Skip to main content

5 posts tagged with "MLU690"

寒武纪 MLU690 芯片详解

View all tags

寒武纪业绩兑现期：从扭亏 20.59 亿到下一代涨价 20%-30% 的逻辑

· 3 min read
AI Compute Cards Wiki Editorial
Industry Research Team

过去国产 AI 芯片常被诟病「故事丰满、报表骨感」。寒武纪 2025 年报与 2026 一季报给出的是另一组数字：全年扭亏为盈、单季营收同比增近 160%、尚未发布的产品先涨价两三成。国产算力正在进入业绩兑现期。

财务数字一览​

指标数值口径
2025 全年营收64.97 亿元（同比 +453.21%）年报
2025 归母净利20.59 亿元（扭亏为盈）年报
2026Q1 营收28.85 亿元（同比 +159.56%）一季报
市值曾超万亿元（2026 年 6 月）二级市场

单季 28.85 亿、全年 64.97 亿——2026 年一季度一个季度的收入接近 2025 全年的一半，增长斜率仍在变陡。

MLU690：从纸面到量产​

思元690（MLU690）2026 年初量产后，进展沿两条线推进：

  • 产品线：双 die Chiplet 封装、中芯国际 N+2 制程，FP16 算力 700+ TFLOPS 为当前国产纸面算力天花板之一；2026-09 价格口径为官方指导价 190,000 元、市场均价 175,000 元（市场流传价，非公司公告价）。
  • 集群线：WAIC 2026 落地 256 卡规模超节点集群方案，已在多地智算中心部署；运营商、金融、互联网行业规模化部署。

涨价的供需逻辑​

据 Reuters 2026-09-10 报道，受 HBM 成本飙升影响，寒武纪对下一代处理器（暂称思元790，预计 2026Q4 出货、2027 放量）的意向报价较两个月前上调 20%-30%。寒武纪董秘办回应「未发布过相关公告，一切以官方公告为准」——即绝对金额是市场流传口径，相对涨幅是 Reuters 报道口径，引用时务必分层。

涨价背后是三重叠加：HBM 等关键供应紧张（与昇腾 950DT 同因）；国内大模型训练与推理国产化率提升带来的需求确定性；智算中心集中采购把国产加速卡从「备选」变「必选」。订单排到数月之后时，涨价既是成本传导，也是筛选客户、平衡产能的手段。

与华为的双头格局​

徐直军 10 月 6 日在 HC 大会确认昇腾中国份额已超英伟达（华为统计口径），同时强调「能满足国内需求还远远不足」。这决定了国产算力市场的基本格局：昇腾供给受制于产能，寒武纪等厂商承接溢出需求。对采购方，昇腾渠道的排队周期反而给了思元690、MTT S5000 等替代方案真实的验证窗口。

风险提示​

  • 思元690 的千卡以上分布式训练损耗仍高于成熟生态（MLU-Link 与 NVLink 的工程成熟度差距），单机 8 卡与小集群表现优异，大规模预训练场景需实测
  • 价格均为渠道或市场流传口径，采购以正式报价单为准
  • MLU-Link 890Gbps 的互联规格与 NVLink 4 900GB/s 接近，但软件栈成熟度差异无法从规格表读出

相关阅读​

寒武纪 MLU690 规模部署与订单交付最新进展

· 5 min read
AI Compute Cards Wiki Editorial
Industry Research Team

国产 AI 芯片市场对 MLU690 的期待，从 2025 年一路延续到 2026 年。但据公开报道，MLU690 在 2026 年并没有形成大规模出货计划——寒武纪全年晶圆资源优先保障已成熟量产的 MLU580 与 MLU590。摩根士丹利 2026 年 10 月的研报也确认了这一点：MLU690 系列（特别是 MLU692）收入确认延迟，构成三季度业绩不及预期的主因。本文梳理 MLU690 的真实交付进展。

MLU690 交付现状：产能优先给成熟型号​

据公开报道，寒武纪 2026 年的产能分配策略已经明确：全年晶圆资源优先供给成熟量产的 MLU580、MLU590 两款芯片，MLU690 中芯版本暂不规划批量出货，也没有主动向客户推广。核心制约仍是产能瓶颈，即便拿到客户订单也无法按时交付，因此公司索性收缩 MLU690 的市场投放节奏。

