Figure 7c reports forward-plus-backward attention throughput on H100 80GB SXM5; PyTorch is marked OOM at sequence length 16k. The panel covers causal attention with head dimension 64.

Figure 7c reports forward-plus-backward attention throughput on H100 80GB SXM5; PyTorch is marked OOM at sequence length 16k. The panel covers causal attention with head dimension 64. · CiteArk

Measurement results

0/17 assessed

0 supported0 disputed or conflicting17 not assessed

The complete result group stays with its claim. Assessed rows come first; their judgments do not replace the whole-claim conclusion. Expand a row for conditions, sources, and evidence history.

Each row is a measurement with its reported value, latest available observation, and cumulative evidence state.
Metric and conditions	Reported	Latest observation	Evidence judgment
method: PyTorch · sequenceLength: 512	26 TFLOPs/s	—	Not assessed
method: FlashAttention · sequenceLength: 512	104 TFLOPs/s	—	Not assessed
method: FlashAttention-2 · sequenceLength: 512	141 TFLOPs/s	—	Not assessed
method: PyTorch · sequenceLength: 1k	29 TFLOPs/s	—	Not assessed
method: FlashAttention · sequenceLength: 1k	123 TFLOPs/s	—	Not assessed
method: FlashAttention-2 · sequenceLength: 1k	192 TFLOPs/s	—	Not assessed
method: PyTorch · sequenceLength: 2k	31 TFLOPs/s	—	Not assessed
method: FlashAttention · sequenceLength: 2k	136 TFLOPs/s	—	Not assessed
method: FlashAttention-2 · sequenceLength: 2k	232 TFLOPs/s	—	Not assessed
method: PyTorch · sequenceLength: 4k	32 TFLOPs/s	—	Not assessed
method: FlashAttention · sequenceLength: 4k	138 TFLOPs/s	—	Not assessed
method: FlashAttention-2 · sequenceLength: 4k	257 TFLOPs/s	—	Not assessed

Figure 7c reports forward-plus-backward attention throughput on H100 80GB SXM5; PyTorch is marked OOM at sequence length 16k. The panel covers causal attention with head dimension 64.

Reported and observed measurements

Assessments (0)

Runs (0)

Measurement results