4x Furiosa RNGD: EXAONE 3.5 32B, batch 1, 60 tok/s at 4K context
60 tok/s, per-request, batch 1, 4-card tensor parallel
Fields
id
benchmarks/furiosa-4xrngd-exaone35-32b-4k-b1-decodeMetric
decode_tok_s
Value
60
Unit
tok/s, per-request, batch 1, 4-card tensor parallel
Model
EXAONE 3.5 32B
Methodology
VENDOR, confirmed by a second outlet quoting the same sentence. One server, four RNGD cards, tensor parallel across cards, batch size one. EXAONE 3.5 32B at a 4K context window: 60 tokens per second. Furiosa-LLM, described as vLLM-compatible; the engine build is not named. Prompt length and generation length inside the 4K window are not stated. This is the system rate at batch 1, not an aggregate across concurrent users and not a per-card rate. Do not divide by 4.
Measured by
vendor
Reproducible
no
notes
LG AI Research measurement reported by Furiosa on 2025-07-22 and reprinted by GamesBeat the same day. The comparison GPU is unnamed, so the 2.25x performance-per-watt claim is not stored. Hardware is eval and enterprise, not a retail card. Quantization is not stated.
Notes
LG AI Research measurement reported by Furiosa on 2025-07-22 and reprinted by GamesBeat the same day. The comparison GPU is unnamed, so the 2.25x performance-per-watt claim is not stored. Hardware is eval and enterprise, not a retail card. Quantization is not stated.