Qualcomm's Adreno X2 GPU
6 hours ago
- Adreno X2 is an improved iGPU from Qualcomm with higher clock speeds, more shader processors, and better memory hierarchy, nearly doubling theoretical compute throughput over Adreno X1.
- Qualcomm claims Adreno X2 lacks hardware fused multiply-add (FMA) capability, leading to poor throughput in OpenCL's fma() built-in function, unlike AMD, Intel, and Nvidia GPUs.
- The text notes that Vulkan throughput graphs show high FP32 performance, but that does not contradict the FMA claim because FMA is a specific fused operation tested separately in OpenCL.
- Adreno X2 features a 21 MB Adreno High Performance Memory (HPM) for tiled rendering, but only 1 MB of it is usable as compute local memory.
- In graphics benchmarks, Adreno X2 performs well, beating Intel's Meteor Lake iGPU in Cyberpunk 2077 and approaching AMD's larger iGPUs in some tests.
- Compute performance is a weak point: Adreno X2 loses to older discrete GPUs like GTX 1050 in FluidX3D and FAHBench, indicating compute bottlenecks.
- Raytracing results are mixed: excellent in lightweight tests (Solar Bay) but poor in heavy workloads (Cyberpunk 2077 with raytracing).
- Overall, Adreno X2 is a step forward for Qualcomm, but it still has ‘glass jaw’ cases in compute and certain raytracing scenarios compared to competitors.