Slashdot Mirror
IBM Gearing Up Mega Power 8 Servers For October Launch
darthcamaro (735685) writes "Now that IBM has sold off its x86 server business to Lenovo, it's full steam ahead for IBM's Power business. While Intel is ramping up its next generation of server silicon for a September launch, IBM has its next lineup of Power 8 servers set to be announced in October. "There is a larger than 4U, 2 socket system coming out," Doug Balog, General Manager of Power Systems within IBM's System and Technology Group said. Can IBM Power 8 actually take on x86? Or has that ship already sailed?" At last weekend's Linux Con in Chicago, IBM talked up the availability of the Power systems, and that they are working with several Linux vendors, including recently-added Ubuntu; watch for a video interview with Balog on how he's helping spend the billion dollars that IBM pledged last year on open source development.
IBM touts the virtualization capabilities of Power, but I can't find any IaaS providers where I can rent a slice of one. I looked at the Softlayer site, they're and IBM company, and I couldn't find it there either. So, it leaves me to wonder...
Good luck taking on x86 dirt-cheap, "good enough" servers with exorbitantly costly, closed POWER systems which people cannot buy cheaply enough to play at home with and learn on. Where is the system administration and development user base going to come from? All such attempts by other companies have failed in the past.
Also, good luck getting POWER re-adopted by the system administration cand development community, when AIX and the compilers remain closed and the barrier to entry is really high (they are not easily available).
You priced yourself out of the market, IBM. Old school guys like me have all but retired, and when they do, they want nothing to do with AIX or POWER or IBM.
Who is going to bother jumping through the hoops to learn AIX on old POWER systems bought on ebay when they can get a fully loaded x86 server or put together a PC bucket at home and slap Linux or Joyent's SmartOS and just start learning? Who?
Latest Power workstation had Power 5 CPUs. The should make a new workstation.
No workstations => No small computer labs => Weak interest for the OS/Hardware from sudents & hobyists => Future decline of sales in servers.
Look at HP & all the other commercial Unix vendors - decline in server sales is almost directly related with workstation unavailability in the past ~5 years.
1% APY, No fees, Online Bank https://captl1.co/2uIErYq Don't let your $$$ sit in a no-interest acct.
Why go non-X86?
Well, gee, let's see what kind of viruses there are for PowerPC architecture now that Mac has gone Intel.
Uh... None?
If you're building a server farm, who cares about the architecture?
Now, having said that, I do agree with the comment that says there ought to be high-horsepower workstations available. Not all of us are Windoze Gamers. I work at a University and do a lot of SCF chemical simulations. That, my friends, takes guts. If I can't cram in additional CPU/GPU, it kind of leaves me out.
Don't take life too seriously; it isn't permanent.
While some might bash that last pick, the more platforms you have to work on, the more assumptions you can help expel from your code, in case future development goes in a direction you weren't expecting.
Just like all the 16/32 bit assumptions that broke during the migration to 32/64 bit systems, having mixed-architecture current generation hardware has innumerable benefits for ensuring cross platform compatibility and minimizing faulty assumptions or bad coding practices during development.
I work in HPC and have had access to some Power 8 machines, both in pre-production and release. The CPUs are amazing beasts, but it is just not enough to make them competitive over cheap and good enough (Intel/AMD) CPUs. On top of that, for highly FP dependent calculations, the SMT8 mode of the chip doesn't give any performance over SMT4 because there's only 4 floating point units per core. Still core per core, it's much faster than the Power7, sure.
Seeing the headline I almost skipped this one since IBM has such a tendency to build expectation and then under-deliver.
But since x86 is gone to Lenovo, I figured this one might be interesting. They might finally put out something I might need to know about - they might leverage their non-IBM-PC-encumbered mainboard designs to make something really compelling for disposable cloud computing and hire a few guys to make sure, say CentOS 7, is easy to deploy on it. I was reminded of the talk c. 1999 when IBM was going to setup Linux as an 'LPAR' (IIRC) and you could run 256 instances on one of their big-iron machines (this was when nobody was virtualizing anything and VMWare was still at Cornell).
