RTX 5090 prototype allegedly has 24,576 CUDA cores and 800W TDP — two 16-pin connectors present
Is this the future RTX 5090 Ti or RTX Titan Blackwell?

The Nvidia GeForce RTX 5090, announced earlier this year and coming January 30, is poised to be one of the best graphics cards. However, hardware sleuth HXL has unearthed what appears to be an early prototype. Still, given the rumored specifications, it could very well become a GeForce RTX 5090 Ti or RTX Titan Blackwell.
The Chiphell forum user claims the graphics card is a GeForce RTX 5090 engineering card manufactured from July 15 to 21, 2024. Apparently, it's a prototype Nvidia's AIC partners used to design their custom offerings in the early days. The user didn't state how he got his hands on the prototype but was willing to pay good money if someone could supply him with the Nvidia GeForce 570.12 driver, which seems like the only driver that supports the prototype.
According to the specifications, the Blackwell graphics card utilizes fully functional GB202 silicon with 24,576 CUDA cores, compared to the RTX 5090’s 21,760 CUDA cores—13% more. The mysterious prototype also has slightly higher clock speeds (2,100 MHz base clock and 2,514 MHz boost clock) than the RTX 5090 (2,017 MHz base clock and 2,407 MHz boost clock).
The memory configuration reportedly remains unchanged between the two graphics cards, with 32GB of GDDR7 memory across a 512-bit memory interface. However, the prototype seemingly has 32 Gbps chips, pushing the memory bandwidth to 2 TB/s. For comparison, the RTX 5090 leverages 28 Gbps chips, capping the bandwidth at 1.79 TB/s.
Specifications | RTX 5090 prototype / RTX 5090 Ti / RTX Titan* | RTX 5090 | RTX 4090 Ti* | RTX 4090 | RTX 3090 Ti | RTX 3090 |
---|---|---|---|---|---|---|
Architecture | Blackwell | Blackwell | Ada Lovelace | Ada Lovelace | Ampere | Ampere |
GPU | GB202 | GB202 | AD102 | AD102 | GA102 | GA102 |
CUDA Cores | 24,576 | 21,760 | 18,176 | 16,384 | 10,752 | 10,496 |
Base Clock | 2,100 MHz | 2,017 MHz | ? | 2,235 MHz | 1,560 MHz | 1,395 MHz |
Boost Clock | 2,514 MHz | 2,407 MHz | ? | 2,520 MHz | 1,860 MHz | 1,695 MHz |
Memory Size | 32 GB GDDR7 | 32 GB GDDR7 | 24 GB GDDR6X | 24 GB GDDR6X | 24 GB GDDR6X | 24 GB GDDR6X |
Memory Bandwidth | 2 TB/s | 1.79 TB/s | ? | 1.01 TB/s | 1.01 TB/s | 936.2 GB/s |
TDP | 800W | 575W | 600W | 450W | 450W | 350W |
Launch Date | ? | 06 Jan 2025 | ? | 20 Sep 2022 | 27 Jan 2022 | 01 Sep 2020 |
Launch Price | ? | $1,999 | ? | $1,599 | $1,999 | $1,499 |
*Specifications are unconfirmed.
However, the most eye-popping specification from the GeForce RTX 5090 prototype is the alleged 800W TDP claims, 39% higher than the RTX 5090. The owner stated that it utilizes two 12V-2x6 power connectors, which makes sense given the higher TDP. It’s doubtful that Nvidia (or anyone else, for that matter) will release a consumer-grade GPU that will demand more than 600W of power. The RTX 5090 is already pushing it with its 575W TDP.
Even with the RTX 40 series, there were rumors of an RTX 4090 Ti or RTX Titan Ada. An RTX 4090 prototype has been discovered and disassembled, sporting a massive blow-through heat sink and unique PCB configuration. There has been talk that it could have been a testbed to see how Nvidia could cool an RTX 4090 Ti, but that hasn’t been proven. While it is indeed possible that the company is thinking about producing an RTX 5090 Ti, it could also be just an experiment, and the company is waiting to see if there will be enough demand for a more powerful GPU than the RTX 5090 before going forward with production.
If this prototype hits the retail market, it will be the most powerful consumer-grade GPU Nvidia releases for the RTX 50-series generation and will likely be priced accordingly. The RTX 5090 is already very expensive, at $1,999 MSRP, so something better will likely start at $2,500.
The geopolitical tension between the U.S. and China means Nvidia cannot sell its most powerful GPUs in the Chinese market. This will likely dampen Nvidia’s hopes for an RTX 5090 Ti or RTX Titan Blackwell, if there is one, as China is one of its biggest markets for these expensive GPUs. These rumored specifications are just that; unless we see a massive demand for a more powerful GPU, we won’t likely see something better than an RTX 5090.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Jowi Morales is a tech enthusiast with years of experience working in the industry. He’s been writing with several tech publications since 2021, where he’s been interested in tech hardware and consumer electronics.
-
helper800
I mean, if Nvidia can do 575w with a 2 slot blow through with just a wicking vapour chamber, i think it stands to reason a 3 slot version of the card could do 800w, but who knows. The gains in cooling efficiency of good ol' HSF design has been blowing my mind in the last couple years.Lucky_SLS said:800W, we touching exotic cooling levels of power draw here folks! -
Gaidax At that point the card should be coming with a dedicated power supply and external cooler.Reply -
Lucky_SLS
At the same time, I am also blown away by the power draw of high end cards. Gone are days when you had 350W as the top end limit.helper800 said:I mean, if Nvidia can do 575w with a 2 slot blow through with just a wicking vapour chamber, i think it stands to reason a 3 slot version of the card could do 800w, but who knows. The gains in cooling efficiency of good ol' HSF design have been blowing my mind in the last couple years.
Did hwbusters do any transient load test for 1250w + PSU with 2 HVPWR connectors? -
spongiemaster Maybe the prototype uses 800W, there's no way a 10% boost to CUDA cores and slight memory speed bump would require an additional 225W or 40% increase in TDP for a retail product.Reply