Compare
Side-by-side spec sheets for chips, systems and models. Every figure keeps its basis label — a vendor claim next to an estimate is labeled as exactly that.
| NVIDIA B200 Blackwell: two reticle-limit dies fused into one GPU — the 2025 workhorse. | AMD Instinct MI325X MI300X with 256 GB of HBM3E — the memory-capacity flagship of 2025. | |
|---|---|---|
| Overview — Designer | NVIDIA | AMD |
| Overview — Architecture | Blackwell | CDNA 3 |
| Overview — Announced | March 2024 (GTC); volume through 2025 | October 2024; systems from Q1 2025 |
| Overview — Status | Shipping | — |
| Manufacturing — Foundry / process | TSMC 4NPVendor spec | — |
| Manufacturing — Transistors | 208 billion (2 × 104B dies)Vendor spec | — |
| Manufacturing — Die-to-die link | NV-HBI — 10 TB/sVendor spec | — |
| Manufacturing — Packaging | CoWoS-L, 8 HBM3E stacksVendor spec | — |
| Memory — Memory | 192 GB HBM3E (8 × 24 GB stacks)Vendor spec | 256 GB HBM3EVendor spec |
| Memory — Memory bandwidth | 8 TB/sVendor spec | 6 TB/sVendor spec |
| Compute — BF16 tensor (dense) | 2.25 PFLOPSVendor spec | — |
| Compute — FP8 tensor (dense) | 4.5 PFLOPSVendor spec | — |
| Compute — FP4 tensor (dense) | 9 PFLOPSVendor spec | — |
| Power — TDP | up to ~1,000 (HGX, configurable); 1,200 in GB200 configurations WVendor spec | 1,000 WVendor spec |
| Physical — Form factors | SXM (HGX B200), GB200 superchip | — |
| Physical — Interconnect | NVLink 5 — 1.8 TB/sVendor spec | — |
| Price — Price | ~$30,000–40,000 per GPU indicated by NVIDIA leadership in 2024Reported | — |
| Compute — Compute | — | Same published CDNA3 throughput as MI300X (FP8 dense 2,615 TFLOPS)Vendor spec |
"—" means that record does not publish this line; different vendors disclose different things, and we do not fill gaps with guesses.