← AI Hashrate 中文

MI300X 192GB vs Tesla P4 8GB for Local LLMs

The MI300X 192GB has about 2660% more memory bandwidth (5,300 vs 192 GB/s). Batch-1 decode tracks bandwidth closely, so a model that fits both cards should decode roughly 2660% faster on the MI300X 192GB. The MI300X 192GB fits more of the catalog at Q4/8K — 57 vs 16 of 69 models (192 vs 8 GB VRAM). Largest model that fits the MI300X 192GB but not the Tesla P4 8GB: MiMo-V2.5 (310.0B). MSRP is $15,000 for the MI300X 192GB vs $80 for the Tesla P4 8GB. MSRP is a launch list price, not live retail — Amazon prices move; click through for the current price. Not retail-shoppable: MI300X 192GB (datacenter-class; cloud rental is the realistic way to use it). In the Q4/8K head-to-head below (39 shared models), the MI300X 192GB posts the higher tok/s in 39 rows vs 0. Methodology.

Spec comparison

MI300X 192GBTesla P4 8GB
VendorAMDNVIDIA
VRAM192 GB8 GB
Memory typeHBM3GDDR5
Memory bandwidth5,300 GB/s192 GB/s
TDP750 W75 W
MSRP (list)$15,000$80
Buy—Search Amazon (affiliate)

MSRP is the launch list price, not live retail. Buy links are Amazon affiliate links — the price after you click is the live price.

Q4 / 8K head-to-head — top 15 of 39 shared models

ModelParams (B)MI300X 192GB tok/sMI300X 192GB fitsTesla P4 8GB tok/sTesla P4 8GB fitsWinner
MiniCPM5-1B1.083122.9 est.Yes113.1 est.YesMI300X 192GB
Qwen3.5-2B2.01686.4 est.Yes61.1 est.YesMI300X 192GB
DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-3.8-mlx2.31466.4 est.Yes53.1 est.YesMI300X 192GB
Qwen3.6-35B-A3B36.01124.2 est.Yes17.0 est.NoMI300X 192GB
Qwen3.5-35B-A3B34.71124.2 est.Yes18.2 est.NoMI300X 192GB
GLM-4.7-Flash31.21124.2 est.Yes21.0 est.NoMI300X 192GB
Qwen3-30B-A3B-Instruct-250730.51124.2 est.Yes21.9 est.NoMI300X 192GB
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF1630.01124.2 est.Yes24.1 est.NoMI300X 192GB
Ornith-1.0-35B35.01124.2 est.Yes17.0 est.NoMI300X 192GB
qwen35b-a3b-fable-sft-abliterated34.71124.2 est.Yes17.3 est.NoMI300X 192GB
Llama-3.2-3B-Instruct3.21054.0 est.Yes38.2 est.YesMI300X 192GB
gpt-oss-20b21.5936.9 est.Yes33.6 est.NoMI300X 192GB
gemma-4-26B-A4B-it25.2887.6 est.Yes23.9 est.NoMI300X 192GB
Qwen3-4B-Instruct-25074.0843.2 est.Yes30.5 est.YesMI300X 192GB
NVIDIA-Nemotron-3-Nano-4B-BF164.0843.2 est.Yes30.5 est.YesMI300X 192GB

est. = bandwidth-based estimate; measured = curated real benchmark that overrides the estimate. Fits = weights + KV(8K) + 1 GB overhead ≤ 95% of VRAM.

Fit coverage at Q4 / 8K

MI300X 192GBTesla P4 8GB
Catalog models that fit57 of 6916 of 69
Largest exclusive fitMiMo-V2.5 (310.0B)—

MI300X 192GB details → · Tesla P4 8GB details →