NVIDIA, the biggest name in the graphics card market, continues to make a name for itself with the products developed with Ada Lovelace architecture. The company introduced Ada Lovelace RTX GPUs for laptops at the GPU Technology Conference (GTC) 2023.

RTX 4000 SFF is based on an AD104

NVIDIA combines an AD104 GPU with 20GB of GDDR6 memory at 16Gbps on a 160-bit bus. It comes with 20GB of GDDR6 memory with graphics card error correction code (ECC) with 6144 active CUDA cores, 192 Tensor cores and 48 RT cores. It is stated that the product, which has a single sensitivity of up to 19.2 TFLOPs, offers Tensor-based computing performance up to 306.8 TFLOPs.

RTX 4000 SFF

NVIDIA RTX 4000 SFF key features;

GPUNVIDIA RTX 4000 SFF
GPU Memory20GB GDDR6 with error correcting code (ECC)
Display Ports4x mini DisplayPort 1.4a*
Maximum Power Consumption70W
Graphic BusPCIe Gen 4x16
form factor2.7″ H x 6.6″ L, dual slot

On the other hand, NVIDIA has also updated all mobile workstation GPUs, introducing the RTX 5000, RTX 4000, RTX 3500, RTX 3000 and RTX 2000 Ada Generation. The introduced products perform in line with all consumer desktop graphics cards, with a few exceptions, including the RTX 3500 with 5,120 CUDA cores. The RTX 4000 SFF Island Gen workstation desktop GPU is listed at $1250.

Comparison list to give an idea;

RTX 5000
Island Laptop
RTX 4000
Island Laptop
RTX 3500
Island Laptop
RTX 3000
Island Laptop
RTX 2000
Island Laptop
CUDA Cores97287424512046083072
Tensor Cores30423216014496
Acceleration Clock~2.19GHz~2.26GHz~2.25GHz~2.16GHz~2.36GHz
memory clock18 Gbps GDDR618 Gbps GDDR618 Gbps GDDR616Gbps GDDR616Gbps GDDR6
Memory Bus Width256 bit192 bits192 bits128 bits128 bits
VRAMs16 GB12GB12GB8GB8GB
ECCPartial
(DRAMA)
Partial
(DRAMA)
Partial
(DRAMA)
Partial
(DRAMA)
NO
single precision42.6 TFLOPs33.6 TFLOPs23 TFLOPS19.9 TFLOPS14.5 TFLOPs
Tensor Performance
(FP8, Rare)
681.8 TFLOPs538.0 TFLOPs368.6 TFLOPs318.6 TFLOPs231.6 TFLOPs
NVENC/NVDEC2 of 22 of 22 / 111th11th
Max TGP80-175W80-175W60-140W35-140W35-140W
GPUAD103AD104AD104AD106AD106
ArchitecturalAda LovelaceAda LovelaceAda LovelaceAda LovelaceAda Lovelace
Production processTSMC 4NTSMC 4NTSMC 4NTSMC 4NTSMC 4N