NVIDIA released its Ampere series in September of last year, with the RTX 3090 at the head of the pack. Although technically advanced, the generation suffered from poor availability and exorbitant price markups. These are finally disappearing, just in time for the release of the RTX 4090.
The 4090: What do we know about it? Not much yet, but enough to pique the interests of both serious designers and avid gamers alike! For your convenience, we have gathered all reliable rumors and leaks in one location.
Nearly nothing about the 4090 RTX is officially known because GPU manufacturers work hard to keep their cards' technical specifications a secret until pre-launch events. The only two details we are certain of are the name of the generation's microarchitecture, Ada Lovelace, and the fact that it is built on TSMC's 5nm manufacturing process.
Fortunately, a few leakers have provided some further details. For a clearer picture, consider the following table that contrasts the RTX 3090 with what is rumored to be its replacement:
|RTX 3090||RTX 4090|
|Manufacturing process||Samsung 8nm||TSMC 5nm|
|Memory||24GB GDDR6X||24GB GDDR6X|
|TFLOPs||35.6||Up to 90|
Let's start with the new die since there is a lot to discuss here. From the 28 million transistors in the 3090, the transition to 5nm will undoubtedly result in a large increase. However, no one has yet said by how much.
The base and boost clock improvements are even more suggestive of a significant projected performance increase. According to early leaks, the boost clock would reach a maximum of 2500MHz; however, reliable leakers Greymon55 and Kopite7kimi later amended this to a maximum of 2750MHz.
The most recent assessment from Videocardz indicates that the early boost clock rumors were accurate. They write that a chiphell forum leaker has confirmed that the 4090's boost frequency is 2520MHz, however third-party partners are sure to put their own spin on the information.
The issue of CUDA cores is another. The estimate by Kopite7kimi was 16,128, however it was later changed to 16,384. Regardless of the outcome, the 60% upgrade is a very reliable sign of outstanding performance.
We can't be certain that the cache information is reliable, because it was obtained as a result of a data breach. If so, it would be one of the most intriguing facts we are aware of. It is possible that AMD's efforts to reduce the memory bandwidth problems that the 6000-series cards would otherwise experience were what led to the fifteen-fold increase in L2 cache.
For instance, the RX 6900 XT smooths the communication between its first two cache levels and the memory while having around half the memory bandwidth of the RTX 3090 (128MB of Infinity Cache). It's possible that NVIDIA paid close attention and will take advantage of this breakthrough because AMD successfully used a similar strategy with the new Ryzen 7 5800X3D.
Calculations indicate that the new card can produce 66 TFLOPs or such, but the leaker Greymon55 has insisted for months that the actual number is 90. TFLOPs aren't as precise an indicator as they once were, but a generational rise this large still requires consideration.
What can we conclude from all of this regarding the 4090's performance in the real world? It will unquestionably be a significant improvement over the 3090, which continues the trend started by Ampere and Turing. On how much the rise will be, we can only make educated guesses.
Another of Kopite7Kimi's more cryptic tweets claims it is simple to double the 3090's output. If that's the case, the 4090 will be a fantastic boost for 3D modeling, video encoding, and content creation. When it comes to gaming, you can anticipate it to run even the most difficult or inefficiently optimized games at or near three-digit framerates in 4K Ultra.
Kopite7Kimi had something more substantial to say a few days after our piece went live. In the 3D Mark Extreme benchmark, they assert that the RTX 4090 can get a score of more than 19,000 points. That is a third more than what the RTX 3090 Ti achieved when paired with a HEDT CPU in the hall of fame. The RTX 3090 achieves 10 to 12K scores when combined with a processor like the i9-12900K, indicating that the 4090 might indeed double that performance.
There is no such thing as a free lunch, therefore it appears that owners of 4090 units will have to pay for it with higher electricity costs. According to Kopite7Kimi, the card's TBP is 450W, which is in line with the specifications for the RTX 3090 Ti. However, they also imply that the AD102 will have a mind-blowing 800W power limit.
It's unclear if this is referring to the 4090 or a possible Ti variant that was allegedly created using the same die. The latter is more likely, hence a 600W limit for the standard card makes more sense. However, even with that substantial increase, enthusiast PC builders will be prompted to reevaluate their PSU selections. To operate such a system comfortably, you'll need a 1,000+W PSU, and even that isn't certain to be sufficient until high-end Raptor Lake requirements are established.
On Chiphell Forums, a more dubious rumor suggests that the RTX 4090 will come in 450W and 600W versions. Does this imply that there will be two variations of the card? Will the greater power limit be saved for the most sophisticated partner cards and the lower limit apply to the Founder's Edition?
According to recent reports, the base TGP will certainly be 450W, and some models may be able to reach the maximum TGP of 660W by turning on the boost in their BIOS.
How much the RTX 4090 will set you back when it is released is a mystery for now. Since of the turbulence in the cryptocurrency market, many scalpers are finding it difficult to sell used 3000-series cards because consumers are hesitant to acquire them, even at a discount. The debut of the 4090 shouldn't be as troubled by shortages and price increases as its predecessors were because the motivation for mining has been much reduced.
The RTX 4090 should be simpler to get at MSRP, but what will that price be? NVIDIA will need to strike a compromise between maintaining its competitiveness and paying for the extra obligations imposed by Ada Lovelace. They will need to account for the far more powerful cooling required to handle, say, a 450W TGP GPU.
We can reasonably predict that the RTX 4090 won't be less expensive than the 3090, so budget at least $1,500. According to the most prevalent MSRP ideas, the cost is between $1,699 and $1,799. We wouldn't be shocked if the final cost increased even further.
RTX 4090 Release Date
There is no need to continue speculating about when the 4090 will be released because most current indications point to October.
On NVIDIA's ambitions, Wccftech provides the most recent information. The RTX 4090 will debut at the GTC developer conference on September 20th, according to one of their sources. Its actual launch should happen sometime in October after that.
There is additional data to support this. Just rumor, Tiah. The AMP Extreme model from Zotac, which is typically one of the most expensive devices, has appeared on the same website in multiple leaks. To say that the card in the image is large would be an understatement. People, get your PC cases ready since the enormous cooler will need four PCIe slots to be installed!
Via Videocardz, more news also surfaced at the same time. They verified that Gigabyte has submitted 11 different RTX 4090 models to the EEC. Three of them have water cooling, and the remaining models in the AORUS, Eagle, Windforce, and Gaming lines are air-cooled variations.
While there are still many unknowns surrounding the NVIDIA 4090, the information that is currently available indicates that we're in for a treat! As the publication date approaches, more factual conjecture will be provided. To ensure you don't miss any developments, we'll make sure to keep the overview up to date.