Simultaneous multithreading (SMT) is widely adopted in modern x86 processors to improve core utilization by sharing hardware resources between co-located threads. However, SMT threads often suffer from performance interference due to contention on shared resources. We observe there is substantial performance potential in identifying SMT-aware workload combinations (up to 10%), but finding optimal co-scheduling is challenging due to the huge complexity and diversity of modern CPUs. In this paper, we propose SMTcheck, a lightweight and accurate methodology for predicting SMT performance interference across diverse x86 platforms. SMTcheck uses carefully designed code snippets (Diags) to extract hidden features of the most performance-critical shared resources. With these extracted features, SMTcheck builds novel per-resource microbenchmarks (Injectors) that are capable of applying pinpoint pressure to specific target resources. SMTcheck then constructs a hardware aware contention model to predict performance interference between arbitrary workloads without exhaustive profiling. SMTcheck achieves high prediction accuracy by up to 95.63% (94.98% on average) across four Intel and AMD CPUs. To show its effectiveness, we implement the contention-aware scheduler in the Linux kernel. The results show our scheduler improves system throughput by up to 1.061× and reduces tail latency by up to 27.5%. We also demonstrate its practicality in data center scenarios by showing that it only incurs negligible profiling overheads (≈ 0.113%).