Innosilicon, a Chinese startup that provides graphics and memory solutions, introduced its first discrete GPU today at the “Fantasy One GPU Product Press Conference.”
Innosilicon presented its intentions for the Fantasy One GPU at the event. This CPU is built on Imagination graphics IP (BXT in particular), which adds a slew of new features to the Chinese company’s discrete GPU options.
Four products, including dual-GPU and single-GPU options, have been introduced. Type A is a consumer/workstation graphics card with a single Fantasy One GPU (multi-chip (chiplet) architecture). Unfortunately, none of the Chinese press sources on this announcement included the actual configuration of the Fantasy One.
According to Innoslicon’s statistics, this GPU has up to 5 TFLOPS of single-precision computation capacity and a fillrate of 160 GPixel/s. It has DisplayPort 1.4, HDMI 2.1, and VGA ports. This card will have up to 16 GB of GDDR6(X) memory and a 128-bit interface. G6X technology has previously been confined to NVIDIA/Micron Ampere GPUs, however, Innosilicon appears to have done extensive study in PAM4 signaling and was able to unlock up to 19 Gbps memory bandwidth for its GDDR6X implementation. However, a relatively narrow memory bus will have an impact on the highest theoretical bandwidth, which is set to reach 304 GB/s (so somewhere in between Radeon RX 6700XT and 6600XT).
The card on the right is an ‘entry-level’ model based on the Fantasy One GPU, with less RAM than a dual-fan solution, although those specifics were not mentioned in press stories.
Type B (seen below in a triple-fan or fanless configuration) is a dual-GPU system with two Fantasy One GPUs linked by an Innolink interface. The business promises up to 10 TFLOPS of computer power and fillrates of 320 GPixel/s. This card can support 32 simultaneous 1080p/60FPS streams or 64 720/30FPS streams. It will have up to 32GB of GDDR6(X) memory but will be limited by twin 128-bit GPU connections. All of these cards have a PCI-Express 4.0 interface that spans the whole X16 width.
Graphics APIs such as OpenGL, OpenGL ES, OpenCL, Vulkan, and DirectX are among the numerous IPs enabled on this GPU. DirectX support is the most intriguing announcement here. Regrettably, the organization did not specify whether the version of the Microsoft graphics API is supported.
At the event, the GPUs were live demoed, including benchmarking with the Unigine Heaven OpenGL demo. Unfortunately, no Chinese news outlets reported the results of these tests.
The business has already revealed that it is developing the Fantasy 2 and 3 GPU families, which will be presented next year. For these GPUs, Innosilicon intends to use 5nm manufacturing technology. The process technology of the Fantasy One GPU, on the other hand, has yet to be validated by any news outlet.
A closer look at the Fantasy One GPU, including schematics of the 9 GPU cores.
In October of last year, Imagination and Innosilicon announced a relationship. Imagination has announced that it will license the BXT design to the Chinese business, which will provide up to 6 TFLOPS of single-precision computation capability. These figures are somewhat higher than those of the Fantasy-One GPU.
