Two Geekbench 7 listings reveal NVIDIA's RTX Spark laptop superchip will ship in both 20-core and 18-core CPU configurations. The full 20-core variant (10+10 Cortex-X clusters at 4.0 GHz, 6,144 Blackwell CUDA cores) scored 2,570 single-core and 23,126 multi-core points. The cut-down 18-core SKU (10+8 clusters at 3.9 GHz, 5,120 CUDA cores) posted 2,541 single-core and 21,776 multi-core. Both show strong multi-core performance relative to competitive Windows-on-Arm and x86 mobile parts: the 20-core RTX Spark trails Apple's M4 Max (25,957) by 11% and beats AMD AI Max+ 395 (20,609) by 12%.
Single-core performance lags expectations. At 2,570, the 20-core is roughly 23% behind Apple's M4 Pro (3,315) and 41% behind M5 base. This reflects NVIDIA's decision to use older Cortex-X925 and Cortex-A725 cores tuned for sustained frequency rather than peak single-thread burst—a tradeoff prioritizing GPU and unified-memory workloads over CPU-centric tasks. The 18-core 21,776 multi-core score matches Apple M4 Pro (21,700) despite having fewer GPU cores (5,120 vs. 10,000 in Pro Max).
For architects, RTX Spark's GPU-first design philosophy is clear: multi-core CPU performance at parity or better than competing 18–20-core Windows solutions, but single-core weakness suggests sustained multi-threaded CPU work will underperform. The real differentiator is the 6,144-core Blackwell GPU with unified memory—two orders of magnitude more TOPS (1,000+ FP4) than Snapdragon X2 Elite's NPU (50 TOPS). These are engineering-sample Geekbench results; final retail hardware with optimized drivers should improve, but single-core may remain constrained by thermal/power envelope on thin laptops launching fall 2026.