摩根士丹利的研报从财务角度印证了同一件事：预计寒武纪 2026 年第三季度收入为 36.5 亿元人民币，低于此前预期的 47.7 亿元。不及预期的主要原因在于 MLU690 系列（特别是 MLU692）的收入确认延迟以及 MLU580 的交付量有限。寒武纪在第三季度已将 MLU692 交付给主要的 CSP（云服务提供商）进行测试，但预计该产品的大部分收入将在第四季度才能确认。

这就是 MLU690 当前的真实状态——不是"卖不动"，而是"没有足够的晶圆去卖"。

型号2026 年产能地位交付状态
MLU580主力，产能倾斜交付量有限，三季度提价后毛利率仍偏低
MLU590主力，产能倾斜上半年内可交付
MLU690 / MLU692产能优先级靠后已交付主要 CSP 测试，收入多在四季度确认
MLU790（下一代）尚未推进受 690 验证进度影响，节奏更谨慎

据公开报道，MLU690 的合同虽在上半年签订，但交付预计在第三季度。这与摩根士丹利所述"大部分收入在第四季度确认"基本一致。

央国企市场：可见的订单空间​

在消费级与互联网市场之外，央国企与训练场项目是 MLU690 的主要落地方向。据公开报道，2026 年央国企市场的可见商机空间预计在 10 万至 20 万卡，其中上半年预计可落地订单约为 5 万卡。这 5 万卡中，有 3 万卡将用于与京能合作的国家训练场项目，另外 2 万卡来自其他央国企计算项目扩容（如浙江台州、昆山的项目）以及上海、安徽合肥等地的训练场和电信运营商项目。

值得注意的是，央国企订单的竞争格局并不宽松。据公开报道，在央国企市场寒武纪并非绝对的第二供应商，市场地位大约能排进前五；在信创领域华为是绝对的第一供应商，占据超过 60% 的市场份额。对寒武纪构成最大竞争压力的是平头哥和昆仑芯——这两家公司背靠阿里云和百度云，具备算力消纳能力，能通过云服务输出算力，而寒武纪无法做出此类承诺。

超节点布局：停滞状态​

超节点是 2026 年国产算力的主战场，但寒武纪在这条线上基本缺席。据公开报道，寒武纪自己没有做超节点，也没有明确的自建超节点计划。当前主要是给一些做超节点的厂商提供产品做验证和适配，2026 年还没有客户明确要求寒武纪去部署超节点，而且产能也不支持这么做。

现有产能仅能勉强满足字节的八卡基础服务器订单，字节自身也没有超节点部署计划，相关计划更偏 2027 年。腾讯和阿里虽有超节点方向的规划，但寒武纪没有办法保证供货，所以没有真正推进到客户部署层面。目前更多是与 OEM 厂商做平台适配，比如阿里的盘古、浪潮、新华三等平台已经做过适配，但没有量产安排。

还有一个被忽视的连锁反应：MLU690 的延迟量产会进一步拖累下一代 MLU790 的流片进度。MLU690 本身承担 790 前置大规模商用验证任务，缺少足量客户真实场景数据迭代，下一代芯片的研发与流片节奏会持续延后。

研判​

寒武纪当前的处境可以用一句话概括：需求不缺，晶圆不够。据公开报道，摩根士丹利预计寒武纪 2027 年向 CSP 客户的出货量将接近 35 万颗，主要得益于 MLU692 的产能爬坡和下一代 MLU580 产品的推出；代工方面主要依赖中芯国际的 N+2/N+1 产能。HBM3e 仍是关键供应瓶颈，但供应链调研显示寒武纪拥有一定库存，并有望随国内产能爬坡获得长鑫存储的一部分 HBM3e 产量。

财务侧的压力同样真实：HBM 成本上升对单位经济效益造成压力，MLU580 即使三季度提价，毛利率仍将保持在相对较低水平。摩根士丹利维持增持评级，目标价从 1408 元上调至 1700 元，但明确将三季度称为"业绩低谷"。

对客户而言，MLU690 在 2026 年不是可规模化规划的选项；2027 年随着产能爬坡与 790 产品线就位，它才会真正进入放量期。这一年多的窗口期，寒武纪要证明的是晶圆产能能否转化为可持续的交付能力。

