EXACT VARIANT REFERENCE
A30 PCIE 24GB
24GB HBM2 option for inference partitioning and compute workloads
SPECIFICATIONS
| Memory per accelerator | 24 GB HBM2 |
|---|---|
| Memory bandwidth | 933 GB/s |
| Form factor | Dual-slot, full-height, full-length PCIe card |
| Maximum accelerator power | 165 W |
Manufacturer specifications describe the product, not the condition of an offered unit. Checked 2026-09-08
SERVER INTEGRATION
| Cooling | Passive heatsink requiring system airflow. Source ↗ |
|---|---|
| Slot width | Dual-slot, full-height, full-length 10.5-inch reference card. Source ↗ |
| Auxiliary power | One CPU eight-pin auxiliary power connector on the east edge of the board. Source ↗ |
| MIG | Up to four GPU instances; select a supported profile. This does not establish virtual GPU software entitlement. Source ↗ |
| ECC | Enabled by default; configurable through software. HBM2 has native ECC support. Source ↗ |
| NVLink bridge | One bridge supported; clearance and slot spacing must match the installation. Source ↗ |
| Board weight | 1240 grams excluding bracket, extenders and bridge. Source ↗ |
| Memory | 24GB HBM2 per accelerator. Source ↗ |
PLATFORM AND SOFTWARE
- Passive cooling requires qualified server airflow
- CPU 8-pin auxiliary power connector
- MIG supports up to four GPU instances, subject to supported profiles and software
- Check the exact board part number for CEC security-controller differences
- Nominal height is not transcribed from the mechanical drawing; consult the linked brief
- Confirm the exact driver, CUDA runtime, framework and server operating-system combination before purchase
- Historical minimum software versions in the product brief do not establish current release qualification
- vGPU and enterprise software entitlements must be checked separately from ownership of the card
Maximum board rating, not complete system consumption
INTERCONNECT
PCIe Gen4 x16
Host interface; rate not normalized from this product brief
NVLink
200 GB/s. Total maximum Rx plus Tx bandwidth; one bridge to one adjacent A30, not an arbitrary all-to-all fabric
MANUFACTURER SOURCES
NVIDIA A30 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.