COMPARE / MI325X VS MI350X

MI325X vs MI350X

Specs from the chip makers' datasheets; prices from the live Compute Cafe panel, refreshed hourly. Every rate is an on-demand published price with a source receipt.

CHEAPEST MI325X
$1.590/hr
SpotGPUs · median $2.00 · 4 providers
CHEAPEST MI350X
$4.000/hr
CloudRift · median $4.00 · 1 providers
CHEAPER RIGHT NOW
MI325X
live panel, 2026-09-27
MI325XMI350X
MAKERAMDAMD
ARCHITECTURECDNA 3CDNA 4
LAUNCHED20242025
MEMORY256 GB HBM3e288 GB HBM3E
MEMORY BANDWIDTH6 TB/s8 TB/s
TENSOR COMPUTE1.3 PFLOPS FP16/BF16 dense2.3 PFLOPS FP16/BF16 matrix (4.6 PFLOPS FP8, 9.2 PFLOPS MXFP4)
POWER DRAW1,000 W (OAM)1,000 W (OAM, air-cooled; MI355X 1,400 W liquid)
INTERCONNECTInfinity FabricOAM (UBB), Infinity Fabric xGMI

HOW TO CHOOSE

MI325X vs MI350X
On paper - MI325X (CDNA 3): 256 GB HBM3e, 6 TB/s bandwidth, 1,000 W (OAM). MI350X (CDNA 4): 288 GB HBM3E, 8 TB/s bandwidth, 1,000 W (OAM, air-cooled; MI355X 1,400 W liquid). The adjacent-generation AMD question: CDNA 3 refresh versus CDNA 4 - whether the newer chip's memory and compute jump beats the 325X's panel discount.
MI350X vs MI325X
On paper - MI325X (CDNA 3): 256 GB HBM3e, 6 TB/s bandwidth, 1,000 W (OAM). MI350X (CDNA 4): 288 GB HBM3E, 8 TB/s bandwidth, 1,000 W (OAM, air-cooled; MI355X 1,400 W liquid). The adjacent-generation AMD question: CDNA 3 refresh versus CDNA 4 - whether the newer chip's memory and compute jump beats the 325X's panel discount.
Workload fit
MI325X: large-model inference, training, cuda-locked stacks. MI350X: llm training, large-model inference, hpc.

MI325X VS MI350X - FAQ

Which is cheaper to rent, MI325X or MI350X?
By today's published on-demand rates: MI325X starts at $1.590/GPU/hr (median $2.00, 4 providers) and MI350X starts at $4.000/GPU/hr (median $4.00, 1 providers). The cheapest live listing is the MI325X at SpotGPUs. Figures recompute from the panel on every refresh.
MI325X or MI350X - which should I rent?
MI325X: The MI325X is the memory-maximized refresh of the MI300X: same CDNA 3 compute, a third more HBM3e at 256 GB. MI350X: The MI350X is AMD's fourth-generation Instinct flagship: 288 GB of HBM3E at 8 TB/s with FP16 throughput at 2.3 PFLOPS, matching the MI355X's compute at a lower 1,000 W air-cooled power envelope. The side-by-side spec table and live price rows above give the full picture.
Where can I rent MI325X and MI350X?
4 providers publish MI325X rates and 1 publish MI350X rates on Compute Cafe. Every row links to the provider's own published price - committed-use and reserved rates are excluded.

Full pricing: MI325X · MI350X · all comparisons