寒武纪 MLU690 深度解析：700+ TFLOPS FP16 与国产训练卡的寒武纪时刻

· 6 min read
AI Compute Cards Wiki Editorial
Industry Research Team

在华为昇腾与英伟达的双雄叙事之外，寒武纪正在写下国产 AI 芯片的第三条故事线：思元690 以 700+ TFLOPS FP16 登顶国产纸面算力，公司实现首次年度盈利，下一代芯片未发先涨 20%–30%。这个曾经靠情怀撑估值的"AI 芯片第一股"，第一次同时握住了技术、商业与定价权三张牌。

一、思元690 核心规格：196GB HBM3 撑起的算力天花板​

按本站芯片卡收录的规格，思元690（MLU690）采用双 die Chiplet 封装，2026 年初量产，核心参数如下：

项目参数
架构新一代 MLUarch 架构
制程中芯国际 N+2（5nm 级），亦见台积电 4nm 双 die 报道
显存196 GB HBM3
显存带宽3.35 TB/s
FP16/BF16700+ TFLOPS
INT42800+ TOPS
互联MLU-Link 890 Gbps/卡
TDP400–500 W
单价¥14–15 万

与上代思元590 相比，这是一次全面换代；与 H100 相比，则是典型的"错位竞争"：

指标思元690思元590H100 SXM对比 H100
显存196GB HBM396GB HBM2e80GB HBM32.45×
显存带宽3.35 TB/s~2 TB/s3.35 TB/s持平
FP16700+ TFLOPS345 TFLOPS989 TFLOPS~71%
INT8 推理反超（INT4 2800+ TOPS 口径）1380 TOPS1979 TOPS部分场景反超

196GB 显存是思元690 最锋利的差异化武器：单卡装下更大的模型与 KV Cache，在推理场景性价比突出；纯推理场景可达到 H100 的 80%–90%。短板同样清晰——FP16 训练算力与千卡以上分布式训练的互联工程成熟度。为此，WAIC 2026 上寒武纪同步落地了 256 卡规模超节点集群方案，已在多地智算中心部署，从"单卡算力"走向"系统级算力"。

二、架构演进：MLUarch 换代与 HBM3 上量​

思元690 的代际跃迁体现在三处：其一，MLUarch 架构换代（市场披露口径称新一代芯片采用 MLUarch 04），配合双 die Chiplet 封装，把晶体管预算投向张量算力与显存通道；其二，显存从 HBM2e 一举上量到 196GB HBM3，3.35 TB/s 带宽与 H100 持平——在国产供应链里率先解决"内存墙"；其三，MLU-Link 890 Gbps 卡间互联已接近 NVLink 4 的 900 GB/s 量级，为 256 卡超节点铺路。软件侧的 NeuWare 是国内少有的云边端统一栈：PyTorch/TensorFlow 适配版、NeuBLAS、NeuDNN，以及全球首批商用 MLIR 图编译推理框架 MagicMind。

供应链上，台积电因实体清单停供后，寒武纪全线转向中芯国际 N+2，成为中芯先进制程的最大客户之一；N+2 产能已锁定至 2027 年。

三、涨价 20%–30%：国产芯片第一次掌握定价权​

路透社 2026 年 9 月 10 日报道：寒武纪下一代芯片（市场暂称"690"）重新定价，较两个月前上涨 20%–30%。同期昇腾 950DT 报价突破 25 万元、沐曦、天数智芯等国产卡也集体上调 20%–30%——这不是寒武纪一家的孤例，而是 HBM 成本飙升叠加供不应求下的全行业现象。

值得注意的是官方态度：寒武纪董秘办回应"不要听信市场消息，一切以官方公告为准"。渠道价与官方定价之间存在明显温差——涨价是市场现实，但公司不愿为渠道炒作背书。这背后是国产 AI 芯片定价权的历史性变化：过去国产卡必须以"替代品折价"换取订单，如今供给稀缺让定价逻辑转向"稀缺溢价"。在华为、寒武纪、摩尔线程构成的国产三强格局中，寒武纪第一次拥有了"不降价也能卖动"的底气。

四、首次年度盈利：商业模式的拐点​

如果说涨价反映的是市场热度，盈利则验证了商业模式的成色。按 2026 年上半年行业复盘口径，寒武纪已实现首次年度盈利。配套数据同样亮眼：2026 年 Q1 营收 28.85 亿元，同比增长 160%；2026 年 6 月市值突破万亿元。客户结构上，字节跳动是最大客户（思元580/590/690 合计采购超 50 万颗，约占其采购的 65%），阿里、百度、中国移动相继进入订单名单。

