Hands-On And Unboxing NVIDIA'S Beastly GeForce RTX 4090
NVIDIA announced its new GeForce RTX 40-Series GPUs at its GTC Developer’s Conference two short weeks ago. We have been privy to a deep dive of the underlying Ada Lovelace architecture and have gleaned more details from AIB partners, but now we have a couple cards in hand to share with the camera.
We are still waiting for the embargo to lift before we can talk performance and offer our full thoughts, but we are able to unbox the cards. Up for show, we have MSI’s Suprim Liquid X GeForce RTX 4090 as well as NVIDIA’s own GeForce RTX 4090 Founder’s Edition. The Founder’s Edition card arrived with some unique packaging. It looks like an ordinary box from above, but the sides are actually hollowed out. The top of the box swings open on a hinge to reveal the beastly card propped at an angle, ready to be admired.
NVIDIA includes a 12VHPWR adapter cable which is stowed below the GPU. This adapter fuses four 8-pin PCIe power cables into the single 12-pin connector used by the card. Each individual 8-pin PCIe power cable can deliver 150 watts, which sum together to offer up to 600 watts in total.
The overall aesthetics of the GeForce RTX 4090 are very similar to its RTX 3090 predecessor and it retains the passthrough fan to improve airflow. Both are 3-slot designs, though the RTX 4090 is slightly taller by about a quarter of an inch. Other small details include a redesigned fan which has seven blades, each with a bit more surface area, to the RTX 3090’s nine blades. The RTX 4090’s fan cuts slightly into the silvered frame of the card as well.
There are more differences than similarities on the spec sheet. The GeForce RTX 4090 and RTX 3090 both use 24GB of GDDR6X video memory with a 384-bit interface. Everything else about the RTX 4090 dwarfs its predecessor except for the process tech. The RTX 4090 is built using TSMC’s smaller 4N node where the RTX 3090 uses Samsung’s 8N technology.
The GeForce RTX 3090 featured 10,752 CUDA, 84 RT (2nd Gen), and 336 Tensor (3rd Gen) cores which the RTX 4090 bumps up to 18,432 CUDA, 144 RT (3rd Gen), and 576 Tensor (4th Gen) cores in turn. The smaller process also allows the new card to push bigger clocks and can boost up to 2.5 GHz where the RTX 3090 topped out at 1.9 GHz. This is supported by the transistor count which has nearly tripled from 28 billion to 76 billion. The GeForce RTX 4090 also gains valuable (dual) hardware AV1 encoding capabilities.
MSI’s Suprim Liquid X uses a very different setup, though it retains the 12VHPWR connector and display output options. As the name implies, this card uses liquid cooling via a closed loop with radiator. This allows the card itself to be much shorter (280mm vs FE’s 304mm) and occupy just two slots of width. As a trade-off, the radiator also needs to find a home within the chassis, plus routing the tubing between it and the card. The radiator uses dual 120mm fans to cool the chip while the card’s radial onboard fan is used for the VRM.
If you want to know more about how the GeForce RTX 4090 performs, and how much the liquid cooling may help, make sure to check out our full review when it arrives.