DDR6 is a mix of GDDR5, GDDR5X, and even HBM2 features.
GDDR6 has the increased capacity, and increased bandwidth. But the dual-channel feature for workloads with a lot of parallel reads and write to different parts of memory.
The GDDR6 is designed for the new chip structures where chips are stacked. The GDDR5 memory can be best suited in situations where older technology is used which do not opt for stacked chips.
The memory is capable of providing transfer speeds of up to 16 Gbps. The bandwidth supported is up to 72 GB per sec.
| Features / Particulars | GDDR5/5X | GDDR6 |
| Power Consumption | High | Low * |
| Manufacturer | Samsung, Micron, and Hynix | Samsung, Micron, and Hynix |
| Transfer Speeds | 8 Gbps for GDDR5 10 to 12 Gbps for GDDR5X | 14 Gbps to 16 Gbps |
| Availability | FBGA190, 0.65 mm pitch, 14x10mm | FBGA180, 0.75 mm pitch, 14×12 mm |
| I/O Configuration | X16/x32 | X8/x16 |
| Channel Count | 1 | 2 |
| Power Drawn | 1.5 V | 1.3 V |
| Sizes available | 512 MB, 1 GB, 2 GB, 4 GB and 8 GB | 8 GB and 16 GB |
| Features / Particulars | NVIDIA GeForce RTX 2080 Ti (GDDR6) | RTX 6000 | RTX 8000 |
| Number of CUDA Cores | 4352 | 4608 | 4608 |
| Number of Tensor Cores | – | 576 | 576 |
| Boost Clock Speed | 1545MHz | 1730 MHZ | 1730 MHz |
| Memory Clock Performance | 14 Gbps GDDR6 (616 GB/s) | 14 Gbps GDDR6 | 14 Gbps GDDR6 |
| Bus width for memory | 352 bit | 384 bit | 384 bit |
| VRAM Capacity | 11 GB | 24 GB | 48 GB |
| Ray Performance | 10 GRays per sec | 10 Grays per sec | 10 Grays per sec |
| TDP | NA, likely 250 W | NA, likely 250 W | NA, likely 250 W |
| GPU | Turing | Turing | Turing |
| Architecture Support | Turing | Turing | Turing |
| Compute capacity | 13.4 Tflops | 16 Tflops | 16 Tflops |
Quadro RTX 4000
The application performance of Quadro RTX 4000 is real-time, immersive BIM walkthrough. By deploying the Enscape3D plugin and putting on an HMD, step inside a full-scale Autodesk Revit model and make changes in real time.