"首次盈利"对国产 AI 芯片行业的标志意义在于：这个赛道第一次出现了不依赖补贴与融资、靠卖芯片本身赚钱的商业模式。研发投入可以自我造血，意味着产品迭代不再以资本耐心为限——这对动辄三年一代的芯片研发周期，是决定性的正循环。

五、命名之辨：思元系列与市场传闻的"690"​

最后回到一个细节：市场上热议的"690"，官方从未正式发布命名。本站芯片卡以"思元690"收录其规格（已量产口径），而路透报道中的"690"实为市场对下一代芯片的暂称，寒武纪对具体命名、规格与定价均保持谨慎，一切以官方公告为准。这种"市场抢跑、官方缄默"的组合本身就是一个信号：在供给稀缺的卖方市场里，厂商不再需要靠预热造势——把名字留到正式发布，反而是一种定价权。

小结​

思元690 代表的"寒武纪时刻"，本质上是三件事同时发生：技术上，700+ TFLOPS FP16、196GB HBM3 与 256 卡超节点让国产训练卡首次形成完整的系统级方案；商业上，首次年度盈利证明了国产 AI 芯片可以自我造血；市场地位上，20%–30% 的重新定价宣告国产芯片开始掌握定价权。当然，考验依然存在：千卡以上训练的互联工程成熟度、NeuWare 与 CUDA 的生态差距、以及下一代芯片能否如期放量。但至少在 2026 年秋天，国产 AI 芯片第一次同时拥有了技术、利润与议价能力。

China's Domestic AI Chip Triopoly (2026): Ascend, Cambricon, Moore Threads — Who Is the "China H100"?

· 7 min read
Charles Qing
AI Hardware Analyst

Against the backdrop of U.S. export controls, China's AI chip market is forming a "three-way standoff." This article compares the technical routes, product specs, software ecosystems, and commercial progress of the three major domestic AI chip vendors: Huawei Ascend, Cambricon MLU, and Moore Threads MTT.

Key Points​

  • Huawei Ascend: leader in domestic AI training chips; Ascend 950 in mass production; most mature software ecosystem
  • Cambricon MLU690: the "China H100," compute close to H200, clear efficiency advantage
  • Moore Threads MTT S5000: full-function GPU route; achieved Day-0 support for Qwen3.5 and GLM-5.2 in June 2026
  • Shared challenge: affected by U.S. export controls, primarily aimed at the Chinese market, limited internationally

I. Vendor Overview​

VendorFoundedFounderListed2025 RevenueMain Customers
Huawei Ascend2018 (division)Ren Zhengfeiprivate (wholly owned by Huawei)~¥20B (est.)Chinese gov, SOEs, military
Cambricon2016Chen Tianshi (CAS)2020-07 (STAR Market 688256)~¥5.2BByteDance, Alibaba, Baidu
Moore Threads2020Zhang Jianzhong (ex-NVIDIA China)2023-12 (STAR Market 688495)~¥1.5B (est.)gov, SOEs, gaming cos.

Strategic Positioning​

VendorTech routeCore strengthMain challenge
Huawei AscendAI-training-specific (Da Vinci)co-optimized HW/SW, carrier channelssanctions, process limits
CambriconAI-training-specific (MLUarch)high efficiency, competitive priceimmature ecosystem
Moore ThreadsFull-function GPU (MUSA)graphics + AI + general compute, Day-0 supportcompute below dedicated AI chips

II. Flagship Product Comparison​

1. Huawei Ascend 950DT (2026 flagship)​

ItemSpec
BF16 compute1,000 TFLOPS
Memory144GB HiZQ 2.0 (in-house HBM)
Memory bandwidth4 TB/s
TDP400W
ProcessN+2 (improved 7nm)
Released2026-04
Mass production2026-Q2
Unit price~¥80,000 (est.)

