My Problem With Intel’s Haswell Processors
This week, Intel formally announced its Haswell processors. In the mobile world, this is a big deal because unlike the previous generation of processors (Ivy Bridge), this new one promises things like better battery life and better graphics. I’m all for better battery life. Part of that is they had to redesign the ultra low voltage (ULV) chips that go in many computers. They have a lower power consumption than Ivy Bridge. By the numbers I’ve seen in preliminary tests, Intel really did improve battery life.
My excitement, however, is short lived. As usual, I’m enticed by new and shiny laptops – most notably the upcoming Vaio Pro 11 or Vaio Pro 13. The Pro 11 reminds me of a cross between my old Vaio G, the older Vaio T (not the new cheapie one), and maybe a bit of the X505 thrown in for good measure. Why is my enthusiasm dampered? Virtualization.
Here’s why: one of the things I love being able to do is run Hyper-V (and sometimes ESX) under VMware Workstation. I use Hyper-V for most of my demos, but it’s easier to show some things Hyper-V related under VMware since I can virtualize the hypervisor and have nested virtual machines. Besides the standard virtualization (VT-x in the Intel world), what allows you to run a hypervisor and nested 64-bit guests is the Intel VT-x with Extended Page Tables (EPT) feature of the processor. EPT = second level address translation (SLAT) for you geeky types. This requirement is documented by VMware here.
Cue the sad trombone. According to Intel’s whiz-bang Virtualization Technology web page, with the new Haswell chips, only some of the desktop processors got SLAT support; no mobile processor did. Don’t believe me? Click here.
This is a real setback for those of us who do complex demos from time to time like nested VMs. Remoting back home is not always an option. I’m going to have to seriously consider getting another backup Ivy Bridge machine (even if it’s a small form backup like the NUC) because outside of video, I won’t easily be able to demo some things anymore starting with Haswell.
I’m also bemoaning the fate of things like ethernet ports and VGA, but those can be solved with adapters. You can’t add a feature to a processor.
My woes here really fall into 1% of the user base for a mobile processor. I realize this. But if you’re like me, this is a pause before you get all giddy over Haswell.
EDIT: See my follow up post http://www.sqlha.com/2013/06/13/haswell-and-ept-update/. Intel posted the wrong info initially.
From Engadget review it says the cpu in a Sony VAIO Pro 11 is 1.8GHz Core i7-4500U, Intel HD 4400
from Intel’s site for the cpu it says it has VT-x and EPT
http://ark.intel.com/products/75460/Intel-Core-i7-4500U-Processor-4M-Cache-up-to-3_00-GHz
unless I matched up wrong or Intel or Engadget are wrong on their specs
i7-4500U is a ULV processor as I said. Both the Pro 11 and 13 are configurable for processors. Neither can have the i7-4650U; that is only in the Duo 13 right now.
As for the processor, when I first looked, if you looked at the link under Click Here, it said NO for EPT. I never said it didn’t have VT-x; so Intel changed the page from the first time it was up there.
This info seems to be wrong – eg see here: http://ark.intel.com/products/75114/
So the Haswell chips do support EPT (SLAT).
See the other comment – Intel changed the page. I should have taken a screen shot of the original listing that had no.
They did change the page, you’re the second person I’ve seen that checked this out and saw the original page and was disappointed. Thanks for the research.
Just unwrapped my new shiny Vaio Pro 13″ i7 less than 1 hour ago. Really I won’t be able to run centos 64bit guests on VirtualBox? That’s terrible!
Can someone confirm this?
ok it just needed to be activated in the bios settings! All right!
After all the corrections in the comments this page has become quite confusing, potentially misleading… Is the article itself accurate, or was it all a rant around some typos that have now been corrected?
I did a follow up post. I had an exchange with a guy from Intel on Twitter and he confirmed that initially what I posted and saw was up there, but incorrect. Everything has been corrected.