AMD Announces Ryzen 3: 8 CPU Cores, Zen 2 Architecture

At CES 2019, AMD announced that its third-generation Ryzen CPU core and accompanying Epyc will both be available from mid-2019 forward. AMD showed off an eight-core version of its Ryzen CPU and confirmed that it will use a chiplet design, with a 14nm GlobalFoundries I/O die and a 7nm CPU chiplet.

We weren’t sure if AMD would opt for a chiplet design at 7nm or continue with an integrated core and theorized that it might make more sense for AMD to continue to use a unified architecture. Instead, it’ll be chiplets down the stack, though the company isn’t re-using Epyc 14nm die as some had theorized. The chiplet die for Ryzen is physically much smaller (Anandtech estimates it at 25-50 percent the size of the Epyc I/O die). It’s interesting to see AMD taking this step at the consumer level as well as introducing it for servers and we’ll be curious to measure the low-level performance capabilities of the core.

According to AMD’s demo from the stage, the third-gen Ryzen is capable of a score of 2023 in Cinebench 15 at a power consumption of 130W (system power), while Intel’s Core i9-9900K is running at 2042 for 180W of system power. This is a substantial improvement in overall power consumption (the two systems were claimed to be running on identical hardware, apart from their CPUs). The AMD chip was not running at final clocks, which makes sense — clock speeds likely have not been finalized yet for these CPU cores. Neither have their prices. Price is the last launch detail to be finalized.

The performance uplift for 7nm Ryzen in Cinebench compared with previous 14nm parts is 15 percent. Cinebench is an excellent test case for AMD, so we can’t assume the 15 percent will be global across the company, but that’s a solid IPC boost. Without knowing more about clocks, we can’t speak to the overall level of performance improvement the company will over, but the IPC gains alone suggest that we should expect a range at least between 10-20 percent (the former number in worst-case scenarios in which tests don’t respond well to Ryzen’s improvements and the latter assuming AMD is able to increase clocks over and above what they showed on stage today). Relative performance between Intel and AMD is always benchmark dependent, but a 15-20 percent overall improvement would put Team Green on much closer footing with Chipzilla as far as efficiency and IPC are concerned. The power consumption improvements are extremely good overall — AMD clearly hit or exceeded its targets.

The big questions around clock will depend on how well TSMC’s 7nm node is tuned for high-performance silicon. It’s not crazy to ask this question; TSMC has never built desktop x86 CPUs for this space before and its 7nm nodes were initially deployed for the mobile market. AMD doesn’t pay for custom silicon any more, so we know the company isn’t running its own process variant at its foundry partner. If AMD’s clocks are low and TSMC’s 7nm hasn’t hit high frequencies (relatively speaking), then we’re looking at a substantial improvement to IPC that easily puts AMD on par with Intel. If TSMC can match or exceed Intel’s frequencies, then AMD could be set to pull ahead on clock without sacrificing overall performance the way it did during the Piledriver era. Either way, these are good results.

Finally, despite rumors to the contrary, AMD did not launch the Ryzen 3rd Generation family at CES. It did not announce model numbers, clock speeds or core counts for the new parts, beyond the 8-core chip that has been shown. It did not announce a $250 Navi that would compete with the RTX 2070. It did not announce a $100 6C/12T CPU. It did not announce a $449 16-core CPU, though it has hinted that such a part may come to market for desktop Ryzen later this year at an unknown price point. While we mostly covered the Radeon VII announcement in another piece, the fact that AMD’s 7nm Vega will take on the RTX 2080 at $700 means the rumor of a $249 Navi that can tackle the RTX 2070 is exceptionally unlikely — the architectural improvements required to deliver that kind of gain iso-process node relative to Vega would be unprecedented in AMD’s recent GPU history.

Fortunately, AMD never had to make such unlikely pronouncements (given what we knew about its launch timelines) at CES to deliver a solid show. A 15 percent IPC uplift and substantial power reductions compared with Intel’s best CPUs is more than enough to keep the company on the board and competitively positioned for 2019. Vega 7nm is a bit more of an unknown quantity, but the CPU firm appears well-positioned overall for 2019.

Continue reading

AMD Brings Its Zen 3 Architecture, Up to Eight CPU Cores to Chromebooks

AMD is going after the high-end Chromebook market with its newest Zen 3 APUs.

TSMC Announces ‘FinFlex’ 3nm Architecture With Variable Configurations

TSMC has revealed an audacious new customizable "FinFlex" architecture for its 3nm process.

Intel Confirms Entirely New Microarchitecture for Lunar Lake

We'll find out more about this low-power architecture on the 26th.

World’s Largest SSD Claims 100TB Capacity, Multi-Processor Architecture

Nimbus Data is claiming a new record for largest SSD, with 100TB of storage in a 3.5-inch form factor and a new multi-processor architecture.