GPU from InnoSilicon will also be for players, powered by Imagination (PowerVR)

The Fenghua No.1 card, announced by InnoSilicon in mid-November and presented as a server accelerator, revealed with its parameters that it was definitely not created as a server accelerator. The presence of HDMI 2.1, eDisplayPort 1.4 and the support of APIs used in game graphics clearly showed that the ambitions were different. InnoSilicon confirmed this at a press conference, where she introduced a range of Fantasy One cards. It is based on the same GPU, but also targets players.

Although the graphics core is equipped with interfaces such as HDMI 2.1, (e) DisplayPort 1.4, GDDR6X, PCIe 4.0, etc. and therefore acts as a showcase for InnoSilicon, which is currently licensing the implementation of these interfaces, fortunately the 3D core itself is not hers, but it was licensed by itself from Imagination Technologies. It licenses graphics cores primarily used for ARM SoC (mobile phones, tablets), but has long boasted the ability to scale the architecture to the level usable in the PC segment.

For many years, these statements gave witnesses hope that the company was preparing to return to the gaming market, where at the turn of the century it offered Kyro I accelerators and especially Kyro II, which gave performance on the border of the then mainstream and high-end with minimal consumption. To the surprise of some owners, when they upgraded to more powerful hardware over the years, they pulled out a PC card with a dust trapped by a fan that had not been rotated for a long time, but it was not recognizable about the card’s performance and operation. It was the low energy requirements and the minimum of waste heat that motivated Imagination to move its activities to the mobile segment.

The described hopes, vain for many years, are now being realized. The question is whether in a way that would be really optimal in terms of brand return to the PC segment. Imagination states that the existing BXT architecture allows scaling up to 6 TFLOPS. This is almost fully used by the InnoSilicon with a chip reaching 5 TFLOPS. The difference is probably due to slightly lower measures, which are likely to guarantee a reasonable production yield. However, because 5 TFLOPS is currently no longer extreme (even the upcoming Radeon RX 6500 XT with core Navi 24 InnoSilicon has also prepared a version of the card with two GPUs, which reaches up to 10 TFLOPS. It is supposed to use its own interface called Innolink to connect the two GPUs, but more detailed information on how the GPU will work will not be published.

On the one hand, we can expect very energy-efficient graphics cores, on the other hand, they will be equipped with 16-32 GB of voracious GDDR6X memory. On the one hand, there is a product with 10 TFLOPS of performance, which with PowerVR’s tile-based architecture can mean significantly higher performance than AMD and (especially) Nvidia’s 10TFLOPS products, on the other hand, these 10 TFLOPS are achieved by a combination of two GPUs. this solution will not be burdened by some of the trade-offs common to multi-chip graphics cards (key questions are whether the performance scaling will be at least approximately linear, whether the GPU connection will be transparent to applications, or whether it will have a negative impact on frame rate smoothness).

Fantasy One – Type A (single-chip card)

A single GPU card achieves arithmetic performance of up to 5 TFLOPS and a fillrate of up to 160 GPixel / s. 16GB 19GHz GDDR6X is connected via 128bit interface (ie 304 GB / s). The outputs for display devices have a rugged DisplayPort 1.4, HDMI 2.1 and VGA (analog D-Sub).

Type A will be available in a two-fan version, but I’m getting a smaller single-slot low-profile card (whether with lower beats or a trimmed GPU, it’s not clear yet).

Fantasy One – Type B (two-chip card)

The version with two GPUs reaches 10 TFLOPS and 320 GPixel / sav presentation was presented data concerning its deployment as a server accelerator for streaming. For example, it can accelerate up to 32 streams in 1080p / 60 FPS or 64 streams in 720p / 30 FPS. It will carry 32 GB 19GHz GDDR6X (ie 16 GB on each GPU as in Type A).

The two-chip card will be available in a three-fan and passively cooled edition.


From the available data, it appears that a GPU with a tile-based architecture for accelerating game graphics would require significantly slower memory than GDDR6X, which was probably used so that the card did not suffer from sufficient data throughput in computational use. However, this can unnecessarily increase the price of the card from the player’s point of view and at the same time unnecessarily increase its consumption. Although it will be so low, it will still be significantly higher than it would have to be.

Consumption itself was mentioned only with very vague information, which everyone interprets differently. The single-chip two-fan card is equipped with an additional 6-pin connector, but there is information that the 4K resolution (probably game content) can handle the card at ~ 20W consumption and consumes around 50 watts in a cloud deployment. Why 6pin? The hypothetical answer may be the aforementioned computing deployment, during which consumption of over 75 watts can probably go.


Source: Diit.cz by diit.cz.

*The article has been translated based on the content of Diit.cz by diit.cz. If there is any problem regarding the content, copyright, please leave a report below the article. We will try to process as quickly as possible to protect the rights of the author. Thank you very much!

*We just want readers to access information more quickly and easily with other multilingual content, instead of information only available in a certain language.

*We always respect the copyright of the content of the author and always include the original link of the source article.If the author disagrees, just leave the report below the article, the article will be edited or deleted at the request of the author. Thanks very much! Best regards!