IBM is feeling Powerful in the Core Wars, details on the Power9 architecture have arrived

Subject: General Tech | April 7, 2016 - 03:43 PM |

IBM's Power9 processor is scheduled to appear on the scene just over a year from now and finally we have some details about what it will be.  Firstly the core count is to be two higher than Intel, 24 cores and is optimized for use in two socket servers.  The chips are 14nm FinFETs fabbed by GLOBALFOUNDRIES which will be compatible with modern industry standards including DDR4, PCIe 4.0 and NVLink 2.0 so you can even take advantage of Jen-Hsun's latest products. 

The list of customers is quite impressive, Google has moved to Power8 already and described changing to the infrastructure as simple as flipping a switch,  the US Department of Energy will build their next HPCs using Power9 and Rackspace is currently working with Google to develop Power9 server blueprints for the Open Compute Project. 

Several Chinese companies will take advantage of those OpenPower blueprints to develop their own 'partner chips', Power8 and 9 architecture which will be using 10nm gates in 2018 to 2020.  This is somewhat amusing considering the shipping of Xeon processors to China has been banned by the US Government.  Check out more of the slides from IBM's presentation at The Register.

View Full Size

"IBM's Power9 processor, due to arrive in the second half of next year, will have 24 cores, double that of today's Power8 chips, it emerged today.

Meanwhile, Google has gone public with its Power work – confirming it has ported many of its big-name web services to the architecture, and that rebuilding its stack for non-Intel gear is a simple switch flip."

Here is some more Tech News from around the web:

Tech Talk


Source: The Register

Video News

April 7, 2016 | 03:56 PM - Posted by quest4glory

Good, I love competition.

April 7, 2016 | 05:14 PM - Posted by Ophelos

I'll take a SPARC64 XIfx 32core cpu instead. :)

April 10, 2016 | 02:14 AM - Posted by Anonymous (not verified)

Yeah, let me know where I can by those. Right now Power machines are available, Sparc64 machines aren't. Unless Fujitsu and/or Oracle starts selling their hardware trough open channels they aren't an option. X86, Power, Arm, Mips and Risc V. Those are in the game. Sparc64 isn't.

Most of us would never buy X86 if we had to call Intel sales for it. We buy it from local vendors that we have a relationship with. Out of their price list and other forms. It took IBM quite some time to understand this but now that they do, they will eat market share.

April 7, 2016 | 05:28 PM - Posted by StephanS

To go in depth, the chips are manufactured by the government of the islamic state of abu dhaby (ruled under sharia), that acquired/took control of AMD and IBM silicon manufacturing.

April 7, 2016 | 07:20 PM - Posted by Anonymouse (not verified)

And I guess Sharia laws would make it an inefficient undertaking?

April 7, 2016 | 11:11 PM - Posted by johnc (not verified)

Depends on how many plants get blown up.

April 7, 2016 | 05:29 PM - Posted by anon (not verified)

What happened to that dick who would spam these comments with all that "Power8 will be on phones and desktops soon" Does that annoying prick still post here? He must have a massive hard on now

April 8, 2016 | 10:05 AM - Posted by Anonymous (not verified)

What about all the gaming gits out there that think technology is a football game and are loyal for illogical reasons to any maker's BRAND, and the Power8 is a server/HPC SKU! So far the Power9 looks to be keeping Intel outside some of Google's and others server/HPC kit. And the Power ISA could be used for some desktop PC products! in fact some server variants with Nvidia GPUs could be made into a very nice home gaming system. Sure no Phone variants will be coming anytime soon(if ever as there is ARM), but Nvidia could brand a high end gaming PC using maybe an 4 or 8 core Power8 processor tied to Nvidia graphics, and probably do fine. NvLink is not yet available for x86 based systems, but maybe Nvidia could put Intel on the outside of some power8 based gaming SKUs and use Nvlink to tie more than one Nvidia GPU to a 4 or 8 core Power8 based gaming SKU running a Linux OS/Vulkan API based custom gaming ecosystem.

It looks like Nvidia has got their GPU processor thread dispatch/scheduling granularity issues down to the instruction level boundary, so no more having to wait for the end of the draw call to issue new work on its GPU processor threads for Pascal based SKUs. We will see how Nvidia's improvements stack up against AMD's Polaris. One more thing to note about the power9 SKUs is that they will now longer need the pricy Centaur external controller chips so the Power9's will be that much more competitive against Intel's server offerings. Hell Nvidia could do well building a high end home gaming server(Linux/Vulkan) with a 8 core power8, or power9, and an 8 core power8/9 will provide 64 processor threads. It's the third party power8 variants from all the openpower licensees that may make for some affordable compared to Intel's pricy offerings power based gaming servers from Nvidia. Why should Nvidia worry about having an x86 license when Nvidia could get a power8/power license form openpower and maybe start making its own PC gaming systems!

P.S. The Power8/Power ISA is a RISC ISA just like ARMv8A is a RISC ISA so nothing is stopping Nvidia, or AMD and others from beefing up a custom ARMv8A ISA based CPU core to rival the Power8's performance metrics, with power8 having 16 execution pipelines to feed those 8 processor threads per core. In fact AMD's K12 may be the first ARMv8A running custom core that supports SMT, so nothing is stopping AMD, Nvidia, or others from making some powerful ARM variants.

April 8, 2016 | 11:38 PM - Posted by collie

Ahhhhhh, that's the shizzle! Thanks for the hit!

April 8, 2016 | 03:12 AM - Posted by Anonymous (not verified)

Those are going to be some very expensive chips with the massive amount of interconnect they support. It says 32 DDR4 slots, but how many channels? PCIe 4.0 by itself is 31.51 GB/s for an x16 link, and that doesn't even get into NVLINK and some of the other interconnect on these chips. There are some applications where such a system will outperform using a larger number of lower powered processors though.

April 8, 2016 | 06:13 AM - Posted by Pixy Misa (not verified)

Sounds like a dual socket server, and if it's two DIMMs per channel, that's eight channels per socket. Which is a lot, but AMD are talking about something similar for their Xen-based server chips.

April 8, 2016 | 10:25 AM - Posted by Anonymous (not verified)

AMD's Zen based server APUs on an interposer variants will offer much more CPU to GPU bandwidth than even Nvlink, or other high speed coherent connection fabrics. This is because AMD will be able to etch thousands bits of wide parallel direct traces between the CPU/Cores die and its Polaris/Vega GPU die, so AMD will have the ability to directly wire the CPU cores to its GPU die via thousands of traces, just look at the number of traces the HBM memory gets at 1024 bit wide per HBM die stack, and AMD could probably have a dedicated direct parallel data path more than 4096 bits wide from the CPU/cores die to GPU's die and the GPUs thousands of GPU cores.

PCI/NVLink has got nothing on HBM for low clocked high effective bandwidth, so a direct APU on an interposer server/HPC, or even a PC Zen/Polaris APU variant on an interposer is not out of the question for AMD, and Intel would be forced to go with an interposer based SOC if it wanted to compete. It's the interposer's ability to have thousands of bits wide traces etched into the interposer's silicon substrate that will give AMD some very powerful HPC/Server and PC APUs on an interposer based SKUs the will provide more effective bandwidth than any PCI based gaming system.

April 8, 2016 | 01:51 PM - Posted by Anonymous (not verified)

Intel and AMD are making their CPUs dysfunctional without binary blobs.
Save us, IBM, you're our only hope. Well, you and ARM.

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

This question is for testing whether you are a human visitor and to prevent automated spam submissions.