AWS Inf2.48xlarge: 2280 FP16 TFLOPS (12-chip instance)

Vendor-published instance specification. 12 Inferentia2 chips with NeuronLink-v2 at 192 GB/s/chip. Per-chip: 190 TFLOPS FP16, 32 GB HBM, 820 GB/s. Also: 570 TFLOPS FP32, 384 GB HBM, 9.8 TB/s aggregate bandwidth.

Vendor-published peak compute specification. The 2280 TFLOPS is the theoretical FP16 peak for the 12-chip Inf2.48xlarge instance. No inference throughput (tokens/sec) is published by AWS for Inferentia2. The per-chip FP16 figure of 190 TFLOPS is directly from the vendor product page. LOAD-CONDITION AUDIT (bq- 2026-10-04). SCENARIO: NOT STATED by the source - absent, not inferred. A throughput figure with no scenario cannot be placed against an MLPerf Offline or Server number (they are different measurements; on identical hardware the Offline-minus-Server difference is negative in 5 of 9 systems measured). ABSENT CONDITIONS (5 of 7 checked: model, hardware, batch-or-concurrency, input length, output length, engine version, warmup): - engine version: unrecoverable from alt-inf2-arch, asic-aws-inferentia-product, asic-aws-inf2-instances - engine_id=alt-neuron-compiler is named but no build/commit/container digest is published - batch/concurrency: unrecoverable from alt-inf2-arch, asic-aws-inferentia-product, asic-aws-inf2-instances - would need the achieved in-flight batch or the --max-concurrency value - input length: unrecoverable from alt-inf2-arch, asic-aws-inferentia-product, asic-aws-inf2-instances - would need the run's ISL or the prepared dataset's input-length profile - output length: unrecoverable from alt-inf2-arch, asic-aws-inferentia-product, asic-aws-inf2-instances - would need the run's OSL or the harness's configured max_tokens - warmup: unrecoverable from alt-inf2-arch, asic-aws-inferentia-product, asic-aws-inf2-instances - would need the raw harness log's warmup phase or the benchmark command line (loadgen min_duration/min_query_count, or trtllm-bench --warm_up N / --num_runs N) None of these were filled in. A plausible value is worse than an absent one: it would be trusted. METRIC KEY AUDIT 2026-10-05 (agent w6-benchmarks): VALUE UNCHANGED, and the record is already correct in every field except the one a filter reads. The 2280 is a theoretical FP16 peak for a 12-chip instance, not tokens per second, and this record says so. The defect is that metric=tps_aggregate answers a capacity planner's tokens/s query with a FLOP number. Recorded in data/gotchas/bench-tflops-filed-as-tps-aggregate-survives-the-tok-s-filter.json; the same defect affects alt-aws-trn2-fp8-tps, alt-tpu-v5e-bf16-tps and alt-tpu-v6e-bf16-tps, and those four are the complete set in data/benchmarks/ with a tokens/s metric and a FLOP unit.