AMD’s New 6-Core and 8-Core APUs Are a Bigger Deal Than They Might Seem

This week, AMD launched new desktop silicon for gamers and enthusiasts who want the horsepower of the latest Zen 3 CPU but at a more reasonable price. The 5600G won more accolades in that regard than the 5700G, but both chips are a bigger deal than they might seem — and not just because we’re in the middle of an ongoing GPU shortage.

AMD has been talking about combining CPU and GPU on the same piece of silicon for well over a decade. The company’s entire “APU” concept relies on the idea of CPU and GPU as equal co-processors. AMD took a shot at creating this future when it launched its HSA initiative, but not much ever came of it. AMD had too little market power and its CPUs were not attractive enough to attract much developer attention back in 2012 – 2015. While the GPU could be used, in some cases, to accelerate workloads beyond what the CPU could deliver, only a small amount of software was ever HSA accelerated.

APU’s have always had a problem: From an enthusiast certain point of view, the whole is weaker than the sum of its parts. Serious and even midrange gamers have had little use for the integrated graphics, especially since modern GPUs can almost always drive three displays off a single card. Thus, enthusiasts have balked at paying for graphics capability they don’t particularly value. The value from the CPU side of things has been limited by the fact that APUs were historically quad-core chips. Ten years ago that was fine, but six-core and eight-core deployments are growing rapidly.

As recently as last year, dual-core and quad-core systems accounted for two-thirds (65 percent) of all Steam installations. Today, the count is much closer — 52.35 percent versus 45.98 percent. At the current rate of change, CPUs with >4 cores will be deployed in more than 50 percent of systems by early 2022. Enthusiasts are moving to higher core count CPUs, and APUs need to keep pace with that growth. By offering chips with higher core counts, AMD is keeping pace with the growth in higher core count deployments and offering customers who need CPU performance but not much GPU capability an affordable option they can take advantage of.

This shift to 6-core and 8-core chips addresses the limited amount of CPU performance previously available from APUs, while integrating graphics into more chips gives AMD something it’s been lacking: Guaranteed Radeon graphics deployed across a wider range of chips. This has potential long-term significance for AMD’s ability to treat its integrated GPU like an AI acceleration unit.

AMD and Intel already offer this kind of capability in certain circumstances; applications like Topaz Video Enhance AI will run on existing integrated GPUs from both AMD and Intel. In the future, we expect to see this kind of feature become much more widespread. We’ve heard rumors that AMD’s future APUs won’t just offer graphics, they’ll integrate a slice of GPU into every chiplet.

This does not preclude the possibility that AMD or Intel might also develop an integrated AI accelerator intended for the broad x86 ecosystem, but it seems more likely that both companies will go the GPU route for the short-to-medium term. While various mobile vendors have shipped dedicated AI processors, Intel and AMD have not. iGPUs are the logical target for such workloads, especially since Intel has recently invested in improving its integrated graphics (mostly on mobile, granted). The 5600G and 5700G are based on Vega, but we know that an RDNA2-equipped APU will ship at some point in the not-too-distant future, given that the Steam Deck sports this configuration.

RDNA2 should boost integrated graphics performance modestly, but it also ought to offer the same ~1.25x uplift in IPC that characterized RDNA versus GCN. The advent of DDR5 next year will boost total available memory bandwidth to the chip; even systems utilizing DDR5-4800 will offer 76.8GB/s of memory bandwidth compared to 51.2GB/s for existing DDR4-3200 configurations. A system with DDR5-6400 would 102.4GB of RAM bandwidth. It wasn’t so very long ago that one needed a quad-channel DDR3 motherboard to hit that kind of bandwidth and integrated GPUs will benefit from the increase.

Right now, AMD doesn’t have a ton of objective reasons to invest in consumer AI, because it’s share of the overall graphics market is rather small, particularly where higher-end discrete cards are concerned. Improving the capabilities of its on-die graphics hardware will allow AMD to grow its potential user base for such applications by guaranteeing that every customer who purchases a future Ryzen CPU has an integrated GPU whether they use it for gaming or a primary display or not.

This kind of capability addition is a long-term rollout. Intel has arguably put a much larger push behind some of these capabilities, with its work on OneAPI and its focus on optimizing workloads for “XPUs”, which is to say, optimizing workloads to run on CPUs, GPUs, or FPGAs. But AMD’s recent purchase of FPGA manufacturer Xilinx and the growth of AI in every market suggests that AMD has plans for compute that go beyond traditional x86.

Continue reading

AMD’s New Radeon RX 6000 Series Is Optimized to Battle Ampere

AMD unveiled its RX 6000 series today. For the first time since it bought ATI in 2006, there will be some specific advantages to running AMD GPUs in AMD platforms.

Nvidia Will Mimic AMD’s Smart Access Memory on Ampere: Report

AMD's Smart Access Memory hasn't even shipped yet, but Nvidia claims it can duplicate the feature.

AMD’s Retail Market Share Surges Based on European Reseller Data

AMD's European channel sales figures continue to be excellent, with a particular surge in November and December 2020.

AMD’s Reliance on TSMC Isn’t Harming the Company’s Growth Prospects

It has been difficult to buy high-end PC components for nearly six months. There are a number of reasons for this, including pandemic-related impacts, the related surge in demand for all computing hardware, and supply shortages. A lot of eyeballs have been trained on foundries like TSMC, to the point that national governments have put…