Expired50% confidenceFactExact time
Halo's prefill result of 650 tokens/s is nearly double the speed measured for the same chip last year, with the improvement stemming from AMD contributing code to upstream open-source projects such as llama.cpp to advance the maturity of the ROCm ecosystem.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/12/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 10/10/2026(expired)
Sources
AMD Ryzen AI Halo Hands-On: Can x86 Architecture Challenge NVIDIA's DGX Spark?
bilibili量子菠萝_7/7/2026
Related Claims
UnverifiedIn prefill speed tests, Halo achieved approximately 650 tokens/s while Spark achieved approximately 2000 tokens/s, with Spark being nearly three times faster.69% similarUnverifiedA sandbox engine with 8 million parameters (8.3M) achieves a decoding speed of approximately 2,000 tokens/second on a 6-core desktop CPU released in 2019.62% similarUnverifiedPangu 2.0's single-card throughput on Ascend chips is twice that of mainstream open-source models in the industry.61% similarUnverifiedMercury 2 achieved a measured generation speed of over 1,000 tokens per second.60% similarUnverifiedAirLLM 以推理速度为代价，每生成一个 token 耗时数秒乃至数十秒，远慢于高端 GPU 全量加载的每秒数十 token59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/488916API
curl https://kongchang.com/api/v1/knowledge/claims/488916MCP
get_claim(id=488916)