14:22Machine-translated from Azerbaijani by AIRead in Azerbaijani
Sugon’s 16.9 mm Workstation Rivals the Cloud for Local AI
Chinese supercomputer manufacturer Sugon has unveiled a new mobile workstation designed for artificial intelligence. The company describes it as the world’s first mobile workstation with a 64-thread processor. The device will feature a discrete graphics card with 16 GB of video memory. The mobile system promises performance close to that of cloud services when running large AI models locally.
Unlike large portable desktop systems, the new workstation has a chassis similar to that of a conventional laptop. It is 16.9 mm thick, which is close to the dimensions of thin premium laptops. Although Sugon manufactures supercomputers, it does not develop processors. The company’s planned merger with Hygon did not materialize last year, but Sugon remains its partner’s largest shareholder. Sugon has not disclosed which processor will be used, but the chip is expected to be sourced from Hygon. Hygon has offered chips with up to 32 cores and 64 threads for several years. The first four generations of the C86 family were based on AMD’s licensed Zen architecture. The new C86-5G uses the company’s own microarchitecture and supports AVX-512 and an SMT4 configuration with four threads per core. If this chip is selected, the workstation could use a 16-core, 64-thread configuration.
The 16 GB of video memory indicates that the device will have a discrete graphics card rather than an integrated solution. However, Sugon has not yet disclosed the GPU model. The manufacturer claims that the processor and graphics card working together will deliver AI inference speeds 3 to 5 times higher than those of the leading products on the market. With a local 35-billion-parameter Mixture of Experts model, the result reached up to 50 tokens per second. Throughput on standard cloud APIs typically ranges from 30 to 80 tokens per second. This comparison provides the basis for Sugon to present its mobile workstation as a near-cloud-level alternative for local AI workloads. However, the final hardware configuration and results from tests conducted outside the manufacturer are not yet known. Sugon has also not announced the device’s release date or price.
Unlike large portable desktop systems, the new workstation has a chassis similar to that of a conventional laptop. It is 16.9 mm thick, which is close to the dimensions of thin premium laptops. Although Sugon manufactures supercomputers, it does not develop processors. The company’s planned merger with Hygon did not materialize last year, but Sugon remains its partner’s largest shareholder. Sugon has not disclosed which processor will be used, but the chip is expected to be sourced from Hygon. Hygon has offered chips with up to 32 cores and 64 threads for several years. The first four generations of the C86 family were based on AMD’s licensed Zen architecture. The new C86-5G uses the company’s own microarchitecture and supports AVX-512 and an SMT4 configuration with four threads per core. If this chip is selected, the workstation could use a 16-core, 64-thread configuration. The 16 GB of video memory indicates that the device will have a discrete graphics card rather than an integrated solution. However, Sugon has not yet disclosed the GPU model. The manufacturer claims that the processor and graphics card working together will deliver AI inference speeds 3 to 5 times higher than those of the leading products on the market. With a local 35-billion-parameter Mixture of Experts model, the result reached up to 50 tokens per second. Throughput on standard cloud APIs typically ranges from 30 to 80 tokens per second. This comparison provides the basis for Sugon to present its mobile workstation as a near-cloud-level alternative for local AI workloads. However, the final hardware configuration and results from tests conducted outside the manufacturer are not yet known. Sugon has also not announced the device’s release date or price.