GeForce RTX 4090 Founders Edition
We have the GeForce RTX 4090 Founders Edition for review today, this video card will retail for $1,599 MSRP.
The GeForce RTX 4090 is not the full AD102 spec, but it is close, one GPC is disabled. Therefore it has a total of 11 GPCs, 64 TPCs, and 128 SMs. What this translates to is 16,384 CUDA Cores, versus the GeForce RTX 3090’s 10,496 and the RTX 3090 Ti’s 10,752.
The GeForce RTX 4090 has 176 ROPs, 512 Texture Units, 128 3rd generation RT Cores, and 512 4th generation Tensor Cores. For comparison, the GeForce RTX 3090 had 112 ROPs, 328 Texture Units, 82 2nd generation RT Cores, and 328 3rd generation Tensor Cores. The GeForce RTX 3090 Ti had 112 ROPs, 336 Texture Units, 84 2nd generation RT Cores, and 336 3rd generation Tensor Cores.
All cache size capacities are up on the GeForce RTX 4090 versus the previous generation, L1, L2, and register sizes. The GeForce RTX 4090 has a total L1 cache size of 16,384 KB, a total L2 cache size of 73,728 KB, and a total register file size of 32,768 KB. This compares to the GeForce RTX 3090 which has a total L1 cache size of 10,496 KB, a total L2 cache size of 6,144 KB, and a total register file size of 20,992 KB. The GeForce RTX 3090 Ti has a total L1 cache size of 10,752 KB, a total L2 cache size of 6,144 KB, and a total register size of 21,504 KB. As you can see, the L2 cache size increase is very large with the GeForce RTX 4090.
The new TSMC 4N process allows for much higher clock speeds, the GeForce RTX 4090 has a boost clock of 2520MHz. This is much higher than the GeForce RTX 3090 which had a boost clock of 1695MHz, and the RTX 3090 Ti at 1860MHz. The GeForce RTX 4090 has the same memory configuration as the GeForce RTX 3090 Ti with 24GB of GDDR6X memory at 21GHz on a 384-bit memory bus. This provides 1008GB/s of memory bandwidth. This is faster than the GeForce RTX 3090 which has 24GB of GDDR6X memory at 19.5GHz on a 384-bit bus providing 936GB/s of memory bandwidth.
The TGP of the GeForce RTX 4090 Founders Edition is 450W, the same as the TGP on the GeForce RTX 3090 Ti Founders Edition. The GeForce RTX 3090 by comparison has a 350W TGP.
Generational Changes
One of the many worries with this generation was power delivery and power spikes, or transient power. NVIDIA has been aware of these issues and has made it a point to address them with the GeForce RT 40 series and GeForce RTX 4090 video card. The new GeForce RTX 4090 uses a PCIe Gen5 power connector, allowing for scalable power demands on a single cable, with the data sense pins as part of the PCIe 5 ATX 3.0 spec. It also utilizes a 23-phase power supply design.
NVIDIA has paid attention to the power transient currents to design power delivery that is more consistent and tightly timed, with smoother curves and fewer transients. The entire GPU has been moved north to provide a better impedance balance between phases. Two PCB layers have been added to enhance power efficiency. The GDDR6X signal integrity has been improved.
GeForce RTX 4090 Founders Edition Pictures and Design
In crafting the Founders Edition version of this video card, NVIDIA utilized the design language from the previous generation, however, it is much evolved. The same dual-sided, opposing fan format is being utilized for the video card. The PCB is also compact and shortened, while the heatsink mass extends beyond. The heatsink is built from an extruded aluminum alloy and the frame has anodizing treatments throughout. The fins are made from a 99% pure aluminum alloy that is lightweight and rigid.
The memory pedestal has been improved to provide more even GPU to heatsink contact improving pressure from the memory modules, and the TIM thickness has been increased from 1.5mm to 2mm to provide better memory cooling. The vapor chamber and heat pipes have also been improved in many ways. The number of heat pipes has increased from 4 to 6 on the east side and a bigger vapor chamber is being used. The internal design of the vapor chamber has also been improved.
For the new RTX 4090 coolers, NVIDIA has an all-new design tailored to increase airflow. The airflow has been increased to 80 cubic feet per minute, which is 20% more airflow than the RTX 3090. The fan sizes have increased on the RTX 4090 to 116mm versus 110mm on the RTX 3090 and have been swapped to fluid dynamic bearings and uses counter-rotating fans for a cooler and quiet operation. The blades of the fans are made from glass fiber-reinforced plastic.
The GeForce RTX 4090 Founders Edition actually shrinks in length down to 12″ inches, from the RTX 3090’s 12.3″ inches. However, the height (thickness) increases from a 2.7-slot to a 3-slot height. The actual size is 304mm in length, 137mm in width, and 61mm in height. For comparison, the GeForce RTX 3090 measures 313mm in length, 138mm in width and 53mm in height. It has 3x DisplayPort 1.4a and 1x HDMI 2.1.
The official TDP for the GeForce RTX 4090 Founders Edition is 450W, and the recommended PSU for the Founders Edition is 850W. In the box is a 16-pin (video card side) to 4x 8-pin PCIe power connectors adapter. However, you do only need to plugin 3x 8-pin PCIe Power Connectors to operate the video card at default settings, no overclocking. Additionally, you can plug in the 4th 8-pin PCIe power connector to enable more Wattage for overclocking (600W). If you have one of the new ATX 3.0 PCIe 5 power supplies, you can utilize the native PCIe 5 16-pin single cable for power, either 450W or 600W.
GeForce RTX 4090 Founders Edition Specs
GeForce RTX 4090 | GeForce RTX 3090 Ti | GeForce RTX 3090 | |
---|---|---|---|
GPU Codename | AD102 | GA102 | GA102 |
Architecture | Ada Lovelace | Ampere | Ampere |
Process | TSMC 4N NVIDIA | Samsung 8N NVIDIA | Samsung 8N NVIDIA |
Die Size | 608.5mm2 | 628.4mm2 | 628.4mm2 |
Transistors | 76.3 Billion | 28.3 Billion | 28.3 Billion |
L1 Data Cache | 16384KB | 10752KB | 10496KB |
L2 Cache Size | 73728KB | 6144KB | 6144KB |
Register File Size | 32768KB | 21504KB | 20992KB |
GPCs | 11 | 7 | 7 |
TPCs | 64 | 42 | 41 |
SMs | 128 | 84 | 82 |
CUDA Cores | 16384 | 10752 | 10496 |
RT Cores | 128 (3rd Gen) | 84 (2nd Gen) | 82 (2nd Gen) |
Tensor Cores | 512 (4th Gen) | 336 (3rd Gen) | 328 (3rd Gen) |
ROPs | 176 | 112 | 112 |
Texture Units | 512 | 336 | 328 |
GPU Boost | 2520MHz | 1860MHz | 1695MHz |
VRAM | 24GB GDDR6X | 24GB GDDR6X | 24GB GDDR6X |
Memory Interface | 384-bit | 384-bit | 384-bit |
Memory Clock | 21GHz | 21GHz | 19.5GHz |
Memory Bandwidth | 1008GB/s | 1008GB/s | 936GB/s |
TGP | 450W | 450W | 350W |