NVIDIA · 2026

NVIDIA Vera Rubin

NVIDIA's next-generation AI platform unveiled at CES 2026: TSMC N3P logic, CoWoS-L packaging, HBM4 memory, and rack-scale liquid cooling. Available in three variants — NVL72 (mainstream), NVL144 CPX (prefill-optimized), and NVL576 Ultra (silicon photonics).

主要 variants

點任一張卡片進入該 variant 的完整 BOM 與供應商分析。

3 個 variant並排比較 BOM →

Vera Rubin NVL72

2026 H2

72 Rubin GPUs + 36 Vera CPUs in one liquid-cooled rack. 3.6 EFLOPS FP4, 20.7 TB HBM4, 260 TB/s NVLink-6. Mainstream Rubin rack.

Liquid-cooled100 kWTSMC N3P

2026 H2low

~2.5k 台 (1.5k–4.0k) · ASP ~3.5M USD

查看 BOM →

Vera Rubin NVL144 CPX

2027 H1

Prefill-optimized rack with 144 GPUs. 8 EFLOPS, 100 TB fast memory, 1.7 PB/s aggregate memory bandwidth. Paired with NVL72 racks for decode.

Liquid-cooled180 kWTSMC N3P

2027 H1rumor

~1.2k 台 (500–2.0k) · ASP ~5.5M USD

查看 BOM →

NVL576 Ultra

2027 H2

576 Rubin Ultra GPUs in a single rack at ~600 kW. First NVIDIA system shipping with co-packaged optics (CPO) NVSwitch, replacing pluggable transceivers. Densest AI training pod NVIDIA has shipped.

Liquid-cooledCo-packaged optics600 kWTSMC N3P

2027 H2rumor

~200 台 (50–400) · ASP ~28.0M USD

查看 BOM →