CARDINALH100 NVL 94GB
Choose between SXM modules, standard PCIe cards and NVL cards before comparing offers
SPECIFICATIONS
| Memory per accelerator | 94 GB HBM3 |
|---|---|
| Memory bandwidth | 3900 GB/s |
| Form factor | Dual-slot PCIe card |
| Maximum accelerator power | 400 W |
Manufacturer specifications describe the product, not the condition of an offered unit. Checked 2026-09-08
COMPUTE SPECIFICATIONS
Manufacturer peak arithmetic per accelerator. These values do not predict application throughput or latency.
| Precision | Execution | Peak rate | Sparsity | Reference |
|---|---|---|---|---|
| FP64 | Vector | 30 TFLOPS | Dense | Source ↗ConditionsManufacturer peak per GPU; not sustained workload performance |
| FP64 | Tensor | 60 TFLOPS | Dense | Source ↗ConditionsManufacturer peak per GPU; not sustained workload performance |
| FP32 | Vector | 60 TFLOPS | Dense | Source ↗ConditionsManufacturer peak per GPU; not sustained workload performance |
| TF32 | Tensor | 835 TFLOPS | Structured sparse | Source ↗ConditionsManufacturer specification marked with sparsity; full GPU, not an NVL pair or MIG slice |
| BF16 | Tensor | 1671 TFLOPS | Structured sparse | Source ↗ConditionsManufacturer specification marked with sparsity; full GPU, not an NVL pair or MIG slice |
| FP16 | Tensor | 1671 TFLOPS | Structured sparse | Source ↗ConditionsManufacturer specification marked with sparsity; full GPU, not an NVL pair or MIG slice |
| FP8 | Tensor | 3341 TFLOPS | Structured sparse | Source ↗ConditionsManufacturer specification marked with sparsity; full GPU, not an NVL pair or MIG slice |
| INT8 | Tensor | 3341 TOPS | Structured sparse | Source ↗ConditionsManufacturer peak per GPU; not sustained workload performance |
- Peak arithmetic rates are not token throughput, latency or a guarantee of application performance
- Structured-sparse rates require eligible sparsity and supported kernels; do not compare them directly with dense rates
- Dense low-precision tensor rates are not derived by halving sparse ratings
- Figures are per GPU; do not multiply them and describe a pair as a single accelerator
PLATFORM AND SOFTWARE
- 94 GB is per card; 188 GB describes two cards together
- Ask whether bridges and both cards are included
- Confirm CUDA, driver and framework compatibility against the intended server image
- Ask whether any software subscription can transfer with the particular used offer
Manufacturer configurable TDP range
INTERCONNECT
NVLink
600 GB/s. Manufacturer aggregate bidirectional bandwidth; topology depends on platform
PCIe Gen5 x16
128 GB/s. Theoretical aggregate bidirectional interface bandwidth
MANUFACTURER SOURCES
NVIDIA H100 specifications ↗NVIDIA · Checked 2026-09-08NVIDIA H100 NVL product brief ↗NVIDIA · Checked 2026-09-08HOW CARDINAL HELPS
Share your quantity, destination and server model, if known. We help turn those requirements into a clearly specified request.
- Help confirm the exact configuration and platform requirements
- Request supplier records for condition, unit identifiers and available testing evidence
- Clarify included accessories, delivery and proposed warranty terms before you decide
Any additional testing or guarantees must be agreed in the written offer.