Chinese language social media, cloud, and leisure large Tencent on Monday revealed that it has began mass manufacturing of a house brew video transcoding accelerator.
The announcement comes practically two years after the corporate unveiled a trio of customized chips designed to speed up every little thing from streaming video to networking and synthetic intelligence workloads.
In a publish printed on WeChat, Tencent Cloud revealed that “tens of hundreds” of its Canghai chips, that are designed to dump video encode/decode for latency delicate workloads, have been deployed internally to speed up cloud gaming and stay broadcasting.
Tencent says the Canghai chip will be paired with GPUs from a wide range of distributors to assist low-latency sport streaming. When used for video transcoding, Tencent stated a single node outfitted with Canghai can ship as much as 1,024 video channels . We’ll observe that Nvidia, with the launch of its L4 GPUs final month, made comparable claims. With out real-world benchmarks, it is exhausting to say how both agency’s claims stack up.
With that stated, Tencent says its chip can obtain superior high quality on the identical bitrate as competing GPU-based accelerators, with latencies for a single 1080p body of simply 4 milliseconds. However once more, it isn’t acknowledged which GPUs it is evaluating its outcomes in opposition to.
No matter the actual world outcomes, the promised enchancment in compression effectivity ought to permit Tencent to keep up present streaming high quality whereas lowering the bandwidth consumed — though a few of this enchancment could possibly be right down to assist for higher codecs.
The Canghai chip reportedly helps H.264, HEVC, and AV1. As we have beforehand reported, AV1 is a comparatively new codec that along with being royalty free, can also be exceptionally environment friendly. Testing places AV1 at someplace between 20 and 40 p.c extra environment friendly in comparison with standard net streaming codecs, like HEVC.
Customized compute for the Chinese language cloud
On the subject of spinning customized chips to enhance the effectivity and economics of cloud computing, Amazon Internet Companies will get numerous credit score. The American e-tail large and cloud titan has developed every little thing from customized CPUs, AI coaching and inference accelerators, and smartNICs to dump many housekeeping workloads.
And whereas Google has developed an accelerator of its personal, referred to as the Tensor Processing Unit (TPU), most US cloud suppliers have largely caught with commercially out there elements from the likes of Intel, AMD, Ampere, Broadcom, or Nvidia, relatively than designing their very own.
Nevertheless, in China, customized chips look like extra prevalent, with growth an crucial accelerated by US sanctions that imply some tech merchandise cannot be exported to the Center Kingdom.
Along with its Canghai video chips, Tencent additionally created the Zixiao AI “reasoning” chip which it is deployed internally, plus the Xuanling community processor. The latter is designed to dump and speed up networking, storage, and compute workloads, which sounds an terrible lot like a smartNIC.
Tencent’s native peer Alibaba Cloud has additionally created customized silicon. In April 2022 the cloud supplier confirmed off its customized Arm CPUs and DPUs.
The cloudy enterprise unit of e-commerce large Alibaba started previewing its Yitian 710 processor, which reportedly options 128 Armv9 cores operating at 3.2GHz, eight channels of DDR5, and 96 lanes of PCIe 5.0, a few 12 months in the past in its Elastic Compute Service.
In the meantime, Alibaba’s Cloud Infrastructure Processing Unit (CIPU), very like AWS Nitro playing cards, is designed to dump virtualization capabilities related to storage, networking, and safety that might in any other case be dealt with by the primary CPU.
And whereas not absolutely customized, Baidu has tapped RISC-V chipmaker StarFive to assist it develop high-performance merchandise based mostly on the open ISA. Nevertheless, it stays to be seen when, or whether or not, these chips will come into play. ®