DeepSeek 联手华为开源 TileLang：给昇腾芯片补上「高级语言」这一层

2026 年 9 月 30 日，DeepSeek 在官方微信公众号宣布与华为合作，为华为昇腾（Ascend）芯片开发编程工具，并开源了昇腾平台上的编程基础设施。路透社的报道说得很具体：开源的清单里包括一门名为 TileLang 的高级编程语言，以及配套的计算库与通信库。DeepSeek 给出的理由是——要建新一代自主可控的 GPU 软件生态，第一优先级是拥有人人都能用、好写、还能榨干硬件性能的高级语言。这句话里藏着三个互相拉扯的约束，也是本文要讲清楚的地方。

一、这次到底发布了什么

先把事实钉死。路透社 9 月 30 日发自香港的报道给出了三条明确信息：DeepSeek 表示已与华为合作开发昇腾芯片的编程工具；在官方微信公众号的帖子里，DeepSeek 称正在开源昇腾平台的编程基础设施；开源内容包括一门名为 TileLang 的高级编程语言，以及相关的计算与通信库。报道同时点明了这件事的背景——中国科技公司正在加深彼此之间的合作，寻找英伟达生态之外的替代方案。没有发布会，没有产品页，一条公众号帖子就把最敏感的一层软件开源了出去。

# What DeepSeek says it released, restated as a mental model.
# This is a shape for reasoning, not a vendor API.

ascend_software_stack = {
    "high_level_language": "TileLang",          # the headline item
    "compute_libraries":   True,                # open-sourced
    "communication_libs":  True,                # chip-to-chip / collective
    "target_hardware":     "Huawei Ascend",
    "comparison_claimed":  "simpler programming model than NVIDIA CUDA",
}

# The claim worth testing is the language, not the libraries.
# Libraries shave weeks off one port. A language changes who is
# allowed to write kernels in the first place -- and that is the moat.

CUDA 的护城河从来不是硅，而是语言

二、为什么「高级语言」是最难的一层

DeepSeek 在帖子里把优先级讲得很直白：要建新一代自主可控的 GPU 软件生态，第一优先级是建立一门通用、易于编程、同时仍能触及硬件完整性能潜力的高级语言，并称 TileLang 正是为此而生。请注意这三个约束是互相拉扯的：「通用」要求它能跨越不同代的加速器；「易于编程」要求应用团队能直接上手，而不是只有内核专家能碰；「完整性能潜力」则意味着不能因为抽象而付出几倍的性能税。任何一层软件都可以替换，只有语言的替换成本最高——因为语言决定的是谁能写内核、生态里有多少现成的代码可以复用。

# DeepSeek's own framing, quoted by Reuters on 2026-09-30.
# Read it as a requirements document, not marketing:

DEEPSEEK_REASONING = """
To build a new generation of independent, self-controlled GPU
software ecosystem, the first priority is establishing a high-level
language that is universal, easy to program, and still capable of
reaching the hardware's full performance potential.
"""

# Three constraints, and the third one is the hard one.
CONSTRAINTS = {
    "universal":        "one language, many accelerator generations",
    "easy_to_program":  "reachable by app teams, not just kernel specialists",
    "full_performance": "no 2-5x tax versus hand-tuned kernels",
}

三、超级节点：把 128 颗芯片当成一台机器

路透社的报道还提到一件容易被忽略的事：两家公司联合推进了一套基于 128 颗昇腾 950 芯片的超级节点配置，并对集群的算力与芯片间通信做了联合优化。这条信息的重要性在于它指向了真正难的部分。单卡峰值算力是纸面数字，能买到的是可持续算力；在大规模并行下，性能损耗几乎全部来自通信与调度。也就是说，这个超级节点本质是一个软件交付物——它检验的正是「高级语言 + 通信库」这套组合能不能把规模化的损耗压下来。

# A portability check you can actually run. The question is never
# "does the hardware work?" -- it is "how much source has to change?"

