On AMDs upcoming graphics cards
A recent presentation from AMD and SK Hynix regarding the 3D stacked High Bandwidth Memory (HBM) co-developed by these two companies have been leaked online. While showing some performance details regarding the new HBM memory standard, the presentation also suggest that HBM will show up with AMD's upcoming 20nm Pirate Islands GPUs.
Caught by Wccftech.com, the presentation shows that the 3D stacked High Bandwidth Memory (HBM), which was originally developed by SK Hynix and AMD in order to replace GDDR5 memory, will bring some impressive performance gains when compared to GDDR5 memory.
The leaked presentation also shows that HBM will allegedly show up on AMD's upcoming Pirate Islands GPUs, including the R9 380X based on the Fiji GPU and the R9 390X based on Bermuda GPU. While the DDR3 was quite a star, the GDDR5 brought some rather impressive performance gains taking the maximum bandwidth up from 2GB/s to 28GB/s. The 3D stacked HBM memory standard will push that performance up to a rather impressive 128 to 256GB/s. This translates to about 4.5x the bandwidth of GDDR5 and 16 times the bandwidth of DDR3.
The leaked presentation also revealed some details regarding the future of the HBM memory and while first-generation of HBM will bring 2Gb per DRAM die, 1Gbps speed/pin, 128GB/s bandwidth and 4-HI stack (1GB) the second generation will push that to 8Gb per DRAM die, 2Gbps speed/pin, 256GBps bandwidth/stack and 4/8 Hi stack (4GB/8GB) configuration. AMD and SK Hynix will push this technology and scale it up until mid 2020s and bring even further improvements in performance, capacity and power efficiency, while the availability of first HBM memory samples is expected as of Q4 2014.
While GPUs are certainly its priority, AMD also plans to use stacked HBM memory on APUs as well. While AMD might use 3D stacked HBM memory with its upcoming 20nm Pirate Islands GPUs in early 2015, at least according to previous rumors, Nvidia's plans are to use same technology with its Pascal GPU, which is scheduled for 2016. This will put AMD ahead of Nvidia for about a year, at least when it comes to memory technology.
You can check out more details via link below.
Source: Wccftech.com.
Caught by Wccftech.com, the presentation shows that the 3D stacked High Bandwidth Memory (HBM), which was originally developed by SK Hynix and AMD in order to replace GDDR5 memory, will bring some impressive performance gains when compared to GDDR5 memory.
The leaked presentation also shows that HBM will allegedly show up on AMD's upcoming Pirate Islands GPUs, including the R9 380X based on the Fiji GPU and the R9 390X based on Bermuda GPU. While the DDR3 was quite a star, the GDDR5 brought some rather impressive performance gains taking the maximum bandwidth up from 2GB/s to 28GB/s. The 3D stacked HBM memory standard will push that performance up to a rather impressive 128 to 256GB/s. This translates to about 4.5x the bandwidth of GDDR5 and 16 times the bandwidth of DDR3.
The leaked presentation also revealed some details regarding the future of the HBM memory and while first-generation of HBM will bring 2Gb per DRAM die, 1Gbps speed/pin, 128GB/s bandwidth and 4-HI stack (1GB) the second generation will push that to 8Gb per DRAM die, 2Gbps speed/pin, 256GBps bandwidth/stack and 4/8 Hi stack (4GB/8GB) configuration. AMD and SK Hynix will push this technology and scale it up until mid 2020s and bring even further improvements in performance, capacity and power efficiency, while the availability of first HBM memory samples is expected as of Q4 2014.
While GPUs are certainly its priority, AMD also plans to use stacked HBM memory on APUs as well. While AMD might use 3D stacked HBM memory with its upcoming 20nm Pirate Islands GPUs in early 2015, at least according to previous rumors, Nvidia's plans are to use same technology with its Pascal GPU, which is scheduled for 2016. This will put AMD ahead of Nvidia for about a year, at least when it comes to memory technology.
You can check out more details via link below.
Source: Wccftech.com.
News by Luca Rocchi and Marc Büchel - German Translation by Paul Görnhardt - Italian Translation by Francesco Daghini
Previous article - Next article
comments powered by Disqus
comments powered by Disqus