⋮  

AMD x86 Zen Based Raven Ridge APUs To Be Made of GloFo’s 14nm Process – Amkor To Package Next Generation HBM APUs

Hassan Mujtaba
Posted Mar 20, 2016
143Shares
Share Tweet Submit

This year will see the launch of AMD’s Zen core which is a completely new architecture to be embedded on next generation FX and Opteron CPUs. The Zen core is coming to enthusiast first followed by server parts and there’s no doubt that AMD will also have Zen based APUs next year, featuring some major improvements in both dGPU and CPU department. The updated AMD family of APUs is suggested to be known as Raven Ridge.

AMD’s Raven Ridge APU Family Based on 14nm Process, Highly Efficient GPU and HBM DRAM

APUs have been a major investment front of AMD on the mobility sector. Their entire mobility processor lineup consists of APUs. It’s no hidden fact that AMD has a more modern APU lineup on the mobility front compared to the desktop platforms. The top Carrizo APUs were available first on the mobility platforms and will launch on the desktop platforms this year with AM4. AM4 is the name of the new platform that will deploy two modern lineups from AMD, Summit Ridge and Bristol Ridge. The Summit Ridge processors are the first processors from AMD to utilize their next generation Zen core while Bristol Ridge brings the efficient Carrizo design on desktop PCs.

The Bristol Ridge family is entirely a Carrizo lineup on the desktop platforms. With Carrizo, AMD managed to get a lot of performance per watt improvements. AMD has stated that these improvements are close to 2.4x perf / watt, about the same as what AMD is getting on Polaris GPUs this year. Bristol Ridge will deploy a range of new features, enhancing the I/O capabilities of modern day APUs and delivering better IPC. We even heard of a 16 CU APU being launched under the Bristol Ridge lineup but whether this APU is just a rumor or something that might launch later in the timeline is expected to be seen.

Bristol Ridge is all great for consumers who want to built a modern PC with decent graphics power however it doesn’t bring anything new on the drawing board while Summit Ridge uses the Zen core. For users who want the Zen core on APUs, they should for Raven Ridge which is the first Zen based APU lineup from AMD. The line of APUs not only deploy the next-gen, 14nm Zen architecture but also feature other enhancements that range from higher bandwidth, faster and efficient dGPUs and performance parity with modern day consoles.

GloFo and Amkor Duo To Package 14nm Raven Ridge APUs For 2017

The latest rumor or should we say, update, comes from BitsnChips who have reported that AMD’s Raven Ridge APUs will be built by Global Foundries and packaged by South Korean Semiconductor company, Amkor. Amkor was the same company that packaged AMD’s Fiji GPUs (HBM+GPU+TSV). AMD will be using the GloFo-Amkor Duo to also package their next-gen Raven Ridge APUs that will utilize the next generation of HBM standard to deliver higher bandwidth to the faster discrete-class graphics chip featured on the SOC.

14nm FinFET Process

Samsung’s 14nm FinFET process which Globalfoundries has licensed is going to be the basis of all of AMD’s next generation CPU, APU and GPU products.  The company’s Chief Technology Officer Mark Papermaster made the announcement last year that the 14LPP process – Samsung’s high performance 14nm process – will be leveraged across all of AMD’s future products.

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

  • The 2.5D/3D interposer construction is a viable approach to single package MCM system design and provides several benefits:
  • Enablement of very high bandwidth wide-bus memory implementations
  • Employment of die utilizing the best technology node at the best price and performance, in some cases permitting re-use of critical IP blocks and die
  • Die partitioning and optimization for memory, analog, performance, power management and cost
  • Higher effective die yields obtained through die recovery
  • Accelerated time-to-market
  • Risk reduction in the overall schedule

It is exciting to know that Raven Ridge will be a 14nm design from top-to-bottom. Since these APUs arrive in 2017, we can expect the Polaris GPU architecture to be deployed on these APUs which delivers up to 2.5 time performance per watt improvement compared to current generation GCN architecture.

AMD’s Carrizo APUs currently deploy the same GCN architecture which is found on Tonga and Fiji and Bristol Ridge is expected to keep the same design. Raven Ridge which deploys the GCN 4.0 GPU design will not only help APUs perform faster in graphics performance but also much more efficiently. AMD’s first Polaris demo was of an entry level chip which delivered great performance for a sub-40W design.

But with more graphics power comes more demand of bandwidth which unfortunately doesn’t provide a lot of bandwidth to faster graphics cores. Intel’s solution to feed higher bandwidth was faster eDRAM cache but AMD is deciding to go the HBM route. The HBM will act as fast DRAM which the GPU will have full access to. That’s the part two of the statement by BitsnChips who state that Amkor packaging AMD APUs is a hint that these are going to deploy the fast HBM DRAMs. Which we’ve also seen hints of in official AMD slides.

It’s not known what specific amount of HBM will be featured but a single HBM2 stack can deliver as much as 4 GB memory and 256 GB/s bandwidth. You can clock it up and down but the most entry level stack can still get as much as 128 GB/s. AMD’s HBM partner SK Hynix will begin production of 4 GB DRAMs in Q3 followed by 8 GB DRAMs in Q4 2016. They also have 2 GB but there’s no date issued how ever these low VRAM DRAMs sound most ideal for Raven Ridge APUs.

Regardless of the amount of VRAM, the bandwidth increase will be more than enough for the dGPU on board Raven Ridge and the 40% IPC increase just from the Zen cores will lead to a huge performance gain on AMD’s next gen APUs. The performance of AMD Raven Ridge is expected to reach or even exceed the levels of Xbox One and PlayStation 4 performance due to better CPU, GPU and memory architecture. The Raven Ridge APUs will hit the AM4 platform and will be exciting to see how AMD handles these high-performance mainstream / consumer APUs when they launch in 2017.

AMD Next Gen CPU Families:

WCCFTechAMD Raven RidgeAMD Summit RidgeAMD Bristol RidgeAMD FX "8000 Series"AMD A-Series "7000/8000-Series"
Product SegmentMainstream Desktop and Mobility APUPerformance Desktop Processors "FX"Mainstream Desktop and Mobility APUPerformance Desktop Processors "FX"Mainstream Processors "Kaveri/Godavari"
Product Architecturex86 Zenx86 Zenx86 Excavatorx86 Bulldozer/Piledriverx86 Steamroller
Process Node14nm14nm28nm32nm28nm
Max CPU CoresTBATBATBA84
GPU ArchitectureGCN 4.0N/AGCN 3.0N/AGCN 2.0
TDPTBATBATBA125-220W95W
SocketAM4AM4AM4AM3+FM2+
South BridgePromontory/SOCPromontoryPromontory/SOC990FXA88X
Memory SupportDDR4/HBMDDR4DDR4DDR3DDR3
Launch2017Q4 20161H 20162011-20162013-2016

Share Tweet Submit