[ad_1]
The design is able to run a larger network better than the GPU banks that connect together. But making and operating a chip is a challenge, looking for new ways to make silicon, a design that also incorporates error readings, and an old water system to keep the main chip cool.
In order to create small WSE-2 sticks that can use large AI types, Cerebras has to deal with another engineering problem: how to extract data inside and outside the chip properly. Stable chips have their own memories, but Cerebras developed a memory-chip box called MemoryX. The company has also developed software that allows the neural network to be securely stored in chip-off storage, with only computers being sent to the silicon chip. And it built a software package called SwarmX that connects everything together.
“They can increase the growth of education to a much greater level than anyone else is doing today,” he says Mike Demler, chief executive with Linley Group and editor-in-chief of Microprocessor Report.
Demler says it is unclear what market the market will hold, especially as some of the potential customers are already preparing their special equipment. He added that chip performance, in terms of speed, efficiency, and cost, is not well known. Cerebras has not released any results so far.
“There is a lot of expertise in the new technology of MemoryX and SwarmX,” says Demler. “But as a processor, this is very special; it only makes sense to teach great colors. ”
Cerebras chips are now accepted by labs that require high power. Early clients include Argonne National Labs, Lawrence Livermore National Lab, pharma companies including GlaxoSmithKline and AstraZeneca, and what Feldman describes as “military” legal entities.
This suggests that the Cerebras chip can be used continuously to strengthen networks; The calculations that the laboratory operates also include very similar mathematics. “And they are always thirsting for the power to count,” says Demler, who adds that the chip may be needed in the future using computers.
David Kanter, researcher and World Events is the largest of MLCommons, an organization that tests the functionality of AI algorithms and hardware, says it sees the future market for the largest AI models on a regular basis. “I usually like to trust the data-centric ML, which is why we need big dates that help create larger colors and other components,” says Kanter.
[ad_2]
Source link
