AMD Instinct MI325X 8-GPU Platform
Server nodeShippingCDNA 32025
AMD's latest 8-GPU OAM platform with MI325X accelerators. 256 GB HBM3e per GPU for the largest memory footprint in its class. Infinity Fabric interconnect. Key alternative to NVIDIA for hyperscalers seeking supply diversification. Dell, HPE, and Supermicro offer reference designs.
8
GPUs per system
48 FP8 PFLOPS
Total HBM
2 TB
Host memory
3 TB
Interconnect
Infinity Fabric
4.8 TB/s
Networking
400 Gbps
Storage
30 TB NVMe SSD
Form factor
8U node (OAM)
Weight
150 kg
Rack units
8U
Performance
Manufacturer datasheet values · aggregate system compute
| FP4 PFLOPS | TBD |
| FP8 PFLOPS | 48 |
| FP16 PFLOPS | 24 |
| BF16 PFLOPS | 24 |
| Training effective PFLOPS | 18 |
Power and cooling
Thermal envelope · cooling requirements · efficiency
Rack power
10 kW
Per GPU
750 W
Cooling
air
PUE estimate
1.3
Power draw relative to tracked systems10 kW / 2500 kW max
4.8 FP8 PFLOPS per kW · average across all systems is 4.81
TCO analysis
Hardware amortized over 3 years · power at $0.05/kWh
List price
$280,000
Per GPU effective
$35,000
Cost per GPU per month
$972
TCO per PFLOPS per year
$2,063
PFLOPS per kW
4.8
77% below the average TCO of $9,046/PFLOPS/year across all tracked systems
Available from
AMDD
DellH
HPES
SupermicroKnown deployments
Disclosed in press releases, SEC filings, and conference talks
Quantity
ND MI300X v5 instances
Source
Azure blog
Quantity
inference fleet diversification
Source
Meta earnings callSources
Every data point on this page is reproducible
Other AI systems
Compare across the system landscape
TPU v5p Pod8960x Google TPU v5p · 8100 PFLOPS
Pod / clusterShipping
DGX GB300 NVL7272x NVIDIA B300 · 1440 PFLOPS
Full rackAnnounced
DGX GB200 NVL7272x NVIDIA B200 · 720 PFLOPS
Full rackShipping
TPU v6e Pod256x Google TPU v6e · 230 PFLOPS
Pod / clusterShipping
Maia 100 Rack32x Microsoft Maia 100 · 96 PFLOPS
Full rackRamping
DGX B2008x NVIDIA B200 · 80 PFLOPS
Server nodeShipping