AIDA64 GPGPU Benchmark
We are running AIDA64 Engineer version 6.33.5700 and its GPGPU Benchmark. It runs OpenCL tests and tests various functions of GPGPU Compute performance on GPUs. We have graphed out each test separately, to compare the video cards.
The memory read to measure the maximum achievable memory data transfer bandwidth. According to this test, the AMD Radeon RX 6900 XT is ahead of the RTX 3090 achieving 26GB/s versus 24GB/s read on the RTX 3090.
Memory write bandwidth is also a bit faster on the AMD Radeon RX 6900 XT at near 26GB/s while it is 24GB/s on the RTX 3090.
Memory copy seems to be vastly faster on the AMD Radeon RX 6900 XT versus the RTX 3090. The result above for the AMD Radeon RX 6900 XT is 1.2TB/s, yes TerraBytes per second, while the RTX 3090 is 797GB/s (GigaBytes per second.)
In Single-Precision FLOPS performance the RTX 3090 is much faster than the Radeon RX 6900 XT, it’s a whopping 45% faster.
However, the tables flip when we look at Double-Precision FLOPS performance. The AMD Radeon RX 6900 XT is massively 147% faster than the RTX 3090.
The trend continues in the 24-bit Integer IOPS test, with the AMD Radeon RX 6900 XT being 15% faster than the RTX 3090.
But as we increase up in precision, now at 32-bit Integer IOPS the tables turn and the RTX 3090 is mind-blowing faster, 310% faster than the AMD Radeon RX 6900 XT.
The trend continues with the higher precision 64-bit Integer IOPS also being 260% faster on the RTX 3090.
It looks like AES-256 is also faster on the RTX 3090 by 49% over the Radeon RX 6900 XT.
SHA-1 Hash is also faster on the RTX 3090 by 35% over the Radeon RX 6900 XT.
Single-Precision Julia is also faster on the RTX 3090 by 30% over the Radeon RX 6900 XT.
The tables turn though in the Double-Precision Mandel test with the AMD Radeon RX 6900 XT being 209% faster.