We have been hearing multiple reports about new GPU comers for the past few years but none of them has actually made it to the mainstream market where we could see alternative GPUs to AMD and Nvidia.
Except for Intel, almost every GPU maker has either failed or could not improvise to the point where their GPUs would become truly usable. One of the biggest examples of this is China-exclusive Moore Threads MTT S80 GPU which even though features high-end specs such as 4000 cores and 16GB VRAM, isn’t able to perform consistently well in different titles.
Currently, there are as many as 13 GPU manufacturers in China but none could actually come closer to Nvidia whether we talk about Gaming or even data center GPUs.
But the new GPU maker is different. As reported by Jonpeddie, Unlike, AMD and Intel, Denglin GPU+ technology provides a software-defined on-chip heterogeneous computing architecture that will be compatible with both CUDA and OpenCL programming languages.
While AMD and Intel make use of OpenCL, CUDA has been Nvidia’s proprietary compute model only available to Nvidia GPUs. Compared to OpenCL, CUDA can dramatically speed up computing applications and has been used to develop popular programs like those of Adobe and Autodesk. The company not only received funding from China Net Investment but has also prepared 4 GPUs for AI-based compute acceleration.
The generation is called Goldwasser where the entry-level chip features two power-efficient editions featuring 4-16GB of VRAM and a TDP of 10 and 15 Watts respectively. These are compatible with the PCI-E 4 slots and also include an MXM version for laptops.
For full PCI-E 16 slots, Denglin has Goldwasser L and XL where the floating point and integer operation performance are significantly higher and the GPUs feature 32 to 64GB of memory. These GPUs will operate in a power range of 25 to 120W and the company promises to deliver up to 5 times performance gain in transformer-based models along with reducing the hardware cost in ChatGPT and Generative AI workloads.
Denglin GPUs here will be domestic products that probably won’t see restrictions like other GPU makers. However, it is still not clear from where the company is outsourcing its chips because the last time another Chinese GPU manufacturer called Biren Tech tried to secure 7nm wafers, it was restricted due to the US suspending TSMC from supplying the chips to China.
Currently, the company hasn’t revealed its gaming GPU lineup but as reports say that these 4 GPUs are made for both gaming and AI training, it is possible that the Goldwasser GPUs are also capable of gaming or it may mean that the company will release gaming dedicated GPUs in the same lineup.
Whatever the case is, we can finally see PC gaming growing faster than before considering that the Global GPU industry is projected to grow from $33 billion dollars to $477 billion dollars by 2030. This means, there will be a growing demand from not only high-end software office users but also from gamers.