Catch up on stories from the past week (and beyond) at the Slashdot story archive

 



Forgot your password?
typodupeerror
Businesses Cloud Google

Google's Compute Engine Now Offers Machines With Up To 64 CPU Cores, 416GB of RAM (techcrunch.com) 74

An anonymous reader shares a TechCrunch report: Google is doubling the maximum number of CPU cores developers can use with a single virtual machine on its Compute Engine service from 32 to 64. These high-power machines are now available in beta across all of Google's standard configurations and as custom machine types, which allow you to select exactly how many cores and memory you want. If you opt to use 64 cores in Google's range of high-memory machine types, you'll also get access to 416GB of RAM. That's also twice as much memory as Compute Engine previously offered for a single machine and enough for running most memory-intensive applications, including high-end in-memory databases. Running your apps on this high-memory machine will set you back $3.7888 per hour (though you do get all of Google's usual sustained-use discounts if you run it for longer, too).
This discussion has been archived. No new comments can be posted.

Google's Compute Engine Now Offers Machines With Up To 64 CPU Cores, 416GB of RAM

Comments Filter:
  • by SuperKendall ( 25149 ) on Thursday March 09, 2017 @07:29PM (#54009853)

    I'll have to check out this Google offering too, but since this is pretty relevant topic to stuff I've been doing recently - another online option is FloydHub [floydhub.com]. They are cheaper than AWS, and I like them because they bill per second of use, and you can choose to use GPU or CPU services (the CPU seems kind of slow). They also have really nice support for python and Jupyter notebooks running on the server, along with the ability to upload large datasets (for machine learning jobs) and an API for programatic access to the computational services.

    It's mainly targeted to deep learning stuff so if you need GPU for other things it may not be as useful. But if you are playing with Deep Learning this kind of service makes training models way more feasible for those of us who normally do not buy really expensive GPU's. It'a also nice to do what you can to support companies trying to go up against Amazon or Google, and the FloydHub people have been very responsive to questions I have asked.

    • by lgw ( 121541 )

      If you want a big cloud instance, AWS has a 128 core 1,952 GB instance type (plus a smaller version the size of Google's). The Spot price looks like $3/hour as I type this, cheap to run a few benchmarks.

      • Yeah, was going to post something along these lines.

        Woohoo, Google now includes instances up to half of what you can do on AWS! And, without the flexibility of getting not only the 64 cores and almost 500GB of RAM, but also getting EIGHT dedicated 1900GB NVMe SSD you get with an i3.16xlarge!

  • by FairAndUnbalanced ( 959108 ) on Thursday March 09, 2017 @07:41PM (#54009897) Homepage
    Slowwwwly closing in on Amazon's X1 instance [amazon.com] which offers 1952 GiB RAM and 128 vCPUs at ~$13.338 per hour (dynamic pricing). Google may catch up in a year or two.....
  • Thanks for the fake news article which is actually an advertisement. Slashdot has plunged so far.

    • Slashdot has always published press release stories, because that's who's going to submit the most... but they aren't smart enough to get paid for them.

  • Dammit! (Score:5, Funny)

    by rsilvergun ( 571051 ) on Thursday March 09, 2017 @08:02PM (#54009991)
    I need 417 GB of ram and 65 cores. Oh well, back to the drawing board.
  • Really 64 cores? (Score:4, Interesting)

    by Anonymous Coward on Thursday March 09, 2017 @08:13PM (#54010047)

    Are these actual CPU cores, or just hyper-threads like with Amazon's AWS? If these are still in Google's "n1" class then by their own documentation they are indeed hyper-threaded.

    Hyper-threaded virtual cores, while nice for desktop i7s, are nearly useless for large-scale compute jobs using the likes of ACML, MKL or MPI.

    If these really are hyper-threads rather than physical cores then you're only going to get 32 real threads of compute performance and should pay for such.

  • One limitation of "the cloud" (also called "other peoples' servers") for many HPC applications is the data transfer costs. Transfering data in is cheap or free, but getting your data out again is anything but. Even if the cpu-hours would be cheap enough, it's usually cost-prohibitive to transfer a few tens of gigabytes of results out of the server and back home for each job.

    • by nasch ( 598556 )

      Even if the cpu-hours would be cheap enough, it's usually cost-prohibitive to transfer a few tens of gigabytes of results out of the server and back home for each job.

      Data Transfer OUT From Amazon EC2 To Internet
      First 1 GB / month $0.00 per GB
      Up to 10 TB / month $0.09 per GB
      Next 40 TB / month $0.085 per GB

      And so on, getting cheaper per GB from there. So if you're talking 50 GB per day, that would be $135/month. Peanuts for anything bigger than a mom and pop shop.

      • It's even cheaper if you are able to use AWS DirectConnect. $0.30/hour for 1GbE plus whatever it costs to get a circuit through a peering provider if you don't already have presence in a facility peered with AWS.

        My company was already in a peered facility, so it was just a matter of stringing a fiber line between their edge router and ours, and setting up BGP.

        $225/mo to move as much data as we please back and forth between our data center and our VPC.

    • If you're doing it in the cloud there's no reason to pull the data out of the cloud.

If A = B and B = C, then A = C, except where void or prohibited by law. -- Roy Santoro

Working...