I thought, "they might actually be coming out with a 4-U box with sixteen processors in it that a cloud provider could cost-justify vs. whitebox x86 pizza boxes and offer management advantages, or maybe a blade system that would make it easy to deploy a compute cluster with 96 processors on a shelf and a tuned-assembly library for HPC." IBM has the means to do all of those things and there's a tremendous market for them. Finally, without the x86 albatross, it's POWER's time to shine.
"2-socket system".
IBM POWER - disappointing the industry since 1989.
My God, it's Full of Source!
OUTSIDE_IP=$(dig +short my.ip @outsideip.net)
"We have to let go of this notion that for Apple to win, Microsoft has to lose."
This is more universally true than a lot of people are willing to let on.
They don't need to take on Intel and x86. They just need to make a profit on each system sold.
I seriously don't understand why the tech common wisdom doesn't understand this very basic concept of business logic.
Non impediti ratione cogitationus.
...mostly useful if you want many TBs of shared coherent memory, or have legacy needs.
x86 scales pretty high now (you can get 3 TB, 60 cores, SSDs from a commodity vendor), and when you scale out you have better interconnect options (Infiniband), and of course distributed systems software's come a ways, so bless you if you really need it.
Roll out a machine with 128 of these and tell prospective customers they can implement 20,000 virtual instances IN THE CLOUD.
Man I'd love to know if anyone from ibm POWER is reading this thread !!!
because the SPARC M7 looks great. It has lots of acceleration in silicon to speed up database queries and encryption.
Oracle are claiming that companies are switched back from Intel boxes to large engineered SPARC systems.
Bounty hunt for IBM: try to find the word 'Linux' on the Ubuntu web pages.
Historically Power has been a mess of IBM technology. Recently they're been supporting KVM and OpenStack, which is like a million times better than HMC etc. But it's still pretty bleeding edge, and I wouldn't expect paid clouds just yet. And depending on the pricing, it's probably never going to be as cheap as commodity x86, or ARM.
That 2 socket system is 12 cores per socket. Each core has 8 SMT threads so it looks like 96 logical processors to an OS. It's bigger than you think, but how those 96 threads compete with the latest 2 or 4 socket, 15 core Intel sockets that have 2 way hyperthreading remains to be seen in real world applications.
As for deployment, you can buy Linux only versions of these systems that run IPMI (rather than IBM's proprietary HMC protocol) and a custom Linux KVM install. So, it should be a mostly transparent drop-in for cloud providers, other than the need to boot / install Power linux images rather than x86.
This has huge legs if enterprise level software vendors complie everything for PPC (ahh like Oracle!?!?!). The hardware is great in that its monitored to the deepest levels. Low level checks confirm that the platform is stable for the OS.. Who has had to try to debug a low level Whitebox issue with a memory error, or even main-stream box with a spurious power supply issue? The Benefit of the SMT 2-4-8 will be interesting to head of when coupled with a low latency storage like SSD or flash arrays. Anyway, its great to have options! The cost value will be interesting when you skip the VMWare layer, and bundle in the OS. Should be a great platform if the ISV's pick it up. Can anyone comment on the completeness of PPC vs x86 OS distributions?
Time for a new Political party in the US (or two!) One is off the rails Other cant pony up a leader.
Now that IBM's let Lenovo bastardize their x86 platforms, that only leaves the stuff that no normal person could hope to afford - POWER.
Perhaps they could come up with some entry point that doesn't have EOL written all over it.
Twitter supports and protects racists - by smearing their critics with the "Hate Speech" label.
If you've not followed things, they've tried to drop anything below POWER4 for support. Perfectly fine POWER3-II's get arbitrary cuts.
Twitter supports and protects racists - by smearing their critics with the "Hate Speech" label.
If it goes down you have another $15k box or a cheaper on that can get by, just like you would with the $25k box when you don't want to wait a day for the IBM guy to turn up, three hours to teach the IBM guy about the IBM system and another day for parts to be flown in.
