NVIDIA RTX 4000 ADA and mobile update
At GTC 2023 NVIDIA announced an update to its workstation GPU series.
The RTX Ada is a codename for the NVIDIA workstation GPU series. Today NVIDIA has unveiled as many as five mobile GPUs following the launch of its gaming GeForce RTX 40 series last month.
RTX 4000 ADA SFF, Source: NVIDIA
More importantly, there is an update to the desktop RTX series as well, in a form of RTX 4000 ADA SFF (Small Form Factor) professional GPU. This card is equipped with 6144 CUDA cores, 192 Tensors and 48 RT cores. It offers up to 19.2 TFLOPS of compute in single-precision workloads and up to 306.8 TFLOPS in Tensor-based tests.
This dual-slot design is a small “half height” card with a single fan. It does not require any external power connectors because the TDP has been set to only 70W. The memory specs are 20GB GDDR6 across a 160-bit interface with a memory clock at 16 Gbps. The specs indicate this is an AD104 based card. According to NVIDIA, this card will launch at $1250.
RTX 4000 ADA SFF, Source: NVIDIA
In the mobile GPU space, NVIDIA has five new workstation models: RTX 5000, 4000, 3500, 3000 and 2000. The specs are more or less a copy of the RTX 40 gaming series, with two exceptions: the RTX 3500 has 5120 CUDA cores, a configuration which was not available with consumer models.
Furthermore, there is no 2560 CUDA card, so the lowest AD107 specs will only be part of the RTX 40. The maximum TGP reported is 175W, but each card can be configured to operate up to 140W.
NVIDIA RTX ADA Workstation GPUs | |||||
---|---|---|---|---|---|
VideoCardz.com | RTX 5000 ADA | RTX 4000 ADA | RTX 3500 ADA | RTX 3000 ADA | RTX 2000 ADA |
CUDA Cores | 9728 | 7424 | 5120 | 4608 | 3072 |
RT Cores | 76 | 56 | 40 | 36 | 24 |
Tensor Cores | 304 | 232 | 160 | 144 | 96 |
Boost Clock | 2.19 GHz | 2.26 GHz | 2.25 GHz | 2.16 GHz | 2.36 GHz |
Memory | 16GB GDDR6 | 12GB GDDR6 | 12GB GDDR6 | 8GB GDDR6 | 8GB GDDR6 |
Memory Bus | 256-bit | 192-bit | 192-bit | 128-bit | 128-bit |
Bandwidth | 576 GB/s | 432 GB/s | 432 GB/s | 256 GB/s | 256 GB/s |
TGP | 80W-175W | 60W-175W | 60W-140W | 35W-140W | 35W-140W |
FP32 Compute | 42.6 TFLOPS | 33.6 TFLOPS | 23.0 TFLOPS | 19.9 TFLOPS | 14.5 TFLOPS |
Tensor Performance | 681.8 TFLOPS | 538.0 TFLOPS | 368.6 TFLOPS | 318.6 TFLOPS | 231.6 TFLOPS |
RT Core Performance | 98.5 TFLOPS | 77.8 TFLOPS | 53.3 TFLOPS | 46.0 TFLOPS | 33.5 TFLOPS |
Source: NVIDIA