Nvidia hits roadblocks as it launches first-generation Blackwell B200 series processors Issues affecting productionand some unconfirmed reports overheat The server also appeared. However, it is reported semi-analyticallooks like Nvidia’s second generation Blackwell B300 Series Processors Just around the corner. Not only do they have greater memory capacity, they also provide a 50% performance increase with only 200W of additional TDP.
Nvidia’s B300 series processors feature a significantly tweaked design and will still use TSMC’s 4NP manufacturing process (a 4nm-level node tuned for enhanced performance by Nvidia), but the report claims they will offer 50% more computing than the B200 Performance series processors. This performance increase comes at the cost of a 1,400W TDP, which is only 200W higher than the GB200. SemiAnalysis claims that the B300 will be available approximately half a year after the launch of the B200.
The second major improvement to the Nvidia B300 series is the use of the 12-Hi HBM3E memory stack, which will provide 288 GB of memory and 8 TB/s bandwidth. Enhanced memory capacity and higher compute throughput will enable faster training and inference and reduce inference costs by three times, as the B300 will handle larger batch sizes and support extended sequence lengths while addressing user interaction issues delay problem.
In addition to higher computing performance and more memory, Nvidia’s second-generation Blackwell machines may also use the company’s 800G ConnectX-8 NIC. The NIC has twice the bandwidth of the current 400G ConnectX-7 and has 48 PCIe lanes compared to its predecessor’s 32 lanes. This will provide significant scale-out bandwidth improvements for new servers, which is a win for large clusters.
Another major enhancement for the B300 and GB300 is that Nvidia will reportedly redesign the entire supply chain compared to the B200 and GB200. The company will no longer try to sell entire reference motherboards or entire server bays. Instead, Nvidia will only sell the B300 with SXM Puck modules, Grace CPU, and Axiado Host Management Controller (HMC). As a result, more companies will be allowed to participate in the Blackwell supply chain, which is expected to make Blackwell-based machines easier to use.
With the B300 and GB300, Nvidia will give its hyperscalers and OEM partners more freedom to design Blackwell machines, which will affect its pricing and perhaps performance.