On Tuesday, AMD added three CPUs to its second-gen Epyc line (codenamed Rome) that aim to grow its server CPU share in more traditional enterprise environments -- a space where Intel’s (INTC) - Get Report Xeon line is still quite dominant. Among the use cases that AMD claims the products are a good fit for: Database servers, enterprise high-performance computing (HPC) clusters and hyperconverged infrastructures that combine server and storage functionality and can scale out over time.
AMD also announced that several major enterprise server OEMs and software developers, including Dell Technologies (DELL) - Get Report, Hewlett-Packard Enterprise (HPE) - Get Report, Lenovo, Microsoft (MSFT) - Get Report and VMware (VMW) - Get Report, would be supporting the CPUs. Like existing Rome CPUs, the new products rely on Taiwan Semiconductor’s (TSM) - Get Report advanced, 7-nanometer (7nm) manufacturing process node and AMD’s Zen 2 CPU core microarchitecture.
The product launches have provided some fresh fuel for AMD’s recent rally. As of the time of this article, the CPU and GPU developer’s stock is up 7.3% on Tuesday to $54.68, amid a 3.6% gain for the Nasdaq. AMD is now only 8% below a February high of $59.27. Meanwhile, Intel shares were up 2.9% to $60.41
The biggest selling point for the new Rome CPUs is that their CPU cores sport higher clock speeds than existing Rome CPUs featuring the same number of cores. For example, whereas AMD’s Epyc 7402 CPU -- until now its most powerful 24-core Epyc offering -- has a base clock speed of 3GHz and a boost speed of 3.3GHz -- the new, 24-core, Epyc 7F72 has a base speed of 3.2GHz and a boost speed of 3.7GHz.
On the other hand, whereas AMD’s existing Rome lineup tops out at 64 cores, the new CPUs respectively have just 8, 16 and 24 cores. That’s likely to limit their appeal to cloud giants and supercomputer builders, which have been major buyers of 48 and 64-core Rome CPUs.
In addition, AMD is charging healthy premiums for the new CPUs relative to existing products with the same core counts. The 8-core Epyc 7F32 and the 16-core 7F52, respectively, list for more than three times as much as the products that were previously AMD’s most powerful 8-core and 16-core Rome CPUs, while the 24-core 7F72 carries a relatively modest 37% premium.
AMD is wagering that software license savings will nonetheless make the new CPUs attractive to enterprises. A lot of enterprise software is licensed on a per-core basis, and VMware, whose server virtualization software is widely deployed by enterprises, recently overhauled its licensing to charge twice as much when its software is used on CPUs containing more than 32 cores.
As a result, if the higher clock speeds of AMD’s new CPUs lower the number of CPU cores needed to support a particular workload by, say, 20%, a customer might still come out ahead, depending on what software is being run.
Also, AMD claims that its new CPUs still deliver more bang for the buck than comparable Intel CPUs when running popular workloads. For example, AMD claims the 7F32 has a 35% price/performance edge relative to Intel’s 8-core Xeon Gold 6244 CPU when running Microsoft’s SQL Server 2017 database (Intel, of course, might share a somewhat different number).
The CPU launches arrive five weeks after AMD, which has benefited from the delays Intel has seen in launching server CPUs relying on a 10nm manufacturing process node that’s seen as competitive with TSMC’s 7nm node, forecast it would achieve a double-digit server CPU share in Q2. Potentially giving the company a near-term lift: Cloud capital spending still appears to be strong, as tech giants add capacity to deal with strong usage growth amid COVID-19 lockdowns.
Intel plans to ship its first 10nm server CPUs (based on a platform known as Ice Lake) by year’s end, but it isn’t clear right now how large Ice Lake’s 2020 volumes will be. AMD, for its part, says it will launch a next-gen, 7nm, Epyc CPU line (codenamed Milan) in late 2020.