Headed into CES 2020, AMD was in an interesting position. The company is coming off what’s arguably its best year in the past two decades, with strong CPU launches in desktop, server, and workstation, the announcement of new Ryzen-based consoles (the PS5 and Xbox Series X), a new GPU architecture that competes much more effectively with Nvidia’s Turing cards, and new partnerships with data center and server companies. AMD’s Lisa Su has pledged to make 2020 an even bigger year for the company than 2019, before announcing the long-awaited Ryzen 4000 Mobile APU family.
Up until now, AMD’s mobile CPUs have been quad-core and dual-core parts. That’s changing on 7nm, with 8C/16T CPUs debuting in both 15W and 45W TDP brackets. The GPU, meanwhile, has been slimmed down — the Ryzen 4000 family will feature 8 or fewer GPU clusters, compared with up to 11 on the 12nm Ryzen 3000 family. According to AMD, changes to the process node will allow 8 CUs to outperform 11. The GPU is still based on Vega rather than utilizing RDNA, but AMD says it has enhanced the architecture for higher clocks and better IPC.
These new CPUs don’t use a chiplet design or a 14nm I/O die — the CPU and GPU are fully integrated. We’re not surprised to see AMD take this step — the benefits of closely coupling the CPU and GPU likely outweigh the drawbacks of moving the I/O and DRAM controllers down to 7nm, especially when space is at a premium. Here’s the 15W CPU family:
At the low end of the stack there’s the Ryzen 3 4300U, a 4C/4T chip with a base frequency of 2.7GHz, boost frequency of 3.7GHz, 2MB of L2, and just 4MB of L3. Assuming AMD has stuck with 64 cores per CU, this chip has a 320:?:? configuration (we don’t know how TMUs or ROPs scale with the number of GPU cores yet). At the high end, the 8C/16T Ryzen 7 4800U is capable of a respectable 4.2GHz boost clock with 4MB of L2, 8MB of L3, and 8 CUs clocked at 1750MHz.
Integrated GPU performance is typically memory bandwidth bound, so any improvements AMD has made to Vega’s efficiency will pay dividends here. The Ryzen 4000 family uses the same CCX concept as the other Ryzen CPUs AMD has launched, but offers just 4MB of L3 cache per CCX rather than 8MB. These CPUs support up to 64GB of LPDDR4X memory and the IF clock runs independent from RAM clock to allow the CPU to hit lower idle power states. AMD claims to have improved latency when entering and exiting idle states by 80 percent, allowing more of the CPU to power down when in idle mode.
AMD is making several different power and efficiency claims, including up to 2x improved performance per watt, attributed to 30 percent efficiency gains from the CPU and 70 percent process node improvements. Overall SoC power is down 20 percent. AMD’s 15W CPUs will have configurable TDP envelopes up to 25W, while its 45W CPUs will turbo up to 54W for short intervals.
The 45W CPUs have fewer CUs than the 15W chips and much higher base clocks. AMD claims that the lower number of CUs is because both chips are being used with dGPUs, but it’s also possible the company plans to launch a Ryzen 9 with higher turbo and a full GPU. Anandtech reports that AMD has actually built a custom APU for Asus, the Ryzen 7 4800HS, which offers 45W worth of performance in a 35W form factor. AMD is also announcing features like Smart Shift, which allocates power between the CPU and dGPU to ensure maximum performance. It’s not clear, however, which laptops will support Smart Shift right now (AMD claims a 1.1x improvement in The Division 2 and a 1.12x improvement in Cinebench).
These chips are expected to ship in Q1 2020, with actual OEM availability in the March-May timeframe. AMD and Intel have been in a rather odd position of late. Intel took 10nm into mobile first, while AMD launched 7nm for desktop and server. The upshot of this is that both companies have been competing against a leading-edge node with their own last-gen products. A head-to-head comparison of the AMD Surface versus the Intel Surface showed the Intel machine well ahead in many tests, but that’s a 10nm CPU compared with AMD’s 12nm APU built on GlobalFoundries’ process technology. Once the Ryzen 4000 series debuts, we’ll be able to see which company’s leading-edge design is better.
Let’s block ads! (Why?)
Read more here: ExtremeTechComputing – ExtremeTech