Strengths:

  • ✅ High large-model inference throughput: 144GB memory friendly to DeepSeek R1 (671B MoE)
  • ✅ Most mature ecosystem: CANN ~85% operator coverage, supports PyTorch, TensorFlow
  • ✅ Strong carrier channel: China Mobile, China Telecom large purchases

Weaknesses:

  • ❌ Process limited: N+2 below TSMC 4nm
  • ❌ Mediocre efficiency: 400W TDP, 2.5 TFLOPS/W

2. Cambricon MLU690 (2026 flagship)​

ItemSpec
BF16 compute600 TFLOPS
Memory64GB HBM3
Memory bandwidth2 TB/s
TDP280W
ProcessTSMC 7nm
Released2025-Q4
Mass production2026-Q1
Unit price~¥140,000 (est.)

Strengths:

  • ✅ Best efficiency: 280W TDP, 2.14 TFLOPS/W (1.5x H100)
  • ✅ Competitive price: ~$20,000, 33% cheaper than H100
  • ✅ Top-tier customer orders: ByteDance, Alibaba, Baidu

Weaknesses:

  • ❌ Small memory: 64GB limits large-model training scale
  • ❌ Immature ecosystem: NeuWare ~75–85% coverage; complex LLMs need manual tuning

3. Moore Threads MTT S5000 (2025 flagship)​

ItemSpec
FP16 compute~1,000 TFLOPS (est.)
Memory80GB GDDR6X
Memory bandwidth1.6 TB/s
TDP~350W
ProcessTSMC 4nm (est.)
Released2025-02
Mass production2025-Q2
Unit price~¥50,000 (est.)

Strengths:

  • ✅ Full-function GPU: graphics + AI + general compute, broader scenarios
  • ✅ Strong Day-0 support: June 2026 Day-0 support for Qwen3.5, GLM-5.2, MiniMax M3
  • ✅ Lowest price: ~¥50,000, high cost-performance

Weaknesses:

  • ❌ Compute below dedicated AI chips: FP16 ~50% of H100
  • ❌ Low memory bandwidth: 1.6 TB/s (48% of H100), limits large-model training

III. Compute Comparison (BF16/FP16)​

ChipBF16 computeMemoryBandwidthTDPEfficiency
Huawei Ascend 950DT1,000 TFLOPS144GB4 TB/s400W2.5 TFLOPS/W
Cambricon MLU690600 TFLOPS64GB2 TB/s280W2.14 TFLOPS/W
Moore Threads MTT S5000~1,000 TFLOPS80GB1.6 TB/s~350W~2.86 TFLOPS/W
NVIDIA H100989 TFLOPS80GB3.35 TB/s700W1.41 TFLOPS/W
NVIDIA H200989 TFLOPS141GB4.8 TB/s700W1.41 TFLOPS/W

Key insights:

  1. Ascend 950DT has the highest compute (1,000 TFLOPS) but mediocre efficiency
  2. Cambricon MLU690 has the best efficiency (2.14 TFLOPS/W), TDP only 280W
  3. Moore Threads MTT S5000 wins on full-function versatility but low bandwidth

IV. Software Ecosystem​

VendorStackFramework supportCoverageMaturity
Huawei AscendCANNPyTorch, TensorFlow, MindSpore~85%⭐⭐⭐⭐ (4/5)
CambriconNeuWarePyTorch-Cambricon, TensorFlow-Cambricon~75–85%⭐⭐⭐ (3/5)
Moore ThreadsMUSIFYPyTorch, TensorFlow, ONNX~70%⭐⭐⭐ (3/5)
NVIDIACUDAall~99%⭐⭐⭐⭐⭐ (5/5)

Ecosystem Maturity Assessment​

Huawei Ascend CANN:

  • ✅ Strength: highest operator coverage, supports MindSpore (in-house framework)
  • ❌ Weakness: steep learning curve, incomplete docs

Cambricon NeuWare:

  • ✅ Strength: PyTorch/TensorFlow compatible, low migration cost
  • ❌ Weakness: complex LLMs need manual tuning

Moore Threads MUSIFY:

  • ✅ Strength: strong Day-0 support, ONNX support
  • ❌ Weakness: lowest operator coverage, dual graphics+AI engine complexity

V. Commercial Progress​

Vendor2026 commercial progressMain customersShipments
Huawei AscendAscend 950 mass production; China Mobile large purchaseChina Mobile, China Telecom, gov~100K/yr (est.)
CambriconMLU690 mass production; ByteDance, Alibaba ordersByteDance, Alibaba, Baidu~50K/yr (est.)
Moore ThreadsMTT S5000 mass production; Day-0 Qwen3.5gov, SOEs, gaming cos.~30K/yr (est.)

