Xiaomi: New CPU matches Apple cores single threaded, much faster multithreaded
Xiaomi: New CPU matches Apple cores single threaded, much faster multithreaded
小米：新款 CPU 单核性能比肩苹果，多核性能大幅领先
Daniel Lemire@lemire: Xiaomi is the Chinese tech giant. Their phones compete with iPhones. Their new CPU roughly matches Apple cores on single threaded tasks, and is much faster in multithreaded execution. Of course, Apple may soon announce their next processor, so this edge may not last long. And you may find it difficult to find a phone with the next CPU (Xring O3). But the new Xiaomi processor is worth discussing further as it reveals an important trend.
Daniel Lemire@lemire：小米是一家中国科技巨头，其手机产品与 iPhone 展开竞争。他们的新款 CPU 在单线程任务中大致能与苹果的核心相媲美，而在多线程执行方面则快得多。当然，苹果可能很快就会发布其下一代处理器，因此这种优势可能不会持续太久。此外，你可能很难买到搭载这款新 CPU（Xring O3）的手机。但小米这款新处理器值得深入探讨，因为它揭示了一个重要的趋势。
The chip has a lot of cache (44 MB in total). It is more than most laptop CPUs. If you have an Intel processor in your laptop, chances are good that you have less cache. The biggest cores on the Xring O3 are the C1-Ultra. C1-Ultra are really powerful cores. They support SME2 (Scalable Matrix Extension 2) for matrix/AI acceleration, SVE2 for data parallelism (SIMD). It is astonishingly wide, with 21 execution ports, six of which support SIMD operations (128 bits). This is more execution ports than you have on your Intel/AMD processor.
这款芯片拥有巨大的缓存（总计 44 MB），超过了大多数笔记本电脑的 CPU。如果你的笔记本电脑使用的是英特尔处理器，那么它的缓存很可能比这更小。Xring O3 上最强大的核心是 C1-Ultra。C1-Ultra 是性能非常强劲的核心，它们支持用于矩阵/AI 加速的 SME2（可扩展矩阵扩展 2）以及用于数据并行（SIMD）的 SVE2。它的架构极其宽泛，拥有 21 个执行端口，其中 6 个支持 SIMD 操作（128 位）。这比你目前在英特尔/AMD 处理器上看到的执行端口还要多。
The AMD Zen 5 has the upper hand because it can do 4x512-bit but 6x128-bit is the best you can do on an ARM chip as far as I know. So the trend is clear. We are getting cores that are massively parallel in terms of the number of execution units. We get better SIMD (more units) and many more units capable of doing arithmetic. This means that you can do many, many independent additions or multiplications per cycle. And much more cache. This is where all the transistors go.
AMD Zen 5 占据优势，因为它能实现 4x512 位运算，而据我所知，ARM 芯片目前能做到的极限是 6x128 位。因此，趋势很明显：我们正在获得在执行单元数量上实现大规模并行的核心。我们拥有了更好的 SIMD（更多的单元）以及更多能够进行算术运算的单元。这意味着你可以在每个周期内执行大量独立的加法或乘法运算。再加上更多的缓存，这就是所有晶体管的去向。
Ice Universe@UniverseIce: Xiaomi’s Xring O3 delivers a massive performance leap, scoring 3,945 in Geekbench single-core and an unprecedented 15,221 in multi-core. Ice Universe@UniverseIce： 小米 Xring O3 带来了巨大的性能飞跃，Geekbench 单核跑分达到 3,945 分，多核跑分达到史无前例的 15,221 分。
Jon Clegg@jonclegg77: This smells somewhat legitimate and somewhat BS. The architecture itself looks impressive — TSMC 3nm, a 4+ GHz prime core, tons of cache, and a very wide execution engine. And the claimed Geekbench numbers, around 3,945 single-core and 15,221 multi-core, are especially impressive. Jon Clegg@jonclegg77： 这看起来既有些真实，又有些吹牛的成分。架构本身确实令人印象深刻——台积电 3nm 工艺、4GHz 以上的主频核心、海量缓存以及非常宽泛的执行引擎。而所宣称的 Geekbench 分数（单核约 3,945 分，多核 15,221 分）尤其令人惊叹。
Chris Allen@theodorvaryag: SVE2 with 21 execution ports? good golly miss molly I could make that thing scream. Chris Allen@theodorvaryag： 带有 21 个执行端口的 SVE2？天哪，我能让那玩意儿性能炸裂。
noname@l1bertypr1me: Tensor G6 also has C1-Ultra core, but can only get ~2700 single core. Something feels off here. noname@l1bertypr1me： Tensor G6 也采用了 C1-Ultra 核心，但单核跑分只有 2700 左右。这里面感觉有点不对劲。