Briefly: China’s first homegrown GPUs are impressive on paper, but the firm that makes them doesn’t seem too serious about selling them to patrons. As an different, it’s pitching them as vitality-atmosphere pleasant compute accelerators for cloud programs, regardless that the spec sheet says they toughen DirectX and Vulkan.
In November, we realized that a Chinese firm known as Innosilicon had been establishing a GPU that would maybe well well be a an analogous success yarn as that of YMTC and its 128-layer 3D NAND memory. The fresh graphics card — dubbed Fenghua 1 — didn’t receive a flashy announcement love Nvidia and AMD are known for, but what caught our leer became that it became designed to enhance the DirectX and Vulkan APIs former in plenty of fashionable PC video games.
This week, TechNode’s Stewart Randall shared a instant video on Twitter of what looks to be proof that China’s first homegrown GPU isn’t factual a mirage. The video presentations a mini-tower PC equipped with Innosilicon’s Fenghua 1 graphics card working the GFXBench T-Rex benchmark. There’s no frames-per-2nd counter on the show masks, and we don’t glean to leer the take a look at consequence, but since right here is extra of a cell-oriented benchmark, it’d be unfair to plan comparisons to gift graphics playing cards from AMD and Nvidia.
A minimum of on paper, the Fenghua 1 must aloof match low-mid range graphics playing cards. There are two variants, a Style-A in a position to turning in up to five teraflops of FP32 efficiency and a Style-B that would maybe well double that in consequence of a twin-chip structure. They’re both in line with PowerVR IP from Creativeness Technologies, which must aloof impact them pretty vitality atmosphere pleasant no subject being manufactured the utilization of a 12 nm direction of node. They moreover toughen DirectX 11, DirectX 12, and Vulkan 1.2.
For reference, Nvidia’s GTX 1660 SUPER can raise a minute over five teraflops of FP32 efficiency, and the non-Ti RTX 2080 gives factual a hair over 10 TFLOPS. Alternatively, the specs equipped by Innosilicon don’t seem to add up. On the one hand, the Fenghua 1 playing cards use 19 Gbps GDDR6 and GDDR6X memory for up to 304 GB per 2nd of bandwidth, and the firm says the pixel maintain price of the Style-A configuration is a decent 160 GPixels per 2nd. Alternatively, the unparalleled vitality consumption ranking is a tiny 50 watts, and there’s no capacity to state how these specs translate into gaming cases.
Innosilicon says it would maybe well well promote reasonably a couple of variants with 4, 8, and 16 gigabytes of VRAM, and the advertising and marketing supplies point to those are primarily aimed toward the conducting market. Now not a lot is known about retail availability, but we would maybe well moreover be taught extra about these mysterious graphics playing cards next year.