inference/research
verifiedmedium · 0.70updated 2026-10-16

FuriosaAI press release: RNGD unveiled at Hot Chips 2024

blog

Fields

id
sources/furiosa-blog-rngd-hot-chips-press-release
Publisher
FuriosaAI
Kind
blog
Published
2024-08-26
Accessed
2026-10-16
notes
Vendor press release, Santa Clara, 2024-08-26, republished on the current furiosa.ai blog. States: a single RNGD PCIe card delivers 2,000 to 3,000 tokens per second throughput for models with around 10 billion parameters, depending on context length. Also states 48 GB HBM3 and that Llama 3.1 8B runs on a single card. The same page quotes TDP 150 W; the current product page says 180 W. No engine version, batch size, or exact prompt and output lengths. A range, not a point measurement.

Notes

Vendor press release, Santa Clara, 2024-08-26, republished on the current furiosa.ai blog. States: a single RNGD PCIe card delivers 2,000 to 3,000 tokens per second throughput for models with around 10 billion parameters, depending on context length. Also states 48 GB HBM3 and that Llama 3.1 8B runs on a single card. The same page quotes TDP 150 W; the current product page says 180 W. No engine version, batch size, or exact prompt and output lengths. A range, not a point measurement.

Cross-references

This record references no other records.