AMD's (AMD - Get Report) launch event on Wednesday late afternoon for its newest server CPU family featured some heady performance claims, as well as votes of confidence from a slew of major server makers and big-name Internet companies.
Optimism about the new chips helped propel the chipmaker's stock price up more than 14% to $33.39 by late morning Thursday.
At its event in San Francisco, AMD officially launched its second-gen Epyc server CPU line, which has been code named Rome. As previously indicated, the most powerful Rome CPUs will feature 64 cores, or twice as many as the most powerful CPUs for the first-gen Epyc line, which was code named Naples and launched in mid-2017. The CPUs rely on Taiwan Semiconductor's (TSM - Get Report) advanced 7-nanometer (7nm) manufacturing process, as well as a revamped CPU core microarchitecture known as Zen 2.
AMD's shares rose 4% in after-hours trading Wednesday following the event, and built on those gains on Thursday. Shares had sold off last week, paring some of their big 2019 gains in response to light revenue guidance that AMD largely blamed on soft game console demand ahead of expected 2020 console launches.
AMD also stated Rome CPUs, which feature up to 8 small CPU chips (known as chiplets) that are connected to a larger chip that handles input/output (I/O) and memory connectivity, will deliver a sizable increase in memory bandwidth relative to Naples CPUs and double the amount of memory supported per CPU socket to 4 terabytes (TB). In addition, whereas Intel's (INTC - Get Report) latest Xeon server CPU platform (code named Cascade Lake) supports the older PCIe 3.0 standard for connecting to networking and accelerator cards, Rome CPUs will support the newer PCIe 4.0 standard, which can deliver twice as much throughput.
Officially, AMD claims Rome CPUs will deliver 1.8 to 2 times as much performance as comparable Cascade Lake CPUs. The company said its flagship, 64-core, Rome CPU, the Epyc 7742, beat a high-end Cascade Lake CPU, the Xeon Platinum 8280L, by over 80% in popular benchmarks. It also stated Rome systems have set 80 world records -- cutting across many types of server workloads -- in tests conducted by OEM partners HP Enterprise (HPE - Get Report) , Lenovo, Gigabyte and Super Micro (SMCI) .
"Early indications from AMD and its OEM partners show that 2nd Gen Epyc is performing exceptionally well and has advantages in many different workloads, but not all workloads," said tech analyst Patrick Moorhead after taking in AMD's performance data. Among other things, Moorhead was impressed with the performance of AMD-powered systems when running analytics, high-performance computing (HPC) and Java application workloads. At the same time, he noted the AI-related optimizations of Cascade Lake CPUs, and their support for Intel's Optane next-gen memory, will give Intel an edge with certain AI and in-memory computing workloads.
Most of AMD's after-hours gains came after the company officially shared pricing information for the 19 products in its Rome lineup. With the qualifier that major customers will likely get discounts, list prices run between $450 for AMD's 8-core Epyc 7232P CPU, to $6,950 for the 64-core Epyc 7742.
AMD made a point of showcasing how various Rome CPUs were cheaper than Cascade Lake CPUs that (based on its performance numbers) they comfortably outperformed. In the case of the Epyc 7742, AMD noted the CPU's list price is nearly 50% below that of Intel's 28-core Xeon Platinum 8280M, which officially goes for $13,012; the 8280L, which had appeared in benchmark comparisons, lists for $17,906.
Execs from server OEMs such as HPE, Dell Technologies (DELL) , Lenovo and Cray (set to be acquired by HPE) made appearances during AMD's event. HPE, for its part, announced that three Rome-powered systems are launching today, and that it expects to have 12 workload-optimized systems a year from now.
Microsoft (MSFT - Get Report) , Google (GOOGL - Get Report) and Twitter (TWTR - Get Report) execs also appeared at the event, and an Amazon Web Services (AWS) exec voiced AWS' support for Rome through a video. Twitter said it plans to deploy Rome CPUs this year, while claiming they allow the company to cram 40% more CPU cores per server rack (and along the way, achieve major power efficiency and total cost of ownership (TCO) gains).
Microsoft announced its Azure cloud platform, which was one of the first public cloud platforms to support Azure, is rolling out a slew of new Azure cloud computing instances that are powered by Rome CPUs -- including ones specifically meant for HPC, memory-intensive and virtual desktop workloads. A Microsoft representative indicated that Rome instances could be generally available in November, and could (after general availability) be about 10% cheaper than comparable instances currently are.
Notably, Google indicated that it not only plans to offer Rome-powered computing instances on the Google Cloud Platform (GCP), but has also begun deploying Rome for a variety of internal workloads.
Intel has long maintained a dominant position in the server CPU market. However, with the help of a manufacturing process edge -- the 7nm TSMC process used by Rome is more advanced than the 14nm process used by Cascade Lake -- Rome provides AMD an opportunity to gain meaningful ground.
Last month, Intel indicated volume shipments for its Ice Lake Xeon server CPU platform, which relies on a 10nm process that's seen as competitive with TSMC's 7nm process, won't start until the second half of 2020. Around that time, AMD is expected to launch a next-gen Epyc platform (code named Milan) that relies on a slightly more advanced, 7nm, TSMC process and a CPU core architecture known as Zen 3.
AMD has previously said it believes it had a mid-single-digit server CPU market share as of the end of 2018, and that (with Rome's help) it aims to achieve a double-digit share within four to six quarters of having reached a mid-single digit share.