Latest as of June 2026​

Huawei Ascend:

  • ✅ Ascend 950DT fully ramping
  • ✅ ¥1B procurement agreement with China Mobile

Cambricon:

  • ✅ MLU690 in volume shipment
  • ✅ ByteDance order ~20K units

Moore Threads:

  • ✅ Day-0 support for Qwen3.5, GLM-5.2, MiniMax M3
  • ✅ MTT S5000 2nd-gen released

VI. Selection Advice​

Scenario 1: Trillion-parameter training (GPT-4 class)​

Recommended: Huawei Ascend 950DT

  • ✅ 144GB large memory supports super-large models
  • ✅ Most mature ecosystem (~85% coverage)
  • ✅ Strong carrier channel, Chinese government backing

Alternative: Cambricon MLU690 (high efficiency, but small memory)

Scenario 2: Tens-to-hundreds-of-billions parameter training​

Recommended: Cambricon MLU690

  • ✅ Best efficiency (2.14 TFLOPS/W), low TCO
  • ✅ Competitive price (~$20,000)
  • ✅ Validated by top customers (ByteDance, Alibaba)

Alternative: Huawei Ascend 920 (more compute, mediocre efficiency)

Scenario 3: Cloud AI inference​

Recommended: Huawei Ascend 950PR (inference-specific)

  • ✅ Well-optimized inference throughput
  • ✅ 128GB memory friendly to MoE models
  • ✅ Mature stack, low deployment cost

Alternative: Moore Threads MTT S5000 (full-function GPU, inference + graphics)

Scenario 4: Edge AI / on-device inference​

Recommended: Moore Threads MTT S5000

  • ✅ Full-function GPU, graphics + AI
  • ✅ Lowest price (~¥50,000)
  • ✅ Strong Day-0 support

Alternative: Huawei Ascend 310 (low power, 8W TDP)

Scenario 5: Domestic substitution (gov, SOEs)​

Recommended: Huawei Ascend 950DT

  • ✅ Chinese government first choice, carrier bulk buys
  • ✅ Co-optimized HW/SW, stable performance
  • ✅ Supported by national semiconductor fund

Alternative: Cambricon MLU690 (high efficiency, competitive price)

VII. Future Roadmap​

Vendor2026 H220272028
Huawei Ascend950DT ramp960 (FP8 ~2 PFLOPS)970 (N+3 process)
CambriconMLU690 rampMLU790 (5nm, BF16 ~1,000 TFLOPS)MLU890 (3nm)
Moore ThreadsMTT S5000 2nd-genMTT S6000 (HBM3, FP16 ~1,500 TFLOPS)MTT S7000

VIII. Summary: Who Is the "China H100"?​

DimensionAscend 950DTMLU690MTT S5000
Compute⭐⭐⭐⭐⭐ (5/5)⭐⭐⭐ (3/5)⭐⭐⭐ (3/5)
Memory⭐⭐⭐⭐⭐ (5/5)⭐⭐ (2/5)⭐⭐⭐ (3/5)
Efficiency⭐⭐⭐ (3/5)⭐⭐⭐⭐⭐ (5/5)⭐⭐⭐⭐ (4/5)
Ecosystem⭐⭐⭐⭐ (4/5)⭐⭐⭐ (3/5)⭐⭐⭐ (3/5)
Price⭐⭐⭐ (3/5)⭐⭐⭐⭐ (4/5)⭐⭐⭐⭐⭐ (5/5)
Overall⭐⭐⭐⭐ (4/5)⭐⭐⭐ (3/5)⭐⭐⭐ (3/5)

Final conclusion:

  • Huawei Ascend 950DT is the domestic AI training chip closest to H100, strongest overall
  • Cambricon MLU690 is the most efficient domestic AI chip, lowest TCO
  • Moore Threads MTT S5000 is the cheapest full-function GPU, suited to edge AI and graphics+AI

References​

Disclaimer: Data based on public sources; actual specs per vendor official. MirrorFrog continuously updates domestic AI chip data — corrections welcome.

Changelog: 2026-06-23 initial release

Cambricon MLU690 vs NVIDIA H100: In-Depth Comparison — Can a Domestic AI Chip Replace the H100?

