September will mark two years since NVIDIA launched its Ampere lineup with the RTX 3090 leading the pack. While impressive from a technical standpoint, the generation was mired in availability issues and ridiculous price markups. These are finally going away, just in time for the RTX 4090 to drop.
What do we know about the 4090? Now that the GTC 2022 event finally happened – quite a lot! We’ve compiled all relevant news & credible rumors in one place so it’s easier for you to keep up.
GPU manufacturers try to keep their cards’ tech specs under wraps until pre-launch events, so next to nothing about the 4090 RTX was known officially until September 20th. The only two bits of information we knew of for the longest time is that the generation’s microarchitecture is called Ada Lovelace and is based on TSMC’s 5nm production process.
Luckily, several leakers have shed some light on a handful of specifics that turned out to be true. Here’s a table comparing the RTX 3090 with what’s speculated about its successor for a better overview:
|RTX 3090||RTX 4090|
|Manufacturing process||Samsung 8nm||TSMC 5nm|
|Memory||24GB GDDR6X||24GB GDDR6X|
|TFLOPs||35.6||Up to 90|
There’s a lot to unpack here, so let’s start with the new die. Moving to 5nm is sure to significantly bump the transistor count up from the 3090’s 28 million. No one has suggested by how much yet, though.
The increases to base & boost clocks are even more indicative of a large expected performance bump. Initial leaks claimed that the boost clock would go up to 2500MHz, which credible leakers Greymon55 and Kopite7kimi later revised to around 2750MHz.
It turns out that the first wave of speculations was more credible as NVIDIA has since confirmed a 2235MHz base clock & 2520MHz boost clock. CEO Huang also claims that their internal lab testing produced overclocks as high as 3GHz. Whether or not partner cards will reach limits anywhere close to that is still to be seen.
Then there’s the matter of CUDA cores. Kopite7kimi’s original estimate was 16,128, revised to 16,384.
The cache info came about as the result of a data breach, though we can’t be sure it’s credible. If it is, it would be one of the most interesting tidbits we know. The fifteen-fold increase in L2 cache was likely inspired by AMD’s attempts to mitigate the memory bandwidth shortages 6000-series cards would otherwise have.
For example, the RX 6900 XT has around half the memory bandwidth of the RTX 3090, yet 128MB of its Infinity Cache smooths out communication between its first two cache levels and the memory. AMD also successfully applied a similar approach to the new Ryzen 7 5800X3D, so It’s likely that NVIDIA paid close attention and will capitalize on this development.
Calculations would suggest that the new card can achieve around 66 TFLOPs, but the leaker Greymon55 has kept insisting for months that it’s a whopping 90. TFLOPs aren’t as clear-cut of an indicator as they once were, but such a drastic generational increase still warrants attention.
So, what can we infer about the 4090’s real-world performance from all of this? It will undoubtedly be a major step up from the 3090, which follows the trend set by Ampere and Turing. We can only speculate on how large the increase will be.
Another of Kopite7Kimi’s more cryptic tweets says doubling the 3090’s output isn’t hard to accomplish. If that’s really the case, the 4090 will be an awesome upgrade for content creation, video encoding, and 3D modeling. On the gaming front, you can expect it to run even the most demanding or poorly optimized titles at or close to three-digit framerates in 4K Ultra.
NVIDIA’s own sources now peg the RTX 4090 as a true generational leap forward, citing 2-4x the performance of the RTX 3090Ti. The larger multiplier refers to the third generation of raytracing cores & substantial DLSS 3.0 improvements that look like they may finally turn rayrtacing from a novelty into a standard game makers will start paying more attention to.
There’s no such thing as a free lunch, and it looks like 4090 owners will have to pay through their power bills. Kopite7Kimi states that the card’s TBP is 450W, which is in line with the RTX 3090 Ti’s requirements. However, they also hint that the power limit for AD102 will be a jaw-dropping 800W.
It’s not clear whether this refers to the 4090 or a potential Ti version supposedly built on the same die. We’re inclined to believe the latter, so a 600W limit for the regular card sounds more reasonable. Even so, that’s still a sizeable increase that will prompt enthusiast PC builders to reconsider their PSU choices. You’ll want a 1,000+W PSU to run such a system comfortably, and even that’s not certain to be enough until high-end Raptor Lake requirements become known.
A more dubious rumor that surfaced on Chiphell Forums claims how the RTX 4090 will be available in 450W and 600W variants. Does this mean the card will be available in two versions? Will the lower power limit refer to the Founder’s Edition and the higher be reserved for the most advanced partner cards?
The latest news suggest that 450W will indeed be the base TGP, with 660W as the max TGP some models may be able to reach by enabling the boost in their BIOS.
There’s no room for guesswork anymore – the RTX 4090 will cost a hefty $1599 at launch. That’s $100 more than the RTX 3090’s launch price was, which we can attribute to inflation and the general trend of slight price increases for electronics goods.
RTX 4090 Release Date
It’s official – the RTX 4090 will be launching on October 12th! There are ačready plenty of leaks from partners like Zotac and Gigabyte that show the card’s most advanced editions will have massive coolers that will take up four slots. It might be time to shop around for a larger case.
Much about the RTX 4090 is no longer a mystery, and the info that’s out there suggests we’re in for a treat! More facts will be available as the release date draws nearer. We’ll be sure to keep the overview fresh so you won’t miss any updates.