Posts Tagged ‘HBM’
The Interposer is what makes HBM possible. Basically, it allows the GPU and memory to communicate with each other. What makes this Interposer special is that it was specifically designed by AMD with high-volume bandwidth in mind to accommodate the stacked DRAM. Honestly, it’s the HBM that deserves most of the attention, but if you’re interested in how the Interposer is made, check out the slides below.
HBM (High Bandwidth Memory) is the new technology that is revolutionizing graphic cards as we know it. There’s no architectural feature that needs to be traded in order to get better performance. It has far great power efficiency, a much smaller form factor and can be placed much closer to the GPU than typical GDDR units. AMD began developing the technology years ago after they realized the limitations of GDDR5. HBM allows the GPU to continue to be more powerful without eating into the power envelope for the entire card. AMD will be the first to use it but we can expect to see more of HBM in the future. Here are the full list of slides for AMD’s revolutionary technology.
Maxwell did an amazing job of bringing the graphics card industry to a whole new level. Honestly, the architecture involved is quite amazing and we haven’t even seen the “Big Daddy” Maxwell card yet. (The 980 is by no means the top dog coming from Nvidia.) This put AMD under a lot of pressure to deliver something that could compare. The good news is that not only does AMD have a response, but if the leaked specs combined with HBM is true, the R9 390X could be as big of a game changer as Maxwell was a month ago. First, let’s take a quick peek at some of the more prevalent specs. The 390X is slated to sport 4096 GCN Cores. When compared to the 2816 Cores of the 290X, that leap seems pretty massive. It looks like we’re still showing 4GB of memory with a 1Ghz boost clock speed, but the next interesting fact about the Fiji architecture is the Manufacturing Process. These will be the first chips to feature the TSMC 20nm node which should help with efficiency and performance. The tiny taste of Tonga last month showed that AMD was heading down the path of better efficiency with better performance. However, outside of major architectural advantages that while extremely possible have not been seen yet, the improvements here may not seem out of this world. That is until we take a closer look at the memory being used. Some of you may have heard of HBM before but others may be scratching your heads. HBM stands for High Bandwidth Memory and sounds very much...