Finally, it’s here and 100% legit, no more bits of gossip or hypothesis. Today we are going to present to you AMD’s Official presentation slides of HBM. A major yells out to Keith May, who’s an individual from AMD’s Red Team Plus, for imparting these slides. Before we plunge profound into the new slides how about we just rapidly experience the historical backdrop of HBM and its connection to AMD’s High end Graphics Core.
AMD Confirm HBM Memory For Next Gen GPUs, Official Presentation
A year ago SK Hynix, the memory maker that co-created HBM with AMD had distributed its first far reaching the open documentation of the progressive memory innovation. We distributed an article a year ago in which we examined the new memory innovation and its suggestions for the business inside and out.
The presentation discharged by SK Hynix a year ago point by point the memory innovation itself and the critical points of interest it offers over DDR3 and GDDR5 memory. Today, anyway, we get the chance to hear the story from an AMD viewpoint. Furthermore, how the organization plans to wed the new memory innovation to their products and all the advantages that will go with that marriage. So immediately how about we get straight into it.
How To Overcome with Bandwidth Challenge
Generally performance and power efficiency have principally been pushed by the advancement of more propelled assembling procedures in the semiconductor business. Which empowered the making of smaller, denser and more powerful, productive transistors as time passed by, a perception that later got to be what is presently known as Moore’s Law.
AMD and Sky Hynix put in seven years and poured an exceptional sum assets to add to a completely new memory standard for good reasons. High Bandwidth Memory was conceived and exists today out of need. Traditional memory guidelines basically came to a point where they had turned out to be architecturally. This point is spoken to in the diagram above at the convergence between the energy consumption of memory and the processor it is encouraging.
GPUs, kept on scaling quickly. Since they profited the most from the expanding transistor count and productivity. Shockingly memory standards did not and then all the more intense these accelerators (GPUs, SSD, etc.) turned into the more memory transfer speed they needed.
The circumstance wound up being that specialists would need to exchange GPU power for more memory data transmission to keep the GPU sustained. AMD shared a comparable chart almost three years prior when the head of the die stacking program Bryan Black first discussed stacked memory and its required in the business. Also, that is the reason the business expected to take a gander at other memory options that would tackle the execution and power efficiency challenges.
Memory standards such as GDDR5 hit a wall on several fronts. GDDR5 failed to continue to scale effectively in performance and power efficiency. Pushing the frequency of GDDR5 to attain more bandwidth meant sacrificing power efficiency. And designing processors with wider GDDR5 memory interfaces inflated both costs and power consumption.
Advantages Of High Bandwidth Memory (HBM)
GDDR5 likewise has a thickness restriction that would keep its integration into high performance small GPUs. The High Bandwidth Memory, then again is vertically stacked, this thus implied that the associations starting with one DRAM die to the other were much shorter and along these lines more efficient.
Likewise, in light of the fact that vertical stacking empowers much more prominent densities there are huge range funds on the printed circuit board subsequently. Empowering the significantly more reduced compact form factors.
From the above slide we can see that HBM is far more compact than GDDR5 and also keeping in mind that HBM is 9X faster than GDDR5.
The final result is a 300% boost in transfer speed every watt. Accomplished through essentially higher bandwidth memory. All the more interestingly, AMD gauges that 15-20% of the R9 290Xs 250W TDP is in the memory sub-system. Which implies that AMD can lessen the memory power consumption from 40-50W down to 15W fair by changing from GDDR5 to HBM. A 290X that uses HBM therefore would be a 215W GPU, instead of a 250W one. We hope to see this emerge with AMD’s upcoming High-End Radeon GPUs.
AMD is situated to report its 2015 GPUs lineup at Computex on June 3rd.