There are plenty of good reasons but it's not as cut and dried as the post above. The biggest reason is capability - if the cheap box does not handle the job adequately or the architecture/platform is what you need to run your stuff then the expensive option can be more viable in the long run.
The POWER8 is 2x faster than POWER7 in theory and POWER8 holds 1TB RAM per socket and has 12 cores. The new Intel Xeon E7 has up to 15 cores and holds up to 1.5TB RAM per socket, it is in general 1.8x faster than POWER7:
http://www.theregister.co.uk/2014/02/18/intel_releases_mission_critical_two_four_and_eight_socket_xeon_e7_v2_line/?page=3
That is disappointing. Sure, the Intel Xeon only scales to 8-sockets, and all Unix cpus (POWER, SPARC and Itanium) scale to 32-sockets or above (SPARC scales to 64 sockets or even more), but the POWER8 has less RAM than Intel. The Intel is almost as fast as POWER8 and holds more RAM.
The SPARC M6 cpu from yesteryear is faster than the POWER7 cpu, and holds several world records on 32-socket servers. The largest Intel servers have 8-sockets (unless you go up to clusters such as Linux SGI Altix and UV2000 servers, or ScaleMP servers which both has 10.000s of cores and 100TBs of RAM - just like a small supercomputer cluster).
The new SPARC M7 coming next year, has 32 cores and 8 threads each and holds up to 2TB RAM. The SPARC M7 cpu will be 3-4x faster than the SPARC M6 cpu. So the largest Oracle SPARC server next year will have 32 sockets, 1024 cores, 8192 threads and 64TB RAM. Of course, the largest Fujitsu SPARC server has 64 sockets and are more tailored for number crunching, with fewer very powerful threads:
http://www.theregister.co.uk/2014/08/07/fujitsu_takes_nextgen_hpc_chip_on_the_road/
The SPARC M7 server has many threads and focuses on throughput, business work with many users. Excellent for very large database configurations running Oracle. Warning: very technical information about the new SPARC M7 supercomputer on a chip:
http://www.enterprisetech.com/2014/08/13/oracle-cranks-cores-32-sparc-m7-chip/
The SPARC M7 will eat the POWER8 server's lunch. There are also speculations (by the die hard IBM fanatic Timothy Morgan Prickett) that IBM has problem with scalability on POWER8, so the largest POWER8 server might have only 16 sockets, thus it will have 16TB RAM too.
http://www.enterprisetech.com/2014/07/28/ibm-forging-bigger-power8-systems-adding-fpga-acceleration/
"...Depending on how many customers are hitting the performance ceiling on the Power 795, IBM could skip putting out a 32-socket Power8 machine and just got with the 16-socket machine with 16 TB of memory...."
We all know that IBM only does high margin business. The earlier POWER6 was 10x more expensive than x86 servers and several times faster. The POWER7 is 2x faster than x86 and 3x more expensive. The POWER8 is 10% faster than Intel Xeon and 2x more expensive. Margin is dropping on POWER servers. When margin goes down the drain, IBM will pull the plug. There is no point of developing servers when IBM does not earn money. And we all know that IBM is exiting hardware business and has sold of many hardware divisions (x86 servers, hard disks, printers, etc etc etc). Coincidentally, the roadmap for IBM POWER stops after POWER8. There is no talk about POWER9:
http://www.enterprisetech.com/2014/07/28/ibm-forging-bigger-power8-systems-adding-fpga-acceleration/
And coincidentally IBM have confirmed that AIX will be killed and replaced by Linux in the long run:
http://news.cnet.com/2100-1001-982512.html
Are we there now? There will be no POWER9, as POWER8 is only marginally faster than x86, and much slower than SPARC? IBM knows that POWER9 will be slower than x86 and must be cheaper -> low margin -> pull the plug?
The power systems are very impressive. I think the last time I used them was around the Power5. Still, very good hardware, excellent performance characteristics, ran Linux like a beast.