Mythic M1076 AMP
2022-08- Peak
- 25 TOPS (INT8)
- TDP
- 3W
- MSRP
- $199
Boutique silicon, custom architectures for non-CUDA AI workloads.
Cerebras, Groq, SambaNova, Tenstorrent, Mythic, Graphcore, and Hailo are the boutique chip designers that bypass conventional GPU architectures for specialized AI compute. They deliver competitive performance-per-watt on specific workloads, wafer-scale training, ultra-low-latency inference, sparse models, analog compute, but struggle with software maturity and general-purpose adoption. Their honest value is for builders willing to tightly couple workloads to fixed-function hardware.
Showing 2 of 42 catalogued products. All affiliate links use tag fredoline-20.
Affiliate disclosure: As an Amazon Associate, MyAIHardware.com earns from qualifying purchases at no cost to you. Prices vary; verify on Amazon. Tag: fredoline-20.
This segment is consolidating around the inference economics opportunity, Groq, Cerebras, and SambaNova are racing to offer cheaper-than-NVIDIA hosted inference APIs. Tenstorrent (RISC-V open ISA) is the wildcard that could change the competitive landscape entirely if their open ecosystem reaches critical mass.
Use our public benchmark widget on your own site, free, attribution-only. Auto-updates as new AI ASIC Specialists hardware ships.
Get weekly AI hardware news, benchmark updates, and deals in your inbox. Founding-subscriber list, be one of the first.