Nvidia가 JEDEC HBM 표준을 우회하고 있다고 Blocks & Files가 보도했다.

GPU leader and dominant HBM buyer Nvidia has effectively sent JEDEC back to the drawing board with its high-bandwidth memory (HBM) standard.

HBM is interposer-connected stacked memory for GPUs that bypasses the capacity and bandwidth limits of socket-connected x86 DRAM. The Joint Electron Device Engineering Council (JEDEC) published its first HBM standard, JESD235 (restricted access), in 2013. Samsung, SK Hynix and Micron built compliant HBM chips, which were bought mainly by Nvidia, its system-building partners and other GPU builders. Now, 23 years later, Nvidia has rebelled against having HBM memory controller functionality in its GPU chip and has moved it to the HBM base die, freeing up space for more GPU cores.

The change is part of Nvidia's NVHBM custom HBM standard, announced in August as part of its NVLink Fusion announcement. Nvidia claims NVHBM delivers up to 30 percent greater memory bandwidth, 15 percent lower HBM power consumption, and up to 25 percent more area on its GPU compute die compared with standard HBM4E.

Two Nvidia GPU chip layout images illustrate the effect. Currently, HBM has a wide physical interface with many pins that connect it via an interposer to the GPU, where the HBM memory controller is located. NVHBM has a narrower, custom PHY interface, with the memory controller function moved off the GPU. That leaves more room for AI-processing GPU cores as well as more HBM bandwidth.

However, NVHBM is specific to Nvidia and is not an industry standard.

Because Nvidia is such a massive HBM buyer, the three HBM suppliers are willing to go along with it and support Huang's colossus. According to reported Morgan Stanley data, Nvidia will purchase 37 percent of the global HBM supply in 2027. The suppliers have even accepted Nvidia dictating the semiconductor foundry process they must use: TSMC's 3nm. Reuters reported that Samsung will accede to this request. South Korea's Seoul Economic Daily said Samsung would still build JEDEC-standard HBM in-house for other customers.

Amazon's Annapurna Labs is also supporting NVHBM as an Nvidia NVLink partner.

**The JEDEC Conundrum**

JEDEC has not issued any public comment about NVHBM.

NVHBM is effectively an industry standard because Nvidia is such a dominant HBM buyer and there are only three HBM manufacturers, all of which are supporting Nvidia.

When JEDEC announced the HBM4 standard in April last year, the HBM subcommittee chair was Barry Wagner, Director of Technical Marketing at Nvidia. His continued chairmanship may now be untenable.

JEDEC could respond in one of three ways. It could accept Nvidia's encroachment on its HBM standard-setting scope and restrict its HBM specification work to the stack above the HBM base memory controller die. It could continue as before, with Nvidia absent. Or it could move in the NVHBM direction and add a base memory controller die with narrower, faster physical channels. This would build on the reduced HBM PHY pin count work it is already doing.