Today, Nvidia announced two blockbuster products, including the massive Hopper H100 GPU and a 144-core Grace CPU Superchip. In addition, the chipmaker also refreshed its RTX professional lineup of mobile and desktop workstation solutions for professionals to tackle next-generation workloads.
On the desktop side, Nvidia revealed the RTX A5500, which slots in between the RTX A5000 and RTX A6000. It would appear that the RTX A5500 will replace the existing RTX A5000. The new model comes with 80 streaming multiprocessors (SMs), 16 more than the RTX A5000 and only four less than the RTX A6000. In addition, the CUDA core bump allows the RTX A5500 to deliver up to 34.1 TFLOPs of FP32 performance, 23% higher than the RTX A5000.
The RTX A5500 preserves the same memory configuration as the RTX A5000. The 24GB of GDDR6 ECC memory runs at 16 Gbps across a 384-bit memory interface to supply a memory bandwidth of 768 GBps. Nvidia didn't specify the RTX A5500's TDP, but Leadtek rates the graphics card with a 230W maximum power consumption.
The RTX A5500 features a dual-slot design with a blower-type cooling system. Measuring 267mm (10.5 inches) long, the graphics card slides into a PCIe 4.0 x16 expansion slot. It draws power from a single 8-pin PCIe power connector and offers four DisplayPort 1.4 outputs.
Meanwhile, Nvidia's mobile RTX series of professional graphics cards receive the most attention, with the chipmaker adding up to six new models. The latest additions include the RTX A5500, RTX A4500, RTX A3000 12GB, RTX A2000 8GB and RTX A500.
The RTX A5500 has 58 SMs compared to the RTX A5000's 48 SMs, resulting in a 21% increase in CUDA cores and a potential 28% boost in FP32 performance. However, it still has the same 16GB of GDDR6 memory, and it's unlikely that the 256-bit memory interface has changed. Nvidia apparently endowed the RTX A5500 with 16Gbps memory, which is how the graphics card can offer up to 512 GBps of memory bandwidth.
The RTX A4500 takes the place of the RTX A4000, with 46 SMs — four more SMs than the latter. In addition, Nvidia has doubled the memory on the RTX A4500 from 8GB to 16GB. The RTX A4500 also shows higher memory bandwidth with a rating up to 512 GBps, 33% higher than the RTX A4000. That again indicates the use of 16Gbps GDDR6.
Meanwhile, the RTX A3000 and RTX A2000 only obtained a memory upgrade. As their names denote, the RTX A3000 12GB and RTX A2000 8GB now have 12GB and 8GB, respectively, twice the memory of vanilla RTX A3000 and RTX A2000. It's not just more memory, but faster memory. The RTX A3000 12GB and RTX A2000 8GB supply up to 336 GBps and 224 GBps, respectively, 27% more bandwidth than the RTX A3000 and 17% on the RTX A2000.
The RTX A1000 and RTX A500 are newcomers to the RTX family. Both models sport 16 SMs or 2,048 CUDA cores and 4GB of GDDR6 memory. The differentiator is the memory bandwidth. The RTX A1000 is good for 224 GBps, whereas the RTX A500 is limited to 112 GBps. It would appear that the A500 is limited to a 64-bit interface with 14Gbps GDDR6, while the A1000 gets the full 128-bit interface that the GA107 GPU offers.
The RTX A5500 (desktop) graphics card is available today, while the RTX mobile offerings will arrive starting this spring.