SHA-256 vs SHA256-90R – Multi-Backend Performance (AWS Intel Xeon Platinum 8375C @ 2.90 GHz)

Algorithm / Backend	Rounds	Block / Output Size	Cycles/Byte (cpb)	Bytes/Cycle	Latency (ns)	Throughput/Core (Gbps)	Slowdown vs Standard	System Specs
SHA-256 (Scalar C)	64	512b / 256b	~60.0	0.017	~240 ns	~0.46 Gbps	– (baseline scalar)	Portable C, no SIMD
SHA256-90R (Scalar C)	90	512b / 256b	~45.0	0.022	~300 ns	~0.60 Gbps	~30% slower vs 64r scalar	Same C impl, 40% more rounds
SHA-256 (AVX2)	64	512b / 256b	~7.1	0.14	~20 ns	~4.2 Gbps	~9× faster vs scalar	AVX2 vectorized, 4-way blocks
SHA256-90R (AVX2)	90	512b / 256b	~11.0	0.091	~24 ns	~2.7 Gbps	~36% slower vs 64r AVX2	AVX2 unrolled 90 rounds
SHA-256 (SHA-NI)	64	512b / 256b	0.18	5.56	~18 ns	154.8 Gbps	– (baseline fast)	Intel SHA-NI instructions
SHA256-90R (SHA-NI)	90	512b / 256b	0.26	3.85	~24 ns	106.7 Gbps	~31% slower	Same system, same SHA-NI
SHA-256 (FPGA Sim)	64	512b / 256b	~5.0 (est)	0.20	~50 ns	~12.3 Gbps	Baseline FPGA	200 MHz pipelined sim
SHA256-90R (FPGA Sim)	90	512b / 256b	~6.5 (est)	0.15	~67 ns	~12.8 Gbps	~4% faster (pipeline depth)	90-stage FPGA pipeline
SHA-256 (GPU est.)	64	512b / 256b	~0.02	50.0	<0.1 ns	45–50 Gbps	~100× scalar	CUDA/OpenCL batch
SHA256-90R (GPU est.)	90	512b / 256b	~0.03	33.0	<0.1 ns	50+ Gbps	Comparable (batch parallel)	Needs warp-level opt