NVIDIA described in detail the features of the reference version of the new generation GeForce RTX 4090 Foundations Edition of the flagship video map. Although the overall design of the map was very similar to GeForce RTX 3090 Ti Foundations Edition on the Ampere architecture, the manufacturer made a number of changes.
The external metallic frame of the GeForce RTX 4090 Foundations Edition was given a more pronounced metallic shade, and the video card radiator unit was modified to improve the air flow between the two cooling system fans on each side of the accelerator.
The fans themselves are also new here. According to the NVIDIA, she's gone over 50 different fan options until she's stopped at one particular model that provides up to 20% more air flow than the GeForce RTX 3090 Ti Foundations Edition fans. In addition to the radiator and fans, the company has changed the volatilizer plate of the cooling system. It has become larger and more efficient to take heat from the GPU and the memory chips surrounding it. The heat from the power subsystem elements of the card is transferred to the support plate. The manufacturer has also revised the distribution of heat-conducting tubes for better heat distribution.
The new flagship booster GeForce RTX 4090 Foundations Edition uses the same compact printing fee as the reference versions of the older models GeForce RTX of the 30th series. The card is fed through a new 12+4-point interface of 12VHPWR of the PCIE 5.0 standard. It is able to transmit to the accelerator up to 600 W.NVIDIA that Ada Lovelace does not require a power unit that meets ATX 3.0 and PCI-Express 5.0.
Although the declared energy consumption of GeForce RTX 4090 is 450 W, the power limit can be increased to 600 W. This may be useful for the manual acceleration of the video map. The video map is sustained by a 23-phase power subsystem. NVIDIA has changed the power subsystem architecture to ensure stable voltage management. This is done specifically to minimize the surges or variations in the voltages observed in the previous generation ' s high-performance video maps, such as GeForce RTX 3090 Ti and Radeon RX 6900 XT. These voltage surges may result in the spontaneous removal of certain power units, even if the latter have a higher than required nominal power.
The NVIDIA graph above shows how the GeForce RTX 4090 current peaks are softened by the new VRM architecture, which uses a feedback PID control loop. If GeForce RTX 3090 can rise sharply and fall, the new map has relatively stable peak values and follows the current load pattern, due to a 10-fold acceleration of the power control system. It is also worth noting that the GeForce RTX 3090 peak is higher than the GeForce RTX 4090. Even though the first reported TDP is 100 W lower than the new flagship.
The company paid special attention to the memory of the video card by reducing its operating temperature by 10 degrees. With the support of Microsoft, the manufacturer has selected chips on a new process with a capacity of 16 Gbit. As a result, all GDR6X chips are placed on one side of the video card ' s printed board and are cooled more effectively.
GeForce RTX 4090 will be sold on October 12, the recommended cost of the video card is $1,600. The new generation's flagship accelerator NVIDIA uses the AD102 processor on Ada Lovelace, which is built using the 4nm TSMC N4 process, has 16384 CUDA kernels, 512 texture blocks and 192 stripping units. The map has received 24 Gbyte GDR6X memory with 384 bits support and 21 Gbit/s contact speeds.