With the new RTX 4090 exceeding expectations both in terms of features and performance, we thought that it would be interesting to see how this new GPU compares to the last generation’s flagship GPU, the RTX 3090 Ti. In this guide, we will look at the various features and hardware developments that come with the new generation of Nvidia cards and find out how exactly they stack up against the older cards.
Now, though we can be certain about the fact that RTX 4090 will be faster than the previous generation card, the question that remains is by how much? Is the value proposition of the RTX 4090 undeniable, or is there something everyone is missing? Let’s find out.
RTX 4090 vs RTX 3090 Ti: Specs Sheet
Before we go on to explore all the new features and details of the RTX 4090, let’s first have a general overlook at the specifications. The first thing you will notice with the RTX 4090is that there’s a significant bump in almost all aspects of the card, from the CUDA cores to the clock speeds.
Furthermore, these specs bumps are a generational upgrade from the 3000 series and not a minor spec bump like what we saw a few years ago with the coming of Nvidia’s Turing (2000-series) GPUs. That said, here’s how the RTX 4090 compares to RTX 3090 Ti on paper:
|Specification||RTX 4090||RTX 3090 Ti|
|Manufacturing Process||4nm TSMC||8nm Samsung|
|Transistor Count||75 Billion||28 Billion|
|GPU Boost Clock||2520 MHz||1695 MHz|
|VRAM||24 GB GDDR6X||24 GB GDDR6X|
|Launch Price||$1599||$1999 (at launch)|
RTX 4090 vs RTX 3090 Ti: Architecture & Cores
Now that the general specs overview is done, it’s time to explore the biggest change in the new GPU, and that’s the architecture.
The RTX 4090 is built on the new Ada Lovelace architecture. It is based on TSMC’s 4nm process node, which is a significant upgrade from the 8nm Samsung process that Nvidia was using for 3000-series cards. This move towards a smaller manufacturing node has allowed Nvidia to cram more transistors (75 billion of them) on a similar size die. This is an almost 165% upgrade over the RTX 3090 Ti, which is bonkers.
More transistors have allowed NVidia to pack even more graphics cores on the GPU. And this is evident, as the new RTX 4090 has almost 62% more CUDA cores compared to the RTX 3090. Manufacturing improvements have also led to a massive increase in the GPU boost clocks. They have been raised to 2520 MHz on the 4090, from the paltry maximum 1650 MHz boost clock of the RTX 3090 Ti.
The other big change at the level of GPU architecture is the small bump that the new 400-series card has seen in both RT and Tensor cores. The RTX 4090 has 84 RT cores against 3090 Ti’s 82 cores and 336 Tensor cores compared to 328 on the RTX 3090 Ti. This might not seem like a big jump in terms of raw numbers, but as Nvidia has claimed, these cores are different from the older ones and can’t objectively be compared.
Well, the RT cores in the RTX 4000 series cards are third-generation RT cores that offer twice the ray-tracing performance of “Ampere” RT cores. The new RT cores also introduce two new hardware components: an Opacity Micromap Engine and a Displaced Micro-Mesh Engine. Nvidia claims that these new components rapidly accelerate the calculation of alpha textures, common in elements such as video game foliage, providing for a better experience while gaming.
Coming to the tensor cores, the new cards include the 4th generation Tensor cores that come with exclusive features like the Optical Flow accelerator, the magical tool that makes DLSS 3 possible (more on this later).
The only core component of the GPU that hasn’t seen any massive change is video memory. The RTX 4090 card comes with the same 24 GB GDDR6X memory on a 384-bit bus that the RTX 3090 Ti shipped with. However, some reports suggest that the RTX 4090 might be using the full bandwidth of the GDDR6X, which is around 24 GB/s, rather than the 21 GB/s offered by the RTX 3090 Ti.
RTX 4090 vs RTX 3090 Ti: Design Comparison
Now coming to the realm of design, at first sight, one could easily think that the form factor of the new RTX 4090 is the same as that of the 3090 Ti. But if you look at the design carefully, you would come to realize that the basic design principle (the dual axial flow-through) is the same as the RTX 3090, but there are subtle differences, like in the way that the fans are designed.
The new fans on the RTX 4090 look to have been slightly extended, and because of that, they seem to be protruding into the outer railing. Even the fan blade design has changed as it now features a flatter blade. Nvidia claims that the new fan design will improve the cooling performance by 10%. But more importantly, the fan redesign seems to be a way to improve the airflow over the memory modules, which were considered inadequately cooled on the 3000 series.
