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... .
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.
There was a lot announced at [the April 7] OpenPower Summit in San Jose, California. Here's what went down... .
[...] The Power9 will be a 14nm high-performance FinFET product fabbed by Global Foundries. It is directly attached to DDR4 RAM, talks PCIe gen-4 and NVLink 2.0 to peripherals and Nvidia GPUs, and can chuck data at accelerators at 25Gbps.
[...] The Power9 is due to arrive in 2017, and be the brains in the US Department of Energy's Summit and Sierra supercomputers.
Summit's peak performance should be 300 peta-FLOPs, thrashing China's leading 55 PFLOPS Tianhe-2, but a good chunk of the American system's performance will come from the Nvidia Volta GPUs rather than the Power9s.
There's lots more; read the rest at: http://www.theregister.co.uk/2016/04/07/open_power_summit_power9/
takyon: A more recent article: Power9 Will Bring Competition To Datacenter Compute
(Score: 2) by Nerdfest on Friday April 22 2016, @06:12PM
rebuilding its stack for non-Intel gear is a simple switch flip.
This is why open source rocks. You tie yourself to a proprietary software supplier and you're at their mercy.
(Score: 2, Funny) by Anonymous Coward on Friday April 22 2016, @06:20PM
Not really. With enough reverse-engineering and emulation you can make closed-source software do anything and run on any hardware you want. Open source just makes you lazy.
(Score: 1, Funny) by Anonymous Coward on Friday April 22 2016, @06:32PM
So true. I hacked and hacked and finally managed to get closed-source Windows running on my PC. Not too stable yet; more of a proof-of-concept, but it's a start.
(Score: 0) by Anonymous Coward on Friday April 22 2016, @06:39PM
Wake me when you've installed Windows Millennium on your ARM-based phone. Yes, I've done it, and it was enough of a proof-of-concept to host a game server that was actually playable.
(Score: 0) by Anonymous Coward on Friday April 22 2016, @06:46PM
IPv4 and AMD aren't goog enough for Google.
(Score: 2) by bob_super on Friday April 22 2016, @08:54PM
> can chuck data at accelerators at 25Gbps
I'm pretty sure they mean per trace pair, and it would be surprising if there was only one pair. Otherwise that's pretty dang slow compared to even a boring PCIe gen 3 x16 (128Gbps).
(Score: 3, Insightful) by c0lo on Friday April 22 2016, @10:03PM
Too small a foothold in mobile devices [vox.com] (where the majority of the market is), now the data centres goes for Power.
No wonder the layoffs [soylentnews.org]
https://www.youtube.com/watch?v=aoFiw2jMy-0
(Score: 2) by Gravis on Friday April 22 2016, @11:24PM
I don't think data centers are all going POWER, i think they are just going with the most cost effective systems. there is the speed and cost of the system and then there is the cost of actually providing electricity to the system until it's removed. i think the POWER is providing the best balance of these two things, at least for now. i'm hoping that in a few years we'll see fast, inexpensive and low energy consuming RISC V systems. however, if you a locked into x86, your stupid ass is going to be paying a premium to continue running x86 software on servers.
(Score: 2) by bitstream on Saturday April 23 2016, @03:22AM
x86 is that something to heat the house with during cold winter months? ;-)
I heard P4 is especially good at this. :p
(Score: 2) by RamiK on Saturday April 23 2016, @07:58AM
The general case is still not in favor of POWER just yet. The total cost of ownership is different for Google since they maintain their own software stack, but other companies license Java from either Oracle for the x86, or IBM for POWER which keeps Intel slightly better off for now.
However, the linchpin isn't POWER or the x86, it's ARMv8. As ARM encroaches on the low-end servers and desktops markets, they're forcing Intel's to realign in unfavorable ways: Much of Xeon's BoM depends on Intel's ability to repackage defective high-end chips as lower frequency servers and desktop products. So, losing desktop and low-end server sales has a direct impact on the profitability of their high-end Xeon sales. When Intel finally started acknowledging this problem a few years ago, they begun pushing Atoms towards low-end servers in the hope to at-least cutting-off their direct loses there. But, they've failed. Their Atoms are being sold at a loss and are still not good or cheap enough.
So now, they're hemorrhaging twice as bad at the low-end. Not only are they not recuperating their Xeon costs, their Atoms aren't able to keep AMD's and Qualcomm's ARMv8 products from the market. Worse, if you look at medium-end MediaTek mobile SoCs, you'll find 10cores operating at 2ghz which is more than competitive with anything Intel has to offer at that price range. They've reacted by firing off 10% of their workforce in the hopes their next production node will carry them through. But they're likely too late seeing how Android is already baking-in laptop features while products like http://www.cnx-software.com/2016/04/23/sub-100-remix-os-laptops-powered-by-allwinner-a64-processor-are-coming-our-way/ [cnx-software.com] are setting an impossible price tag for Intel to compete against.
Regardless, both ARM and IBM have deep pockets to fund this war for decades to come so even if Intel pulls a hail Mary and survives the next production node, their production model is irrevocably damaged without the ability to recuperate loses so they'll just face worse odds in 2 years time anyhow.
compiling...