Home/Events/Kog Aims for 30x Faster LLM Inference on Conventional GPUs

Kog Aims for 30x Faster LLM Inference on Conventional GPUs

Developing

Confidence

70%

Impact: 60%

Updated 1h ago

Consensus Brief

French startup Kog is focusing on optimizing conventional GPUs for faster AI inference, claiming to achieve 30x faster LLM inference. The company has demonstrated a speed of 3,000 tokens per second using a small model and aims to prove its approach works on larger models. Kog has attracted significant interest, with 200 business leads reported by CEO Gaël Delalleau.

Sourced from

TechCrunchT2

Primary: TechCrunch

What Changed Since Last Update

1h ago

Kog's approach promises to unlock new capabilities on existing hardware through software optimization, contrasting with the trend of developing purpose-built chips.

Claim Ledger

4 claims tracked across sources

Official Claim

Kog aims for 30x faster LLM inference.

Confirmed Fact

Kog demonstrated 3,000 tokens per second using a small model.

Confirmed Fact

Kog received 200 tangible business leads.

Confirmed Fact

Kog is supported by Scaleway and backed by France’s Bpifrance and French Tech 2030’s program.

Role-Based Impact Analysis

Source Timeline

1 source corroborating

Kog is going deeper to squeeze more inference out of GPUs

TechCrunch·1h ago