Nvidia has also made minor changes to the color scheme as the newer RTX 4090 comes with a gilded gold metal railing, while the RTX 3090 had silver ones. All in all, there is no significant design change when it comes to the cooling setup, but this was expected as the 3090 Ti (Founder’s Edition) was already the best in its class.
The cooling solution might be similar, but the part of the card that has drastically changed is the power connector. With RTX 3090 and 3080 GPUs, Nvidia came up with a proprietary 12-pin power pin called Molex Micro fit 3. It was a revolutionary standard that had the capability of delivering up to 600 watts of power. But with the release of the RTX 4090, it seems that the 12-pin standard is dead. The new RTX 4090 and 4080 cards come with a single 16-pin PCIe 5.0 power cable that’s radically different from the old solution. In terms of power delivery, though, it’s the same – with a peak power capability of 600W.
RTX 4090 vs RTX 3090 Ti: What Are the New Features?
The new RTX 4090 graphics card doesn’t just come with core hardware upgrades but also includes a plethora of newly added features. Just like every generation, Nvidia has been hard at work to deliver us the most cutting-edge features. Nvidia made it clear that if they wanted to reach a new level of performance, just architectural upgrades weren’t going to cut it.
With the RTX 4090, the 4x performance upgrade over the 3090 Ti card is a combination of architectural and non-architectural changes in which features play an important part. In the section below, we will look at some of our favorite new features, ones that we think will have a huge impact on performance going forward.
The biggest feature upgrade that comes with the new RTX 4000 series cards is the inclusion of DLSS 3, which is a totally new integration of Nvidia’s prior AI-enhanced super sampling techniques – DLSS and DLSS 2.0. This feature essentially allows the GPU to generate entirely new frames instead of just pixels.
This is possible because the new cards have the 4th-Gen Tensor cores that come with a new hardware component called the “Optical Flow Accelerator“. How does it work? In an official blog post, Nvidia mentions that DLSS 3 works by taking information from both sequential frames and optical flow field (captured by optical accelerator) to generate new frames that are high-quality and accurate. Further, it claims that DLSS 3, when combined with older DLSS features like DLSS Super Resolution and NVIDIA Reflex, can provide up to 4x improvements in frame rate and up to 2x improvements in latency compared to native resolution rendering.
The second big feature that we see in the RT cores is the inclusion of Shader-execution reordering. The working of this mechanism is, well, complicated. But to simplify it, one can think of it as a more efficient scheduling system that makes the process of calculating ray-tracing data much more efficient.
Usually, ray-tracing is a heavily single-threaded task, which causes the ray-tracing data to wait, at times, before it can be calculated. With Shader execution, though, this process becomes multi-threaded (sort of) and gets calculated at a much faster pace. Nvidia claims that this feature gives ray-tracing games a 25% bump in performance when compared to Ampere cards.
AV1 Encoding Support
The final feature we will be looking at wasn’t talked about at the Nvidia GTC conference but is heavily referenced in the blog post. This new feature is support for AV1 encoding. Now, at first glance, support for a brand new codec might not seem like a big deal, but we think it is.
AV1 is seen as the holy grail for content creators because it has the ability to capture more details at a much lower bandwidth. Nvidia claims that AV1 encoding has the ability to improve efficiency by 40%. This means streams that support AV1 codec will have significantly more detail than current streams that use H.264 codecs like NVENC.
RTX 4090 vs RTX 3090 Ti: Performance
Now that we’ve gone through the drudgery of looking at the details of the specifications and features, it’s finally time to move on to the most important question – how does the RTX 4090 compare in relation to the RTX 3090 Ti? Well, Nvidia claims that it’s a generational improvement over the previous card, but how much of it is marketing, and how much of it is the truth? Let’s have a look:
At the GTC conference, Nvidia made the recurring claim that the new 4000-series cards will be twice as performant as the corresponding 3000-series card in classical-rasterized games and up to four times as fast in ray-tracing games.
Now, at the outset of it, this claim seems plausible. The new cards don’t only come with a massive upgrade in terms of core specifications like core counts and clock speeds, but also include newer technological features such as shader-reordering and DLSS 3. To illustrate this upgrade, Nvidia showed gameplay footage from Flight Simulator and Cyberpunk, where we saw that the new RTX 4090 getting twice as many frames in Flight Simulator when compared to the 3090 Ti and up to 4x frames in games like Cyberpunk 2077.
