There has been a development that could change the balance in the artificial intelligence hardware market. Tenstorrent company announced that the new generation Galaxy Blackhole servers it developed provide a serious competitive advantage in terms of both performance and cost against Nvidia GB300-based systems, thanks to the RISC-V architecture. This system, which offers high efficiency especially in running deep learning models and productive artificial intelligence tasks, aims to accelerate enterprise-wide big data processing processes. Tenstorrent claims that in its tests, its architecture outperforms the solutions offered by the giants in the industry, especially in delay-sensitive applications.

RISC-V Architecture Pushes Performance Boundaries

Galaxy Blackhole systems developed by Tenstorrent promise lower latency compared to their Competitors in high-performance artificial intelligence tasks, especially thanks to the “Blitz” mode.

In the tests carried out on the DeepSeek-R1-0528 671B model, more than 350 tokens were produced per second, surpassing the performance of leading systems in the industry such as Groq and Cerebras. In addition, the first token time with 100K context length is under 4 seconds, proving the efficiency of the system in the workflow.

Tenstorrent’s new solutions deliver up to a 10x performance increase over existing GPU-based systems, especially in video generative AI tasks.

Technical Details and Hardware Structure Stand Out

Tensix cores at the center of Galaxy Blackhole servers are equipped with five customized RISC-V processors.

These processors offer an integrated structure with matrix multiplication units, vector processing blocks and high-speed local SRAM memories. Each core can be optimized for different artificial intelligence workloads thanks to its fully programmable structure.

The 56 800G Ethernet ports offered by the hardware allow seamless integration of the system in data centers requiring high bandwidth.

The price tag, starting at $110,000, aims to redefine the cost-performance balance for companies looking for high-performance AI infrastructure.

The Future of Competition in the Sector is Taking Shape

The fact that a RISC-V-based alternative offers such high performance figures against Nvidia’s market dominance is interpreted as a harbinger of a new era in the hardware world. Tenstorrent aims to reduce the costs of businesses that focus on large language models (LLM) and video production processes, with the scalable architecture it offers. The hardware’s 1 TB DRAM with 16 TB/s bandwidth and 6.2 GB SRAM capacity running at 2.9 PB/s reveals the power of the system in data-intensive calculations.

Do you think these new generation servers based on RISC-V can shake the market leadership that Nvidia has maintained for a long time? Share your opinions about this technology and your priorities in hardware preferences with us in the comments section.