-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 06/11/2017 01:09 PM, Johnysecured88 wrote:
The problem is there are no heavy duty apps for the power arch. Photoshop, Premiere, All Adobe apps, all Autocad/3ds max, the list goes on an on.
Respectfully, if you are using those applications, you have far larger issues to worry about than whether the underlying system can be owner controlled -- you have zero control over the application layer and (probably) the kernel and firmware due to the increasing DRM requirements for those applications.
That being said, as POWER systems become more widespread I expect ports of big name software to become available; at least software that already had a Linux port. For others, the best course of action is really to somehow get development of the competing open source solution off the ground and up to feature parity with the commercial offering(s) -- by the time you get a vendor's attention with a multi-million dollar proposition, you could just directly fund development of one of the open source packages and bring it up to a high degree of polish.
I myself am building an opteron server build but I also plan
on taking you up on the upcoming offering.
The question is, what options are there for running x86 on power9? If there were a good performing emulation layer it would be great. Ive seen the qemu demo and it doesnt seem to be on par with normal virtualizations. Ive heard of hardware assisted emulation but I dont believe power9 chips will have that?
POWER has had hardware virtualization extensions since at *least* POWER7, and IIRC significantly before that. However, they help POWER run on POWER, and do not directly help x86 run on POWER.
The biggest issue with QEMU cross-arch emulation for heavy-duty apps is really the fact that there is no direct mapping of vector instructions. If someone could invest the $$$ needed to make that happen, x86 apps could run with only a several time slowdown versus the hundredfold or more slowdown due to guest vector instructions being serialised on the host platform. There are a few demos from some Chinese researchers exploring this idea (x86-->ARM), and the results are very impressive.
Of course, if you start relying on this for anything other than academic purposes you *might* run into the same legal problem that prevents the manufacture of non-Intel and non-AMD x86 chips -- you'd be using the patented, copyrighted x86 instruction set without the express permission of Intel. The reason I say "might" is that if your x86 apps are old enough, or support an old enough version of the x86 instruction set, you might be able to use a modified version of QEMU that avoids any patent infringement and still have your x86 application(s) run.
- -- Timothy Pearson Raptor Engineering +1 (415) 727-8645 (direct line) +1 (512) 690-0200 (switchboard) https://www.raptorengineering.com