Conclusion
We have now pitted AMD’s flagship GPU against NVIDIA’s flagship GPU in compute, content creation, and professional workstation application benchmarks. We have also recently fully reviewed the GeForce RTX 3090 Ti FE, Overclocked it, as well as the Radeon RX 6950 XT, and compared overclock vs overclock in many games at 4K, with Ray Tracing at 1440p and with DLSS and FSR. These compute benchmarks today should provide the full scope, combined with the gaming performance as to how the GeForce RTX 3090 Ti and Radeon RX 6950 XT compete. Depending on your work case and workload scenario, you should be able to decide which video card will provide the best experience for you.
Compute Performance
Starting off in Geekbench, we looked at OpenCL and Vulkan compute performance. We found that in both APIs, the GeForce RTX 3090 Ti FE was faster. In OpenCL it was 13% faster and in Vulkan, it was much faster at 88% of a difference compared to the Radeon RX 6950 XT. When we benchmarked CUDA, it was faster still, beating OpenCL performance by a lot.
Next up was Blender, and the GeForce RTX 3090 Ti FE simply dominated performance in Blender 3.2.0 versus the Radeon RX 6950 XT. In monster it was 119% faster, in junkshop it was 139% faster and in classroom, it was 125% faster. When it comes to Blender, you just cannot beat OptiX on GeForce RTX video cards, it’s so much faster, and can also render Ray Tracing so much faster. AMD’s OpenCL implementation cannot hold a candle to it, even NVIDIA’s CUDA is faster than AMD’s OpenCL in Blender.
In our next benchmark we utilized LuxMark’s Hall Bench and Food benchmark scenes. Hall Bench tests path tracing and global illumination, while Food is all about brute force path tracing. In both scenes, the GeForce RTX 3090 Ti FE came out on top, even though both video cards are using OpenCL here. The RTX 3090 Ti FE was 32% faster in Hall Bench and 15% faster in Food.
Next we moved on to AIDA64’s GPGPU Benchmark which tested a myriad of GPGPU compute functions. Firstly, it does seem the Radeon RX 6950 XT technically has a higher compute memory read, write, and especially copy bandwidth. However, that doesn’t help raw compute workloads like Single-Precision, where the RTX 3090 Ti FE smoked the Radeon RX 6950 XT by 57%. In Double-Precision though, it flipped and the Radeon RX 6950 XT was a large 154% faster. This trend continued in 24-bit Integer but then fell off a cliff in 32-bit and 64-bit Integer where the GeForce RTX 3090 Ti FE smoked it again. The GeForce RTX 3090 Ti FE continued to be much faster in AES-256 by 48% and in SHA-1 Hash by 38%. The RTX 3090 Ti FE was also faster in Julia by 25% but the Radeon RX 6950 XT smoked it in Mandel by 206%.
Moving on to PerformanceTEST and PassMark Advanced GPU Compute testing the GeForce RTX 3090 Ti FE was the fastest in almost all tests. In NBodyGravity it was 21% faster, in Quaternion Julia Fractals it was 71% faster, in Mandelbrot it was 47% faster. However, in Fluid Simulations the Radeon RX 6950 XT was 11% faster and in OpenCL the RX 6950 XT was 51% faster.
Finally, we come to SPECviewperf 2020 and testing professional workstation application rendering performance at 4K. Once again we saw mostly a dominance of the GeForce RTX 3090 Ti FE in performance, save for a few specific applications. In 3dsmax the RTX 3090 Ti FE edged out performance by a small 3%, but that grew larger in catia where it was 13% faster, in maya it was 19% faster and in solidworks it was 63% faster. However, the Radeon RX 6950 XT does have some strong points. In energy it was 8% faster, in medical it was 35% faster and in snx it was a whopping 267% faster.
Final Points
At the end of the day, the GeForce RTX 3090 Ti FE seems to have more gains and wins over the Radeon RX 6950 XT than it loses. The GeForce RTX 3090 Ti FE is very strong in some specific cases, it seems to have overall faster OpenCL and Vulkan performance, and of course supports CUDA, something the Radeon RX 6950 XT cannot claim.
In CUDA accelerated applications, the RTX 3090 Ti FE is generally going to perform much higher than OpenCL. In addition, in applications like Blender, it can use OptiX, which takes it even further and just dominates 3D rendering performance. It seems to be very strong in encryption performance as well, as proven by AIDA64 and does well with high bit-depth Integer performance. With the large 24GB of VRAM, it’s great for workstation applications.
The Radeon RX 6950 XT does have a few specific wins in very specific workstation application scenarios. The RDNA2 Radeon RX series has always seemed to be very strong in energy, medical, and snx performance. If that’s your bag, then look no further than the Radeon RX 6950 XT. Its 16GB of VRAM is also welcomed in such scenarios.
Overall though, when you look at the flexibility in performance advantages offered by the GeForce RTX 3090 Ti FE in compute and workstation-class applications, with its added ability to utilize CUDA, OptiX, and render Ray Tracing faster, with 24GB of VRAM, the GeForce RTX 3090 Ti FE seems to be the overall, more well-rounded winner, for compute workloads.