NVIDIA unveiled its Ada Lovelace architecture along with the RTX 4090 & the RTX 4080 last month. These GPUs bring massive improvements from generation to generation along with new features such as DLSS 3.0. The RTX 4090 showed a 60% CUDA improvement against last-gen in a leaked benchmark. Today, the RTX 4090 has been benchmarked in Blender and it turns out to be almost 90% faster than the RTX 3090 Ti.
The RTX 4090
The RTX 4090 is built on the AD102 GPU, which is unparalleled in terms of performance. The GPU features a massive 24GB of Samsung-made GDDR6X memory running at 24Gbps for 1152GB/s of effective bandwidth. Packed with 16384 CUDA Cores, this powerhouse is home to 75 billion transistors in total, more than 2.5x than the RTX 3090 Ti. Priced at $1599, it has raised a few eyebrows from community. Read more about the 4090 here.
The RTX 4090 in Blender shows a massive increment against Ampere. The GPU scores 11873.55 points which puts it 90% ahead of even the RTX 3090 Ti.
The actual results from GPU-GPU are shown below. The difference between the RTX 3090 Ti and the RTX 4090 stands at ~90%. This is slightly less than Jensen’s promise of a 2-4x performance uplift, however, that number included results from DLSS 3.0. Read more about the new DLSS 3.0 technology here.
|SKU||Chip||FP32/CUDA||Max Clock||Cache||Memory Bus||VRAM||Memory Spec||Speed (Gbps)||TDP|
|RTX 4000 Titan?||AD102-450||18176||3.0GHz+?||96MB?||384/382-bit||48GB||GDDR6X||24||~800W|
|RTX 4090 Ti||AD102-350||18432||3.0 GHz?||96MB||382-bit||24GB||GDDR6X||24||600W?|
|RTX 4090||AD102-300-A1||16384||2.52GHz||96MB||384-bit||24GB||GDDR6X||21||450W+ (TGP) / 660W (Max TGP)|
|RTX 4080 Ti||AD102||14848?||2.7 GHz?||80MB?||320-bit||20GB||GDDR6X||23||420W|
|RTX 4080 (Variant 1)||AD103-300-A1||9728||2.505GHz||64MB||256-bit||16GB||GDDR6X||22.5||320W(TGP)/ 516W (Max TGP)|
|RTX 4080 (Variant 2)||AD104-400-A1||7680||2.61GHz||48MB||192-bit||12GB||GDDR6X||21||285W (TGP) /366W (Max TGP)|
|RTX 4070 Ti||AD104-300?||7680||3.0GHz?||?||192-bit||12GB||GDDR6X||21||300W?|