PORT_SURFACE = [
    "kernel language",        # TileLang vs CUDA C++ / Triton
    "collective comms",       # chip-to-chip libraries
    "graph capture",          # does the runtime expose CUDA graphs equivalent?
    "quantisation kernels",   # FP8 / INT4 paths
    "profiler + debugger",    # what do you read when it is slow?
    "CI on real silicon",     # can you test without renting a cluster?
]

def portability_score(available):
    missing = [k for k in PORT_SURFACE if k not in available]
    return {"coverage": 1 - len(missing) / len(PORT_SURFACE), "missing": missing}

高级语言决定谁来写内核

四、对开发团队意味着什么

如果你只在一种加速器上跑模型，这次发布短期内不会改变你的预算。真正受影响的是三类人：一是在做硬件多元化、需要给供应链留后路的基础设施团队；二是被 CUDA 特有的内核写法绑死、难以迁移的推理团队；三是评估自研或国产推理集群、需要判断三年后软件成熟度的技术决策者。对第二类人来说，判断标准很朴素：把「可移植性面积」列成清单，逐项对——内核语言、集合通信、图捕获、量化内核、性能分析工具、真机 CI。缺哪一项，就意味着哪一项要你自己补。

# The supernode work Reuters reported alongside the release: a system
# built on 128 Ascend 950 chips, with compute throughput and
# chip-to-chip communication tuned together. Scale-out is a software
# problem, so treat it as a software deliverable, not a spec sheet.

SUPERNODE = {"chips": 128, "model": "Ascend 950"}

def effective_throughput(chips, per_chip, comm_efficiency):
    """Peak flops are easy. Sustained flops are what you buy."""
    ideal = chips * per_chip
    return {
        "ideal": ideal,
        "sustained": ideal * comm_efficiency,
        "loss_pct": round((1 - comm_efficiency) * 100, 1),
    }

# Ask every accelerator vendor for `comm_efficiency` at your tensor
# parallel width, not at their demo width.

五、像工程师一样验证，而不是像分析师一样转述

「比 CUDA 更简单」是一个可以设计实验去证伪的命题。不要拿厂商的演示宽度去问通信效率，要拿你自己的张量并行宽度去问；不要只比峰值吞吐，要同时对吞吐比、P99 延迟比、以及最关键的「是否需要重新训练权重」。一个诚实的评估框架应该包含三个数字：相同模型与精度下的每秒 token 比值、p99 延迟比值、以及完成一次移植所改动的人日。任何没有这三个数字的「已适配」声明，都还停留在演示阶段。

# Vendor claims deserve vendor-grade verification. Put the numbers
# behind a gate so "it works on paper" never becomes a roadmap item.

def evaluate_accelerator_claim(native_kernels, ascend_kernels, prompt_suite):
    """Compare like for like: same model, same batch, same dtype."""
    results = {}
    for name, kernels in (("native", native_kernels), ("ascend", ascend_kernels)):
        results[name] = benchmark(kernels, suite=prompt_suite)
    native, ascend = results["native"], results["ascend"]
    return {
        "tok_per_s_ratio": ascend.tokens_per_second / native.tokens_per_second,
        "p99_latency_ratio": ascend.p99_ms / native.p99_ms,
        "retrain_required": not ascend.weights_compatible,
        "verdict": "port" if ascend.tokens_per_second / native.tokens_per_second > 0.9
                   else "re-benchmark next generation",
    }

把 128 颗芯片当成一台机器来调优

六、对 CUDA 护城河的实际意义

英伟达的护城河从来不只是硅，而是那一层让全世界的内核、编译器、调优经验都长在上面的软件惯性。DeepSeek 这次的动作，打的正是这层惯性里最上游的位置：语言。它不会在明天改变谁买什么卡，但它改变了「替代方案的成本曲线」——只要有一个足够好用、且被真实工作负载验证过的高级语言，上层的库、框架、调优经验就有机会跟着一起迁移。对任何一家不想把三年路线图押在单一供应商上的团队来说，这值得放进观察名单，而不是放进新闻摘要。