NVIDIA has established itself as the premier AI processing hardware, and its latest chip, the Blackwell Ultra GPU, is being hailed by the company as its fastest chip for AI. It boasts an impressive 640 5th-gen Tensor cores and 15 PetaFLOPS of NVP4 compute power. NVIDIA is positioning this new GPU as a big upgrade over its current Blackwell AI chips. Based on current information from the company, it promises some pretty significant gains.

The Blackwell Ultra features 208 billion transistors, which is 2.6 times more than what NVIDIA offered on the Hopper GPUs that came before it. One of the big benefits, NVIDIA says, is that this build structure allows a huge performance boost, but it also keeps the CUDA programming model familiar. NVIDIA says it’s the same programming model that developers have been using for almost 20 years.

The NVIDIA Blackwell Ultra GPU is 50% faster than the non-Ultra GB200 Blackwell chip

NVIDIA’s new Blackwell Ultra GPU brings in more than a few upgrades over the non-Ultra models, and one of those upgrades is its overall performance. It’s reportedly 50% faster than the Blackwell GB200, and that’s a pretty noticeable jump. It also has 288GB of memory compared to the 192GB in the non-Ultra Blackwell chip.

The GPU also features 160 streaming multicore processors, which NVIDIA breaks up into 8 different graphics processing clusters. In terms of performance, the new Blackwell Ultra can deliver “higher throughput and lower latency” for both dense and sparse AI workloads. Whether you’re really piling on tons of tasks or doing something more mild, the Blackwell Ultra promises to push through more of those tasks to completion in less time.

In other words, NVIDIA says this chip is what will allow businesses to effortlessly transition from proof-of-concept AI to production AI factories. According to the company, AI computations that were impossible before are now possible thanks to the Blackwell Ultra chip. While these are generally going to be more powerful than NVIDIA’s gaming GPUs, they’re not designed or built for gaming. Not to mention the cost is likely to be much more substantial.
The post NVIDIA’s AI processing just got a shot of adrenaline with this new chip appeared first on Android Headlines.

Source: ndroidheadlines.com