· 6 min read
Charles Qing
AI Hardware Analyst

In 2026, against the backdrop of U.S. export controls on AI chips to China, Cambricon's MLU690 has drawn intense attention as a "China-made H100." This article compares the two in depth across compute, memory, power, software ecosystem, measured performance, and price to help you make a selection decision.

Core Verdict (Read This First)​

DimensionMLU690H100WinnerGap
BF16 compute600 TFLOPS989 TFLOPSH100+65%
Memory capacity64GB HBM380GB HBM3H100+25%
Memory bandwidth2 TB/s3.35 TB/sH100+68%
TDP280W700WMLU690-60%
Energy efficiency2.14 TFLOPS/W1.41 TFLOPS/WMLU690+52%
Software ecosystemNeuWare (~75% coverage)CUDA (100% coverage)H100large gap
Price~¥140,000~¥200,000MLU690-30%
Availabilitydomestic spot stockexport-controlledMLU690✅

One-line summary: MLU690 delivers roughly 60% of H100's compute, but at only 40% of the power and 70% of the price — a strong fit for AI training and inference in the Chinese market.

1. Detailed Spec Comparison​

1.1 Compute​

PrecisionMLU690H100 SXM5H200 SXM5Note
FP8~300 TFLOPS (est.)3,958 TFLOPS3,958 TFLOPSH100 supports FP8; MLU690 likely does not
BF16/FP16600 TFLOPS989 TFLOPS989 TFLOPSH100 leads by 65%
FP32~150 TFLOPS (est.)60 TFLOPS60 TFLOPSMLU690 estimate; H100 actually higher
INT81,200 TOPS1,979 TOPS1,979 TOPSH100 leads by 65%

Key findings:

  • ✅ MLU690 reaches 60% of H100's BF16 compute
  • ⚠️ H100 supports FP8 (4-bit); MLU690 likely does not (needs confirmation)
  • ⚠️ H100's higher INT8 compute favors inference scenarios

1.2 Memory​

ItemMLU690H100H200Note
Capacity64GB HBM380GB HBM3141GB HBM3eH200 largest
Bandwidth2 TB/s3.35 TB/s4.8 TB/sH200 highest
TypeHBM3HBM3HBM3eH200 uses latest HBM3e

Key findings:

  • ⚠️ MLU690 has 20% less memory than H100 (64GB vs 80GB)
  • ⚠️ MLU690 bandwidth is 40% lower than H100 (2 TB/s vs 3.35 TB/s)
  • ❌ When running 70B+ parameter models, MLU690 may run out of memory (model parallelism required)

1.3 Power​

ItemMLU690H100H200
TDP280W700W700W
Efficiency (FP16/W)2.14 TFLOPS/W1.41 TFLOPS/W1.41 TFLOPS/W
8-card server power~3.5kW~6kW~6kW
Annual electricity (¥0.6/kWh)~¥18,400~¥36,800~¥36,800

Key findings:

  • ✅ MLU690 draws only 40% of H100's power, sharply cutting data-center electricity cost
  • ✅ MLU690 leads efficiency by 52%, better suited to large-scale deployment
  • ✅ For power-sensitive inference, MLU690 has a clear edge

2. Software Ecosystem​

2.1 Framework Support​

FrameworkMLU690 (NeuWare)H100 (CUDA)Note
PyTorch✅ (PyTorch-Cambricon)✅ nativeMLU690 needs an extra plugin
TensorFlow✅ (TensorFlow-Cambricon)✅ nativesame
JAX⚠️ partial✅ nativeMLU690 limited
ONNX⚠️ partial✅ nativesame
vLLM⚠️ in progress✅ nativeMLU690 awaits community port

2.2 Operator Coverage​

CategoryMLU690H100Note
Basic operators✅ 95%✅ 100%conv, matmul, etc.
Transformer operators✅ 85%✅ 100%Attention, LayerNorm, etc.
Custom operators⚠️ hand-written✅ CUDA C++MLU690 harder to develop
LLM inference opt.⚠️ basic✅ mature (FlashAttention, PagedAttention)H100 leads

Key findings:

  • ⚠️ NeuWare is only 5–6 years old, with ~75–85% operator coverage
  • ❌ Complex LLMs (e.g., GPT-4, Claude) may need manual optimization
  • ✅ Common models (Llama, Qwen, GLM) are essentially already supported

3. Measured Performance​

3.1 Training​

ModelMLU690 (time)H100 (time)Speedup
Llama 7B~48 h (est.)~30 h1.6x
Llama 70B~7 days (est.)~4.5 days1.6x
Qwen 72B~8 days (est.)~5 days1.6x

Note: above figures are estimates; real performance depends on software optimization.

