Is It Too Late to Buy Cerebras Systems After the Stock Soared Following Its IPO?

Will AI make the world's archetypal trillionaire? Our squad conscionable released a study connected the 1 little-known company, called an "Indispensable Monopoly" providing the captious exertion Nvidia and Intel some need. Continue »

Cerebras' Wafer-Scale Engine (WSE) chips are, comparatively speaking, enormous. While astir companies person worked to shrink their chips to astir the size of a postage stamp, Cerebras has gone successful the other direction, designing immense chips that are astir 8.5 inches connected a broadside -- astir the size of a regular iPad, not the mini version. Cerebras chips are fundamentally the size of an full spot wafer. This makes them much hard and costly to manufacture.

Manufacturing high-end logic chips successful wide is simply a hard task, arsenic defect rates tin beryllium high. Taiwan Semiconductor Manufacturing is the lone foundry that has consistently been capable to nutrient specified chips with precocious yields of defect-free processors, which is however it has achieved a virtual monopoly successful the space. However, adjacent its chips aren't 100% defect free. When manufacturing accepted GPUs, each silicon wafer is chopped to marque dozens oregon hundreds of idiosyncratic processors. If a fewer of those chips are excessively flawed to ship, it's a manageable issue. But erstwhile you're producing a azygous wafer-sized chip, you can't spend defects that origin an full wafer to beryllium thrown out. To trim the chances of that happening, Cerebras incorporates spare cores into its architecture, enabling its chips to enactment astir flawed sections truthful that they don't interaction the chip's performance.

Another large quality with Cerebras' AI chips is the benignant of representation they use. Most GPUs and different AI chips are packaged with off-chip high-bandwidth representation (HBM), a peculiar signifier of DRAM (dynamic random entree memory), which is needed to assistance optimize their performance. However, fixed the size of its chips, Cerebras chips usage on-chip SRAM (static random entree memory), which is faster but besides physically larger and much complex.