Nvidia RTX GPU. But this time not for gamers

Mar 26, 2022 - 20:48
 0  45
Nvidia RTX GPU. But this time not for gamers

Although Nvidia's latest presentation focused on the 144-core Hopper H100 graphics unit, it also introduced new professional RTX graphics cards aimed at both desktops and laptops.

In the case of desktop PCs, only one new model has been introduced: the RTX A5500, which sits between the RTX A6000 and A5000. It features a total of 10,240 Cuda cores, 80 ray-tracing cores, and 320 tensor cores. The card contains 24 GB of GDDR6 (ECC) memory with a throughput of 768 GB/s and a total compute power of 34.1 TFLOPS in SP.

This model's TDP is 230 W (one 8-pin PCI Express connector is required), and the two-slot cooling system is a conventional blower, meaning it has a single fan that exhales heated air from the cabinet to the outside. DisplayPorts 1.4 can link up to four displays.

Many new laptop models are now available. There are six mobile cards available: the RTX A500, A1000, A2000, A3000, A4500, and the highest model, the RTX A5500.

The RTX A5500's most powerful variant features a processor with 7,424 Cuda cores, 58 ray-tracing cores, 232 tensor cores, and 16 GB of memory with a throughput of 512 GB / s. 24.7 TFLOPS of computing power (SP).

Professional models are intended for graphics, CAD, and other professional applications where certified drivers and extended support are available. Mobile versions for laptops will be available in the first devices in the spring, while the RTX A5500 desktop card is already available.

GeForce RTX 4080/4090 series supports up to 24 GB of video memory, NVIDIA can increase TDP to 600 W

The GeForce RTX 3090 Ti is gradually becoming available, and Nvidia has done extensive testing on this card thus far. It is the first graphics card with a TDP of 450W and a new 12 + 4Pin 12VHPWR power interface, as well as a complete GA102 core and Micron's new 2GB GDDR6X module. Memory speed exceeds 21 Gbps. There could be other reasons why Nvidia places such a high value on this.

According to Igor's lab, the GA102 may be compatible with the next-generation AD102 based on the Ada Lovelace architecture, which means that much of the development work on the GeForce RTX 3090 Ti can now be transferred to the next-generation flagship model, including the GeForce RTX 4080/4090 series.

The reference design of the PCB equipped with AD102 is supposed to indicate that the GPU power supply reaches 24 phases and the memory power supply is 4 phases, for a total of 28 phases, an increase from 20 phases.

The GeForce RTX 3090 appears to corroborate that power consumption is high. The GeForce RTX 4090 series will be more powerful than the GeForce RTX 4080 series. There are rumors that it will have a power output of 600W or higher.

Although the GA102 is pin-compatible with the AD102, it is rumored that Nvidia partners will use this PCB and bespoke GA102 600W BIOS for internal testing before obtaining the Ada Lovelace GPU.