-
- News
- Books
Featured Books
- pcb007 Magazine
Latest Issues
Current IssueInner Layer Precision & Yields
In this issue, we examine the critical nature of building precisions into your inner layers and assessing their pass/fail status as early as possible. Whether it’s using automation to cut down on handling issues, identifying defects earlier, or replacing an old line...
Engineering Economics
The real cost to manufacture a PCB encompasses everything that goes into making the product: the materials and other value-added supplies, machine and personnel costs, and most importantly, your quality. A hard look at real costs seems wholly appropriate.
Alternate Metallization Processes
Traditional electroless copper and electroless copper immersion gold have been primary PCB plating methods for decades. But alternative plating metals and processes have been introduced over the past few years as miniaturization and advanced packaging continue to develop.
- Articles
- Columns
Search Console
- Links
- Media kit
||| MENU - pcb007 Magazine
NVIDIA’s New B200A Targets OEM Customers; High-End GPU Shipments Expected to Grow 55% in 2025
August 7, 2024 | TrendForceEstimated reading time: 1 minute
Despite recent rumors speculating on NVIDIA’s supposed cancellation of the B100 in favor of the B200A, TrendForce reports that NVIDIA is still on track to launch both the B100 and B200 in the 2H24 as it aims to target CSP customers. Additionally, a scaled-down B200A is planned for other enterprise clients, focusing on edge AI applications.
TrendForce reports that NVIDIA will prioritize the B100 and B200 for CSP customers with higher demand due to the tight production capacity of CoWoS-L. Shipments are expected to commence after 3Q24. In light of yield and mass production challenges with CoWoS-L, NVIDIA is also planning the B200A for other enterprise clients, utilizing CoWoS-S packaging technology.
TrendForce anticipates that the B200A will have a lower thermal design power (TDP) compared to the B200, allowing it to use air-cooling solutions with the GB rack. This is expected to mitigate delays caused by the complexities of liquid-cooling designs in 2025. The B200A will feature 4 HBM3e 12hi memory modules, with a total capacity of 144 GB. OEMs are expected to receive the B200A chips after 1H25, providing a suitable timeline to prevent conflicts with the ramp-up of the H200 in 3Q24 and ensuring broader market adoption.
Blackwell to account for over 80% of NVIDIA’s high-end GPU shipments in 2025
TrendForce’s observations reveal that NVIDIA’s high-end GPU shipments in 2024 will primarily be from the Hopper platform. The H100 and H200 models will be shipped to North American CSPs and OEMs, while Chinese customers will primarily receive AI servers equipped with the H20. The H200 is expected to ramp up in 3Q24, becoming NVIDIA’s mainstream model through 2025.
TrendForce notes that the Blackwell series will still be in its early shipping stages in 2024. By 2025, Blackwell will become the main shipment driver, with the high-performance B200 and GB200 rack meeting the high-end AI server demands of CSPs and OEMs. The B100, a transitional product with lower power consumption, will gradually be replaced by the B200, B200A, and GB200 rack after fulfilling existing CSP orders. TrendForce estimates that by 2025, the Blackwell platform will account for over 80% of NVIDIA’s high-end GPU shipments, driving the annual growth rate of NVIDIA’s high-end GPU series shipments to 55%.