AMD and NVIDIA’s rivalry in Tech industry is legendary. Both companies have a rabid fan base. From time to time both firms showcase products that outshine similar products of its competitor. This time AMD seems to have outclassed its arch-rival NVIDIA with its new lineup of Radeon Instinct GPUs with deep-learning & Artificial Intelligence support. AMD launched Radeon Instinct GPUs a few days ago along with commanding MI25 with Vega. This incidentally is AMD’s most powerful graphics architecture yet. Raja Koduri, the Technology Group’s chief over there at Radeon, unveiled one of its kind largely queer gadget which is definitely worth discussing and dissecting.
The contraception we are talking about is a compact 4×3 inch quad GPU cube which holds unbelievable TERAFLOPS of FP16 compute, the FP32 compute has fifty odd TFLOPS. To give a better idea about the power of this cube, let’s put it in this perspective that NVIDIA’s liquid cooled Drive PX2 AL supercomputer gives 16 TFLOPS of FP32 computer and just 8 TFLOPS of FP 15 computer which is found today on many driver-less and autonomous cars. AMD’s latest offering leaves NVIDIA’s counterpart in the dust with the former delivering almost six times more power than the latter. Another noticeable bit that makes AMD’s new Vega Cube futuristic compared to NVIDIA’s Driver PX2 is its size which you can hold in one hand.
AMD Vega Cube Blows Away The Competition With Its Powerful Specs Slotted Inside a Tiny Body
AMD Vega Cube was shown on the stage by Raja Koduri and houses four efficient individual Vega 10 graphics processors. Each one of the four processors has its own dedicated imperceptible circuit board that makes it suitable for use in small devices and gadgets, as well as allows manufacturers to reduce the size of their devices if they opt to go with AMD’s Vega Cube as the main processing unit. AMD has introduced a new style of assembly style in Vega Cube which is designed to bundle one over the other vertically in a special surface that helps to keep its size really small. All this has the potential to allow for the manufacturing of exceptionally competent supercomputers with smaller dimensions.
However, on the other side, such a small device delivering this level of power poses heat management issues. Which is compounded by the fact that each Radeon MI25 Instinct processor is measured at 300W of power, not to forget that MI25 is powered by Vega 10 chip which is heat intensive. NIVIDIA’s Tesla P100 accelerator for deep-learning is rated at 300W but with a lot less power capacity than the AMD’s accelerator.
One way to counteract this heating issue on a small device is to employ the liquid cooling technology, which Nvidia is already using in Drive PX2. AMD is not unfamiliar with liquid cooling since it has already used it in R9 Fury X rated at 275W down to almost 175W in the R9 Nano. Although, this temperature management comes at a cost of 15 percent decline in power. AMD is also working on Vega 10, which is a power-efficient variant with slower clock speeds.
We Are On The Cusp Of Artificial Intelligence That Mirrors The Complexity & Potential Of The Human Brain
Tech fanatics and those familiar with the news would be able to recall that back in 2001, Ray Kurzweil, one of the ardent advocates of the technological singularity hypothesis, prophesied that by 2019 commonplace $1000 computer will match the complexity and processing power of the human brain. Right on cue, AMD’s Vega Cube is about to hit the same benchmark predicted by Ray Kuezweil.
According to report, Vega Cube will be made available to gaming manufacturers and developers starting from next year. This leaves enough time for tech industry to produce a $1000 computing device with an ability to match the human brain in terms of power. However, this requires equivalent advancements in Artificial Intelligence required to catch up with the processing power to actually bring from outside the Artificial realm into the actual intelligence world. It’s not possible to predict the actual time-frame when this will eventually unfold, but its cannot be too far away.
AMD Vega 10 and Vega 11 GPUs Specs:
Graphics Card > > RX 480 | TBA | TBA
Processor > > 14mn | 14mn | 14mn
GPU > > Polaris 10 | Vega 11 | Vega 10
Memory Bus > > 256bit | TBA | 2048-bit
Memory > > 8GB GDDR5 | TBA | 8GB HBM2/16GB
TDP > > 150W | TBA | <300W
Bandwidth > > 256 GB/s | TBA | 512 GB/s
What are your thoughts on this news? Share in comments below.
Also Check out: