Read throughput correctly

At 250 MHz, 500,000 results per second permits 500 cycles per result:

250,000,000 cycles/s ÷ 500,000 results/s = 500 cycles/result

At 300 MHz the same result-rate target permits 600 cycles per result. This does not make 300 MHz easier to route; it only changes the cycle budget after that clock has been justified.

Current generated-RTL interval

The sustained signer trace measured one aggregate interval spanning 48 steady completions in 23,121 cycles:

23,121 ÷ 48 = 481.6875 cycles/result

Normalizing that interval at 250 MHz gives 519,008.693 results/s. This is a useful architecture result because it is below the 500-cycle budget. It remains a normalized generated-RTL rate, not a measured U280 rate, until the exact RTL meets 250 MHz through the required hardware gates.

Service interval versus latency

One signature requires 1,258 compression blocks. Three filled lanes have a theoretical service capacity of three blocks per cycle. Individual job latency is much longer because dependencies, 64-stage lane latency, context scheduling, and output framing remain. Saturated completion interval—not first-job latency—is the relevant throughput metric.

Output traffic

The 2,208-byte fused result produces 1.104 GB/s at 500,000 results/s before shell protocol overhead. A core timing result does not prove the shell or memory system can sustain that traffic.