At the GTC conference organized by NVDIIA, it was revealed what artificial intelligence models can achieve. memory interference MemVerge, claims that its self-developed software can maximize GPU utilization. This may open new doors in training artificial intelligence models.
How will artificial intelligence education change with MemVerge?
MemVerge, in its presentation at the GTC conference Compute Express Link (CXL) He revealed how to accelerate language model training using technology. The company, working together with Micron, Memory Machine X He tested his memory tiering software on various artificial intelligence chips from NVIDIA.
The company’s software; Tested on a Supermicro server hardware equipped with NVIDIA A10 GPU, Micron DDR5 memory and CZ120 CXL. In the scope of testing, GPU+DRAM was compared with MemVerge’s GPU, CPU and CXL data layering software.
Compared to standard methods, thanks to Memory Machine X’s management of memory traffic Up to 50 percent speed increase happened. In addition, GPU usage increased by half, reaching over 91 percent.
MemVerge CEO Charles Fan said it is possible to scale AI models cost-effectively. Stating that the company “feeds GPUs with data” with its own software, the CEO underlined that they maximized efficiency.
Although the importance of the chip in training artificial intelligence models is indisputable, various software can also play a critical role in this training. MemVerge aims to make these hardware most efficient by working with NVIDIA and Micron.
Source link: https://shiftdelete.net/memverge-cxl-yapay-zeka