← All chips

AI accelerator · China

MLU370-X8

Dual-die training/inference accelerator card from Cambricon's third-generation MLUarch03 architecture, positioned in the mid-to-high domestic cloud segment. Officially specified at 256 INT8 TOPS with 48GB LPDDR5 and MLU-Link interconnect.

ChinaShippingCambriconMLU familyLaunched 20237nm

Latest observation ~$2.1K (from ¥15,000 at 2026 avg 7.25 CNY/USD) (per card) as of 2026-08 · researched 2026-08-27

~$2.1Klatest observation · reseller listing · 2026-08 · from ¥15,000
1dated price observations on record
6sourced specs (all linked below)

Specifications

Sourced specs

SpecValueSource
INT8 compute256 TOPSvia cambricon.com2026-08
FP16 compute96 TFLOPS (BF16 96 TFLOPS, FP32 24 TFLOPS, INT16 128 TOPS)via cambricon.com2026-08
Memory48GB LPDDR5, 614.4 GB/svia cambricon.com2026-08
InterconnectMLU-Link, 200 GB/s aggregate bidirectional (4 ports x 16 lanes @ 50 Gbps)via cambricon.com2026-08
TDP250 W max, dual-slot full-height passivevia cambricon.com2026-08
Host interfacePCIe Gen4 x16via cambricon.com2026-08

Pricing

Price history, newest first

DatePriceTypeSourceConfidence
2026-08~$2.1K from ¥15,000 at 2026 avg 7.25 CNY/USD per cardTaobao listings for MLU370-X8/S4 cards from ~¥15,000; low volumes sold, B2B is the primary channelReseller listingvia Taobao listingslow

vendor-official or multiple independent reports · single major report or several secondary · estimate or lone secondary source

Prices in USD, converted at each observation year's average exchange rate; the original figure and rate print under every converted price. How prices are sourced and graded ↗

Context

Notes

  • Card announced 2023 (exact month approximate); no official public price — enterprise procurement via Cambricon/agents.

Where it sits

In the stack

All specs and prices link to their original sources, dated as observed. Figures are market observations, not offers; treat each as of its date.