Oxmiq Labs noted that DeepSeek's R1 model, released in January 2025, was trained on NVIDIA H800 chips and demonstrated competitive performance.
Public source
Publisher name
Public post
๐๐ฒ๐ฒ๐ฝ๐ฆ๐ฒ๐ฒ๐ธ ๐ฝ๐น๐ฎ๐ป๐ ๐๐ผ ๐ฑ๐ฒ๐ฝ๐น๐ผ๐ ๐ญ๐ฒ๐ฌ,๐ฌ๐ฌ๐ฌ ๐๐๐ฎ๐๐ฒ๐ถ ๐๐ ๐ฐ๐ต๐ถ๐ฝ๐ ๐ฎ๐ ๐ฎ ๐ป๐ฒ๐ ๐ฑ๐ฎ๐๐ฎ ๐ฐ๐ฒ๐ป๐๐ฒ๐ฟ ๐ถ๐ป ๐๐ป๐ป๐ฒ๐ฟ ๐ ๐ผ๐ป๐ด๐ผ๐น๐ถ๐ฎ, ๐๐ฎ๐ฟ๐ด๏ฟฝโฆ
Company
Oxmiq Labs
Re-Architecting the GPU Stack: From Atoms to Agentsโข
- Industry
- Technology, Information and Internet
- Location
- Campbell, US
- Company size
- 51โ200 employees
About Oxmiq Labs
Founded by Raja Koduri, OXMIQโข is rearchitecting the GPU full stack from Atoms to Agents to meet the demands of next-generation gaming, graphics and multimodal AI. The company develops licensable GPU software and hardware IP that balances flexibility and performance by integrating breakthrough technologiesโincluding nano agents in silicon based on RISC-V, near- and in-memory computing, advanced light transport, and other innovations. OXMIQโs architecture is designed to scale seamlessly from physical AI devices all the way to Data Center scale. See oxmiq.ai for more info.
See moreLatest activity
Latest activity from Oxmiq Labs
19 signals
Products & Services
Oxmiq Labs announced that Nebius and Palantir are offering a bundled stack including infrastructure and operational intelligence tools designed to meet data residency and security requirements.
Partnerships
Oxmiq Labs announced that Palantir brings the software layer.
Partnerships
Oxmiq Labs announced that Nebius brings NVIDIA-based GPU clusters across the U.S. and Europe.
Discover more
Similar signals
Similar public activity from other companies.
Research & Knowledge
Acasia
Acasia tested an image-classification model using an A100 GPU across thousands of images and classified over 7,000 new images.
Research & Knowledge
Deel
Deel reported that DeepSeek's Multi-head Latent Attention (MLA) technique reduced the KV cache by more than 90% compared to standard attention.
Research & Knowledge
Yantrion Inc
Yantrion Inc measured 975.5 aggregate decode tokens per second across 56 concurrent requests on one 8ร AMD Instinct MI350X node using Kimi-K3.
Research & Knowledge
General Compute
General Compute noted that Cerebras takes a different route by making compute wafer-scale using distributed on-chip SRAM to avoid inter-chip communication bottlenecks while streaming weights.
Products & Services
Marktechpost AI Media Inc