Chinese chip independence is no longer a distant ambition — it’s an accelerating industrial project with concrete silicon milestones. Loongson Technology, a Chinese fabless chipmaker, used its 2025 annual earnings call and 2026 Q1 earnings call to outline progress on two key products: the 3B6600 octa-core CPU and the 9A1000 GPU. The targets are modest by global standards — matching Intel’s 12th Gen Alder Lake processors and AMD’s Radeon RX 550 — but the trajectory tells a more interesting story than the benchmarks do.
Key Takeaways
- Loongson’s 3B6600 CPU targets single-core performance comparable to Intel’s 12th Gen Alder Lake, launched in 2021.
- The 9A1000 GPU aims to match the AMD Radeon RX 550, an entry-level card from 2017 delivering roughly 1.2 TFLOPS FP32.
- Loongson CEO Zhang Ge admits the 3B6600 and 3B7000 lag in multi-core performance but have closed the single-core gap significantly.
- The 9A1000 entered tape-out in Q3 2024, with mass production likely in 2025 and broader market impact targeted for 2027.
- Loongson cannot access TSMC, Samsung, or Intel Foundry Services, and is expected to manufacture via SMIC.
What Loongson’s 3B6600 and 9A1000 Actually Promise
The 3B6600 is an octa-core processor built around single-core performance rather than raw thread count. Loongson CEO Zhang Ge described it during earnings calls as closing the gap significantly on single-core workloads while acknowledging that multi-core performance still trails the competition. The company claims a 20x performance uplift over its first-generation 3A1000, which ran at 800MHz on a 65nm process — a chip that launched over a decade ago. That comparison is generous framing; the real test is against current Western silicon, and there the gap remains real.
The 9A1000 GPU is the more strategically interesting product. Loongson founder Hu Weiwu stated that the 9A1000 delivers comparable performance to the Radeon RX 550 — an entry-level AMD card from 2017 with roughly 1.2 TFLOPS of FP32 compute. The 9A1000 also hints at scientific computing and AI acceleration capabilities, suggesting a design that borrows conceptually from tensor-style compute architectures, though Loongson has not published detailed specifications. Tape-out was targeted for Q3 2024, with mass production expected in 2025.
Is Chinese Chip Independence Closing the Gap on Intel and AMD?
The honest answer is: partially, and slower than Beijing would like. Loongson’s prior CPU, the 3A6000, reportedly achieved IPC figures close to AMD Zen 3 and Intel Raptor Lake — a genuinely impressive architectural result for a domestic Chinese design. The 3B6600 builds on that foundation with a mainstream octa-core configuration, but the multi-core deficit Zhang Ge acknowledged matters enormously in the workloads that drive server and cloud adoption. Single-core parity with a 2021 Intel chip is progress. It’s not a threat to Intel’s current product stack.
The GPU side is starker. Matching the RX 550 — a card AMD launched in 2017 — in 2025 or 2027 puts the 9A1000 several generations behind AMD’s current RDNA lineup and Nvidia’s current architecture. For basic display output, entry-level gaming, and light AI inference tasks paired with Loongson CPUs, it works. For anything demanding, it doesn’t come close. The goal here isn’t to beat Nvidia in data centers. It’s to give Chinese enterprises a domestically sourced option that doesn’t require import licenses or expose them to sanctions risk.
Why Sanctions Are Reshaping the Chinese Chip Market
U.S. export controls have cut Loongson off from leading-edge foundries. The company cannot access TSMC, Samsung, or Intel Foundry Services, leaving SMIC as the expected manufacturing partner for the 9A1000. SMIC’s most advanced nodes lag behind TSMC’s leading edge, which directly constrains what Loongson can achieve in transistor density and power efficiency. That manufacturing ceiling is the single biggest structural obstacle to Chinese chip independence — not design talent, not software ecosystems, but fab access.
The sanctions pressure is also producing unexpected side effects. A Russian outfit, reportedly only eleven months old at the time of reporting, has been marketing 16-core and 32-core processors under Cyrillic branding. The specs — a 16-core variant running at 2.10GHz with 64MB of L3 cache, octa-channel DDR4-3200, 1612.8 GFLOPS of compute, and a 180–200W TDP — match Loongson’s LS3C6000 exactly. The implication is straightforward: Chinese chips are already moving through third-party channels to reach sanctioned markets, with cosmetic rebranding doing the compliance work. That’s a preview of how Chinese chip independence plays out geopolitically, not just technically.
How does the Loongson 3B6600 compare to Intel 12th Gen?
Loongson claims the 3B6600 will match Intel’s 12th Gen Core (Alder Lake) in single-core performance, but this is an unverified company assertion with no independent benchmark data published yet. Alder Lake launched in 2021, meaning the 3B6600 is targeting performance that Intel’s current lineup has already surpassed by multiple generations. Multi-core performance still lags, per CEO Zhang Ge’s own admission.
When will the Loongson 9A1000 GPU be available?
The 9A1000 entered tape-out in Q3 2024, with mass production targeted for 2025 and broader commercial availability expected to ramp through 2027. No global availability has been announced — the product is aimed squarely at the Chinese domestic market to reduce dependence on foreign graphics hardware amid ongoing U.S. sanctions.
Can Loongson chips run standard software?
Loongson uses the LoongArch instruction set architecture, which is not x86-compatible. Software support requires porting or translation layers, which limits out-of-the-box compatibility with Windows applications and most Western enterprise software. The ecosystem is primarily targeted at Linux-based Chinese government and enterprise deployments where software stacks can be adapted to the domestic architecture.
Chinese chip independence is real progress dressed in modest clothing. Loongson’s 3B6600 and 9A1000 won’t threaten Intel or AMD in global markets — the performance targets are years behind the current frontier, and the manufacturing constraints imposed by sanctions aren’t going away. But that’s not the point. The point is to build a domestic supply chain that functions under pressure, and on that measure, Loongson is delivering. The 2027 timeline isn’t a victory lap. It’s the starting line for something more consequential.
Edited by the All Things Geek team.
Source: Tom's Hardware
