H100 SXM
training · high-throughput inference
$1,352/ 30 days
View GPU H100 SXM vs H100 PCIe
Compare memory, node configurations, and fixed weekly and monthly cost before choosing between NVIDIA H100 SXM and NVIDIA H100 PCIe.
H100 SXM provides the same amount of VRAM per card. H100 PCIe costs $160 less per 30-day term . Choose based on whether memory headroom or lower fixed cost is the binding constraint.
training · high-throughput inference
fine-tuning · inference
| Specification | H100 SXM | H100 PCIe |
|---|---|---|
| Architecture | Hopper | Hopper |
| VRAM | 80 GB HBM3 | 80 GB HBM2e |
| Host resources / card | 24 vCPU · 192 GB RAM | 20 vCPU · 160 GB RAM |
| Local NVMe / card | 3.84 TB | 2 TB |
| 7-day price / card | $380 | $336 |
| 30-day price / card | $1,352 | $1,192 |
| 4-GPU node / 30 days | $4,975 | $4,387 |
| Catalog quantity | 690 cards | 510 cards |
Workload decision · updated 2026-09-04
Both models use Hopper and offer 80 GB, but the memory subsystem and host integration differ. The useful question is whether your workload needs a connected multi-GPU platform or can run on one card or independent replicas.
Choose H100 SXM for a communication-sensitive model split across several GPUs, provided that the allocation actually exposes the expected HGX fabric. Memory-bandwidth-sensitive work is another reason to test SXM against PCIe.
Choose H100 PCIe when one GPU is sufficient, workers operate independently, or a lower-priced Hopper configuration meets the measured latency and throughput requirement. Do not assume that multiple PCIe cards create an all-to-all GPU fabric.
| Manufacturer specification | H100 SXM | H100 PCIe |
|---|---|---|
| Architecture | Hopper | Hopper |
| Memory | 80 GB HBM3 | 80 GB HBM2e |
| Published memory bandwidth | 3.35 TB/s | 2.0 TB/s |
| Form factor | SXM | Dual-slot PCIe Gen5 |
| Reference system connection | HGX NVLink/NVSwitch platforms | PCIe; optional bridge to an adjacent H100 |
Published accelerator specifications, not a measurement of AnchorGPU infrastructure. Host topology and delivered fabric must be confirmed separately.
Bottom line. PCIe is the lower-cost Hopper choice for independent work. SXM is a candidate for closely coupled workloads only when the delivered topology supports the intended strategy.
Catalog cost difference: $44 per 7 days and $160 per 30 days. Prices are read from the same catalog as the configurator.
Manufacturer specifications inform the comparison. Recommendations are workload-dependent; no AnchorGPU performance benchmark is claimed.
NVIDIA H100 specifications ↗NVIDIA H100 PCIe product brief ↗