Reg comments22

EPYC leak! No, it's better than celeb noodz: AMD's forthcoming server CPU

Specs and performance deets in the wild

AMD CEO Lisa Su speaking at the firm's 2015 financial analyst day
AMD CEO and president Lisa Su

Media site VideoCardz has leaked two AMD EPYC 7000 server CPU slides revealing core, thread and clock details.

EPYC, the renamed Naples, is AMD's forthcoming server CPU design, to be followed by updates codenamed Rome and Milan.

It features:

  • Up to 32 Zen cores
  • 8 DDR4 Channels/CPU – up to 2,666 MHz
  • Up to 2TB Memory per CPU
  • 128 PCIe lanes
  • Dedicated security subsystem
  • Integrated chipset
  • Socket-compatibility with next-gen EPYC processors

The processor variants and their core/thread counts and clock details are:

AMD_EYPC_7000_A

The minimum is 8 cores.

Here is a slide showing costs and a SPECint benchmark comparison with a Xeon 1-socket processor.

AMD_EYPC_7000_B

We note that Zen has 128 PCIe lanes, a lot compared to Intel, but when you make a 2-socket Zen machine, 64 lanes are used by each to talk to each other, as the architecture is built on PCIe. Both single-socket and 2-socket systems have 1x128 or 2x64 (128) PCIe lanes.

AMD slides say a single-socket system will be better than equivalent Intel Xeon because it will lower power consumption and provide up to 20 per cent lower acquisition cost. Customers can use it to right-size under-utilised servers, and optimise storage and heterogeneous compute.

The leaked slides focus on single-socket EPYC vs single-socket Xeons, which might suggest a 2-socket EPYC vs Xeon comparison would be less advantageous to AMD.

Expect EPYC 7000 to be announced on June 20.

+Comment

AMD's basic appeal seems to be to build cheap Xeon substitutes, with Intel not quite managing to kill AMD off, and so we all avoid Intel having an absolute x86 server monopoly but still pay high x86 prices.

Can AMD struggle up from its current gnat status to be big enough to provide real competition to the Intel elephant? That seems unlikely, unless Intel makes a whole series of marketing/product missteps.

We wonder if state-encouraged Chinese technology groups, such as Tsinghua Unigroup, have cast an acquisitive eye over AMD, thinking buying it could enable them to gain some independence from Intel. Or, alternately, is China's long-term server processor technology plan to have its own server architecture and so gain even more independence?

The x86 architecture is so utterly dominant that buying AMD would seem to be more effective. Would the US government allow such a sale?

Other possible acquirers could be, at a stretch, Dell and HPE, looking to to get their own independence from Intel and bring down server CPU costs.

It does seem to us that AMD needs a relationship with a deep-pocketed organisation rather than remaining independent. ®


Biting the hand that feeds IT © 1998–2017