NVIDIA’s GeForce RTX 5080 and that both graphics cards will be announced at the same time in September of this year, but this prediction will not come true in the end.

It’s still possible that both models will be introduced by the end of 2024, but the latest information I had the chance to consult says that they will leave in early 2025 and that NVIDIA will use CES to present its new product. generation of graphics cards. It makes sense and right now seems to be the most credible date.

As for the selling price, the truth is that I have many doubts. NVIDIA may have decided to keep the price of the GeForce RTX 5090 at the same level as the GeForce RTX 4090, which would bring us to approximately 2000 euros. There is also a possibility that the selling price will increase, in fact NVIDIA would have a very easy time in this generation since it will not have a competitor from AMD in the high line.

The GeForce RTX 5090 will be an elite product, just like the GeForce RTX 4090, so the question of price in this case is something that is not as important as with the lower models due to the type of user it is aimed at. Due to changing the node and using GDDR7 memory I think it makes sense to consider the GeForce RTX 5090 slightly more expensive than the GeForce RTX 4090, but nothing is confirmed yet, so both options are open.

AI generated cover image.