AMD Confirms Next-Gen Vega GPU and Navi GPU – 14nm Vega With HBM2 in 2017 and Navi GPU With NextGen Memory in 2018
AMD has just announced their next generation Vega and Navi GPUs at Capsaicin 2016 event which will arrive in 2017 and 2018, respectively. The next generation GPUs will be offered in the flagship space and will be scalable down to several SKUs. The Vega GPU has also been in several headlines before as we were the first to report the GPU existed and today the GPU is finally confirmed along with its successor, the Navi GPU.
AMD's Next Gen Vega With HBM2 Launches in 2017, Navi GPU With Next Gen Memory and Extremely Scalable Architecture in 2018
So we know that AMD don't just have Polaris planned for 2016 and beyond. In fact, there are two Polaris GPUs, the Polaris 10 and Polaris 11 which are the main attraction of today's event but the major announcement was revealed as a part of a roadmap which talked about AMD's next gen GPUs. In 2016, AMD will bring Polaris GPU which is based on the FinFET process and delivers up to 2.5x performance efficiency. Polaris will adopt a brand new GCN architecture that's going to be called GCN 4.0 as already announced by AMD and will be AMD's bet at the delivering an efficient GPU architecture.
AMD’s Chief Technology Officer, Mark Papermaster, confirmed last year that AMD will be leveraging 14nm FinFET technology across its CPU, APU and GPU products. These will include both Polaris and Vega GPUs while Navi is not yet confirmed whether it will use a new process or the current iteration of FinFET node.
FinFET technology is expected to play a critical foundational role across multiple AMD product lines, starting in 2016, GLOBALFOUNDRIES has worked tirelessly to reach this key milestone on its 14LPP process. We look forward to GLOBALFOUNDRIES’ continued progress towards full production readiness and expect to leverage the advanced 14LPP process technology across a broad set of our CPU, APU, and GPU products. – Press Release
AMD’s Polaris architecture-based 14nm FinFET GPUs deliver a remarkable generational jump in power efficiency. Polaris-based GPUs are designed for fluid frame rates in graphics, gaming, VR and multimedia applications running on compelling small form-factor thin and light computer designs.
AMD's First HBM2 GPU is Vega - Arrives in 2017 For High-Performance Radeon Gaming and Compute Demanding Firepro Markets
We have already talked a lot about Polaris 10 and Polaris 11 but AMD is in a good mood and they are already showing what they plan to bring in the years ahead. First up, we have the Vega GPU which is said to be known as Greenland internally by AMD before they switched to their new naming scheme.
The Vega GPU from AMD is the next generation architecture from AMD that utilizes the 14nm FinFET process node but enhances the architecture in every possible way to deliver the best compute capabilities and gaming performance to users. Vega GPU has already been speculated and alleged to be part of the HPC APUs which are under development by AMD. This makes us believe that not only will Vega be aimed at enthusiast gamers but also the FirePro market which needs serious amount of compute performance. AMD also wants to deploy a new interconnect fabric during the same time as Opteron Zen and Vega FirePro products are introduced to keep up in the HPC space against the likes of Intel and NVIDIA who are quite dominant in this sector.
The GPUs arriving in 2017 will also be the first to deploy HBM2 memory which means that current GPUs will have to stick with either GDDR5/GDDR5X or HBM1 solutions. Knowing that one of the two Polaris GPUs uses GDDR5 memory, it is possible that AMD is reserving HBM2 for their compute oriented cards and makes sense since their collaborative partner, SK Hynix, also plans to begin HBM2 memory production in Q3 2016. NVIDIA on the other hand is expected to release their Pascal GPUs for the supercomputer and HPC market with HBM2 in 2016 since they are said to partner with Samsung to deploy HBM2 memory on their GPUs. Samsung initiated mass production of their 4 GB HBM2 chips in early Q1 2016.
