loading
loading
Names like Sugon or Hygon may not appear on lists of the best CPUs since they often fly under the radar. However, that does not mean they are not making substantial strides in China. According to the latest ITHome report, Sugon has recently teased an upcoming mobile workstation (MWS) designed to push the boundaries of AI performance. The next-generation device boasts a powerful domestic 64-thread processor paired with 16GB of VRAM, a combination that delivers up to 50 tokens per second on local 35B MoE AI models. Sugon’s device is closer to a traditional laptop form factor rather than those pseudo-portable workstations we have seen over the years that look more like a desktop stuffed inside a suitcase. The mobile workstation is 0.665 inches (16.9 mm) thick, so it is right in line with the thin and lightweight premium laptop category, only slightly thicker than ultrabooks. Sugon, a prominent Chinese supercomputer manufacturer, does not actually produce processors. Last year, Sugon attempted to merge with Hygon, which originated from a joint venture between AMD and Chinese partners, in hopes of creating a domestic server superpower to rival global giants such as HPE and Dell, but the deal did not go through. Nonetheless, the failed merger did not alter the already close relationship between Sugon and Hygon, with the former already being the latter's largest shareholder. Sugon did not officially disclose which processor will power the mobile workstation. However, the manufacturer will likely source it from its long-time partner, Hygon. Sadly, knowing the chip has 64 threads does not give us a complete picture of which processor model is inside Sugon's device. For several years, Hygon has offered processors with designs up to 32 cores and 64 threads. Four generations of Hygon chips, spanning from the C86-1G to the C86-4G series, leveraged licensed Zen IP from AMD. Alternatively, Sugon’s mobile workstation may tap Hygon's latest C86-5G chips. These processors are particularly interesting because Hygon claims they utilize a “new self-developed microarchitecture,” signaling the departure from AMD's Zen microarchitecture. The new C86-5G series, which supports AVX-512, features a four-way simultaneous multithreading (SMT4) configuration, or in other words, four threads per core. If Sugon’s mobile workstation is indeed equipped with one of these chips, it would likely be a 16-core, 64-thread configuration that could provide decent processing power for AI. The small teaser specifically highlights the inclusion of 16GB of VRAM. The small detail clearly indicates the mobile workstation will come equipped with a discrete graphics card instead of integrated graphics. According to the vendor, the combination of the 64-thread processor and the mysterious 16GB discrete graphics card will deliver AI inference performance that is between 3X and 5X faster than mainstream products currently available on the market. The performance figures being thrown around are up to 50 tokens per second when running a massive 35-billion-parameter Mixture of Experts (MoE) model locally. By comparison, standard cloud APIs typically offer throughput ranging from 30 to 80 tokens per second. If Sugon can deliver on its promise, its mobile workstation could be a true game-changer for AI workloads. Get Tom's Hardware's best news and in-depth reviews, straight to your inbox. Sugon has not revealed the availability or pricing for the forthcoming mobile workstation. Given the current pricing for memory and storage products, we expect the mobile workstation to cost a small fortune. Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds. Zhiye Liu is a news editor, memory reviewer, and SSD tester at Tom’s Hardware. Although he loves everything that’s hardware, he has a soft spot for CPUs, GPUs, and RAM. - jp7189 It's not meaningful to compare a 16GB laptop to a frontier cloud API. You'd have to use something like 2 bit quantization to get a 35B to fit in 16GB and hit 50 tok/s which would be utter crap compared to a 1T+ model running at higher precision.Reply - j1mm4 Reply Not to mention you still wouldn't hit 50 tk/s of generation. It's weird how they kept throwing that number around but didn't specify if it was prefill or generation. If it's prefill (which it probably is) that's still pretty miserable.jp7189 said:It's not meaningful to compare a 16GB laptop to a frontier cloud API. You'd have to use something like 2 bit quantization to get a 35B to fit in 16GB and hit 50 tok/s which would be utter crap compared to a 1T+ model running at higher precision.
document fetched 2026-09-04 02:52 · canonical https://tomshardware.com/laptops/chinese-supercomputer-manufacturer-reveals-the-worlds-first-64-thread-mobile-workstation-for-ai-domestic-cpu-paired-with-mystery-gpu-with-16gb-of-vram-promises-cloud-level-performance