3.2 Inference​

ModelMLU690 (tok/s)H100 (tok/s)Note
Llama 7B~80 tok/s (est.)~120 tok/sH100 +50%
Llama 70B~20 tok/s (est.)~35 tok/sH100 +75%
Qwen 72B~18 tok/s (est.)~30 tok/sH100 +67%

Key findings:

  • ⚠️ H100 leads inference by 50–75%
  • ✅ But MLU690 draws only 40% the power, with better efficiency
  • ✅ For cost-sensitive inference, MLU690 is more economical

4. Price​

4.1 Hardware Procurement​

ItemMLU690H100H200
Per-card (domestic)~¥140,000~¥200,000~¥300,000
8-card server (turnkey)~¥1,200,000~¥1,800,000~¥2,600,000
Cost gap-+50%+117%

4.2 TCO (3 years)​

ItemMLU690H100Note
Hardware¥1,200,000¥1,800,000MLU690 33% cheaper
Electricity (3y)¥55,200¥110,400MLU690 50% cheaper
Facility¥150,000¥250,000MLU690 40% cheaper
TCO (3y)¥1,405,200¥2,160,400MLU690 35% cheaper

Key findings:

  • ✅ MLU690's TCO is 35% lower than H100's
  • ✅ For large-scale deployment (100+ cards), the cost advantage is pronounced

5. Selection Advice​

5.1 Choose MLU690 if...​

  • ✅ Your business is primarily in the Chinese market
  • ✅ You are affected by U.S. export controls and cannot buy H100/H200
  • ✅ You are power-sensitive (edge data centers, high electricity-cost regions)
  • ✅ Your models use common architectures (Llama, Qwen, GLM)
  • ✅ You have domestic-substitution requirements (government, SOEs, military)

5.2 Choose H100/H200 if...​

  • ✅ Your business is global
  • ✅ You need to train frontier models (GPT-4 class)
  • ✅ Your models use complex operators (need the CUDA ecosystem)
  • ✅ You demand extreme performance (low-latency inference)
  • ✅ You can legally procure H100/H200
ScenarioRecommended
TrainingH100 (high perf) + MLU690 (low-cost scale-out)
InferenceMLU690 (cost-sensitive) + H100 (low-latency)
Domestic projectall MLU690
International marketall H100/H200

6. Outlook​

6.1 MLU690's weaknesses​

  • ⚠️ Immature software ecosystem: 75–85% operator coverage; complex models need manual tuning
  • ⚠️ Small memory: 64GB limits support for 70B+ parameter models
  • ⚠️ Weak interconnect: Cambricon Link bandwidth below NVLink
  • ⚠️ Limited international market: affected by U.S. export controls

6.2 MLU690's improvement path​

  • 📅 MLU790 (2027): expected 5nm process, ~2x compute
  • 📅 Memory upgrade: next gen may adopt HBM3e, capacity up to 128GB
  • 📅 Software: NeuWare ecosystem improving, operator coverage target 95%

7. Summary​

DimensionMLU690H100Recommended scenario
Compute⭐⭐⭐⭐⭐⭐⭐⭐⭐H100 for top-tier training
Memory⭐⭐⭐⭐⭐⭐⭐H100 for large models
Power⭐⭐⭐⭐⭐⭐⭐⭐MLU690 for inference
Ecosystem⭐⭐⭐⭐⭐⭐⭐⭐H100 for complex models
Price⭐⭐⭐⭐⭐⭐⭐⭐MLU690 for large-scale deployment
Domestic⭐⭐⭐⭐⭐❌MLU690 for Chinese market

Final recommendation:

  • 🇨🇳 Chinese market: prefer MLU690 (domestic + low cost)
  • 🌍 International market: prefer H100/H200 (performance + ecosystem)
  • 💡 Hybrid: train on H100, infer on MLU690

References​

Disclaimer: Data in this article is based on public sources and reasonable estimates; actual performance is subject to vendor official testing. MLU690's software ecosystem is evolving rapidly — watch NeuWare updates.

Last updated: 2026-06-23