Navi GPU Is AMD's True Next-Gen Core - Arrives in 2018 and Features Next Gen Memory With Extremely Scalable Design
The true next gen product that the AMD showed on its roadmap has to be Navi which is a GPU totally unheard of but can now be confirmed as AMD's 2018 flagship offering. Little details of the Navi GPU were released by AMD but the two enabling technologies it should feature include support for next-generation memory and a fully scalable design which means the GPU architecture will be able to scale down to entry level, low-power solutions and all the way top to the beefiest compute and performance oriented offerings.
Scalability is a big factor in offering better efficiency across different SKUs based on the same chip architecture. Having just one GPU scaling down or upward to several designs will save AMD a lot of R&D and still deliver a great GPU architecture which won't require the development of several different GPUs for different types of platforms.
AMD's Fury X was the first graphics card to deploy HBM
What's more interesting about Navi GPU is that it will deploy a completely new memory standard which AMD refers to as Nextgen memory. There are no hints to what this memory type would be but just like AMD, NVIDIA also believes that HBM2 is not here to stay for much long. As bandwidth demand increases in the professional and HPC market where Vega and Navi are definitely being pointed towards, there exists a looming memory crisis which was discussed by NVIDIA at SC15.
On further explaining the next generation GPU architectures and efficiency, Stephen pointed out that HBM is a great memory architecture which will be implemented across next generation GPUs including Pascal, Volta and Vega but those chips have max bandwidth of 1.2 TB/s (Volta GPU). Moving forward, there exists a looming memory power crisis. HBM2 at 1.2 TB/s sure is great but it adds 60W to the power envelope on a standard GPU. The current implementation of HBM1 on Fiji chips adds around 25W to the chip. Moving onwards, chips with access of 2 TB/s bandwidth will increase the overall power limit on chips which will go from worse to breaking point. A chip with 2.5 TB/s HBM (2nd generation) memory will reach a 120W TDP for the memory architecture alone, a 1.5 times efficient HBM 2 architecture that outputs over 3 TB/s bandwidth will need 160W to feed the memory alone.
This is not the power of the whole chip mentioned but just the memory layout, typically, these chips will be considered non-efficient for the consumer and HPC sectors but NVIDIA is trying to change that and is exploring new means to solve the memory power crisis that exists ahead with HBM and higher bandwidth. In the near future, Pascal and Volta don’t see a major consumption increase from HBM but moving onward in 2020, when NVIDIA’s next gen architecture is expected to arrive, we will probably see a new memory architecture being introduced to solve the increased power needs.
It is possible that both AMD and NVIDIA will try to get memory manufacturer's to develop a new memory standard to cope with this problem. We might refer to this memory type as HBM3 which will be the third generation of the memory standard but right now we know nothing more than the fact that next iteration graphics cards will indeed be deploying new memory types. There's a lot of good stuff coming ahead and performance in graphics processors is expected to increase ten folds with next generation designs that offer better performance and efficiency.
AMD Radeon Graphics Architectures
|GPU||Process||Transistors In Billions||Performance||Memory||Bandwidth|
|Southern Islands||2012||HD 7970 Ghz||Tahiti||28nm||4.3||4.1 TFLOPS||3GB GDDR5||264GB/s|
|Volcanic Islands||2013||R9 290X||Hawaii||28nm||6.2||5.6 TFLOPS||4GB GDDR5||320GB/s|
|Caribbean Islands||2015||R9 Fury X||Fiji||28nm||8.9||8.6 TFLOPS||4GB HBM1||512GB/s|
|POLARIS||2016||RX 480||Polaris 10||14nm||5.7||5.2 TFLOPS||8GB GDDR5||256 GB/s|
|VEGA||2017||RX Vega 64||Vega 10||14nm||12.5||13 TFLOPS||8GB HBM2||480GB/s|
|NAVI||2019||TBA||Navi 10||7nm||TBA||TBA||Nextgen Memory||TBA|