So, can we use this data to abstract the claim that RTX 4090 is going to be four times as fast as 3090 Ti? Now, this is where things get a bit complicated. On stage at the conference, Nvidia didn’t show us enough games for us to positively conclude that this performance is representative of all games.
There are two issues that we have with the gameplay comparisons that Nvidia has provided everyone. One, for the demo, Nvidia has picked games that have historically performed better on their system like Cyberpunk 2077 and Microsoft Flight Simulator. And two, that they have enabled all of the new features that come with the new RTX cards – DLSS 3, DLSS Super resolution, and ray tracing.
This comparison is unfair, as the number of games that support all of these features is very small. A handful, maybe. The comparison shows us how the cards would perform in the best of conditions, but this is hardly representative of general gaming. If we want to objectively look at the performance upgrade over the 3090 Ti, we have to go into comparisons that are not vetted by Nvidia’s PR team.
RTX 4090 vs RTX 3090 Ti: Benchmark (Leaked)
Now, this is where leaked details about the performance of the RTX 4090 come into play. We haven’t really seen many leaks for the newest RTX 4090 graphics card, but there was one that occurred a few weeks ago. The leak is not without its issues, but it might give us some idea of what to expect from the RTX 4090 in terms of performance in non-optimized games.
The leak that we are referring to came from a now-deleted post on the Chiphell forum. It didn’t specifically say that the screencap showed the benchmark results for an Nvidia RTX 4090, but the mysterious graphic card scored a mind-boggling score of 20,192 in 3Dmark Time Spy Extreme, which according to many graphics card experts lies in the same ball-park as the performance improvement that Nvidia has claimed for their RTX 4090.
If rumors are to be believed and this card is indeed an RTX 4090, it would make it 78% faster than the RTX 3090 Ti (11,400) and 90% faster than its predecessor, the RTX 3090. This, one could argue, makes sense as the RTX 4090 should theoretically be 62% faster on the basis of core count and architectural improvements alone. And if one puts core clock differences to this – 78% doesn’t seem that far off.
But like all leaks, there are many questions that still remain unanswered. The graphics card, for one, was not running at the advertised peak clock speeds of 2520 MHz (as mentioned in the specs sheet above) but was running at 3015 MHz. This difference in clock speeds has made some wonder if this is indeed the RTX 4090 or the rumored 4090 Ti.
If it’s the 4090 Ti, then a direct apples-to-apples comparison with the 3090 Ti would not hold. But what if it’s simply an overclocked 4090? If so, then we can just decrease the score by 17%, and we might have the answer we are looking for – a Timespy score of 16,760. It would make the RTX 4090 46% faster than the RTX 3090 Ti. However, this method is a bit problematic, as the supposed RTX 4090 was running on unoptimized drivers.
RTX 4090 vs RTX 3090 Ti: Pricing
Finally, another aspect in which the GeForce RTX 4090 and RTX 3090 Ti differ is the release price. Nvidia unveiled the RTX 3090 Ti GPU on April 2022 for a retail price of $1,999. This was a huge gamble by Nvidia at the time, as it was nearly $500 more expensive than the RTX 3090. And surprisingly, that was only for the founder’s edition card produced by Nvidia. Third-party cards from brands like ASUS, MSI, or Zotac could reach upwards of $2,100, which made it one of the most expensive consumer graphics cards ever released.
The RTX 4090, on the other hand, has been announced at a more palatable (but still absurd) launch price of $1,599. This price is more akin to the launch price of the RTX 3090, which was released at $1,499. The price drop might make one feel like Nvidia is moving in the right direction, but the thing that one has to keep in mind is that the release of the 3090 Ti was already met with a lot of disdain among GPU buyers.
The outrageous pricing of the 3090 Ti sparked a debate about whether Nvidia was using its market monopoly to twist the arm of high-end GPU buyers. The launch price of the RTX 4090 at $1,599 might seem like a bargain in the context of the last few years, but only a couple of generations ago, something like this would have been considered outrageous.
However, a good takeaway from the pricing, as argued by DigitalTrends, is the fact that Nvidia hasn’t drastically increased the price of its high-end cards this year. This could suggest that 4000-series cards will not be any more expensive than the 3000-series cards. Lower-end cards like the RTX 4060 and RTX 4070 could actually make for very good value.