[Citrini Analyst Questions SanDisk HBF Performance Comparison: Demonstration Parameters May Underestimate HBM's Actual Capabilities] BlockBeats News, August 14: Citrini analyst Zephyr published an article questioning the parameter settings in SanDisk's performance comparison of HBF and HBM during its Investor Day presentation, suggesting that the company may have 'deliberately underestimated HBM's performance' in the demonstration. He pointed out that SanDisk set the total bandwidth of both HBM and HBF to 12.8TB/s in the comparison, equivalent to 1.6TB/s per stack, and used the bfloat16 format to run the Qwen3-480B-A35B model. Zephyr noted that current model inference more commonly uses FP4 or FP8, meaning the required capacity for the model under different quantization methods is approximately 240GB to 480GB. Zephyr further stated that SanDisk fixed the single GPU capacity of HBM at 192GB, whereas a 16-layer HBM4E configuration with 8 stacks can achieve a capacity of 512GB and provide about 32TB/s bandwidth—roughly three times the bandwidth of the parameters used in SanDisk's demonstration. He believes that these parameter choices render SanDisk's claim that 'HBF requires only 1 to 4 GPUs, while HBM requires 8 GPUs' insufficient to represent the actual performance of higher-spec HBM configurations. [Original Link]