Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
It also looks like so far to get rpms to install, one has to:

Code:
export PATH=/i386/home:/i386/usr/local/bin:/i386/usr/local/sbin:/i386/usr/bin:/i386/bin:/i386/sbin:/i386/usr/sbin
export $LD_LIBRARY_PATH=/i386/lib/:/i386/usr/lib/
/bin/sudo /opt/powervm-lx86/bin/powervm-lx86 /i386/bin/rpm -i <rpm> --dbpath /i386/var/lib/rpm/Packages
so far to have the right environment variables, while I try to figure out a way to generate the runx86 script properly
 
So this is the output of Geekbench 2.2.7 before it crashes with PowerVM-LX86, doesn't look too stellar...
Some basic PowerVM-LX86 benchmarks, side-by-side (left PowerVM-LX86, right native ppc64) with bogomips from


I compiled it under i586 OpenSuse 11.2 to have an SLES-11 compatible glibc (glibc-2.11.1-0.17.4 for SLES11 and glibc-2.10.1-10.4 for OpenSUSE 11.2), as compiling it under Tumbleweed live i586 (20230106) results in an incompatible binary since it requires glibc 2.34. Cross-compiling (not possible under Tumbleweed ppc64, i586 cross-compiling is not supported anymore it seems at least under ppc64) under debian ppc64 or ubuntu x86-64 also resulted in incompatible binaries. The only difference in the compilation was that I had to use
Code:
-std=gnu99
under i586 OpenSuse 11.2.

The bottom line is that the performance looks very weak, 5.5% efficiency at best.

Geekbench2 seemed to indicate slightly better performance, at around 15% comparing with native osx ppc64 benchmarks, e.g.


I have noticed that under Tumbleweed generally speaking (both ppc64 and i586), cpu performance is generally weaker than under debian (notably via hardinfo) so I've copied all the PowerVM-LX86 directories (/opt/powervm-lx86 and /opt/at05) to debian ppc64, I will see if I can get it to work there and if it's any better.

It could be that powervm-lx86 does somehow make use of the MSR endianness switch bit under POWER5/6/7 (absent on the G5) but that there is a fallback mode for the G5 albeit much less efficient as there could be an additional byte-swapping step in the dynamic instruction translation, but maybe not.
 

Attachments

  • Screenshot from 2023-01-08 23-36-36.png
    Screenshot from 2023-01-08 23-36-36.png
    288.3 KB · Views: 78
Last edited:
Some basic PowerVM-LX86 benchmarks, side-by-side (left PowerVM-LX86, right native ppc64) with bogomips from


I compiled it under i586 OpenSuse 11.2 to have an SLES-11 compatible glibc (glibc-2.11.1-0.17.4 for SLES11 and glibc-2.10.1-10.4 for OpenSUSE 11.2), as compiling it under Tumbleweed live i586 (20230106) results in an incompatible binary since it requires glibc 2.34. Cross-compiling (not possible under Tumbleweed ppc64, i586 cross-compiling is not supported anymore it seems at least under ppc64) under debian ppc64 or ubuntu x86-64 also resulted in incompatible binaries. The only difference in the compilation was that I had to use
Code:
-std=gnu99
under i586 OpenSuse 11.2.

The bottom line is that the performance looks very weak, 5.5% efficiency at best.

Geekbench2 seemed to indicate slightly better performance, at around 15% comparing with native osx ppc64 benchmarks, e.g.


I have noticed that under Tumbleweed generally speaking (both ppc64 and i586), cpu performance is generally weaker than under debian (notably via hardinfo) so I've copied all the PowerVM-LX86 directories (/opt/powervm-lx86 and /opt/at05) to debian ppc64, I will see if I can get it to work there and if it's any better.

It could be that powervm-lx86 does somehow make use of the MSR endianness switch bit under POWER5/6/7 (absent on the G5) but that there is a fallback mode for the G5 albeit much less efficient as there could be an additional byte-swapping step in the dynamic instruction translation, but maybe not.
Strangely enough, if those low PowerVM-LX86 and native bogomips weren't bad enough, I noticed that on OpenSUSE Tumbleweed (i586 at least) that even for a Core2 Quad 9650 I get bogomips in the 900 range which should be much higher (e.g. on Ubuntu x86-64 / 2*Xeon X5365 I get scores in the 4000 bogomips range) so it could really be an OpenSUSE issue. I need to try Ubuntu or the x86-64 version of OpenSUSE Tumbleweed on the Q9650.

Re PowerVM-LX86 on Debian merely copying the directories doesn't seem to be enough for powervm-lx86-daemon to launch (although it could be a trivial permissions issue, it complains about permissions in /var/opt/powervm-lx86/daemon, which I did fix but to no avail). However, I was able to start the installer by again spoofing it into thinking it's installing on SLES11, and since my Debian distro has rpm installed, it might work. I need to rebuild the powervm-lx86 rpm to adapt it to Debian packages or apt-install the required deps and remove their requirement from the rpm, let's see how that goes

Cheers,
 
Strangely enough, if those low PowerVM-LX86 and native bogomips weren't bad enough, I noticed that on OpenSUSE Tumbleweed (i586 at least) that even for a Core2 Quad 9650 I get bogomips in the 900 range which should be much higher (e.g. on Ubuntu x86-64 / 2*Xeon X5365 I get scores in the 4000 bogomips range) so it could really be an OpenSUSE issue. I need to try Ubuntu or the x86-64 version of OpenSUSE Tumbleweed on the Q9650.

Re PowerVM-LX86 on Debian merely copying the directories doesn't seem to be enough for powervm-lx86-daemon to launch (although it could be a trivial permissions issue, it complains about permissions in /var/opt/powervm-lx86/daemon, which I did fix but to no avail). However, I was able to start the installer by again spoofing it into thinking it's installing on SLES11, and since my Debian distro has rpm installed, it might work. I need to rebuild the powervm-lx86 rpm to adapt it to Debian packages or apt-install the required deps and remove their requirement from the rpm, let's see how that goes

Cheers,
So on Ubuntu 20.04 (x86-64) I get bogomips in the 1900-2000 range for the Core2 Quad 9650, as opposed to 900-1000 under OpenSUSE Tumbleweed (i586). I don't know if this is due to OpenSUSE Tumbleweed per se; I will try it on OpenSUSE Tumbleweed x86-64 to exclude any differential performance between 32- and 64-bit.

Another note on hardinfo, after careful examination, the PM G5 Quad (running Debian ppc64 this time; performance is clearly lower under OpenSUSE Tumbleweed ppc64) outperforms the MP2,1 (high-end, 8x3GHz X5365 version) on Fibronacci, Cryptohash and N-Queens, hands down; for FPU raytracing, Zlib and GPU drawing (Radeon HD 5770) performance is comparable (Radeon HD5870 for the MP2,1). The performance for Fibronacci, Cryptohash and N-Queens on the G5 is incredibly comparable to that of the MP5,1 (8x2.4GHz version). For Blowfish, FFT the MP2,1 and MP5,1 win hands down though.

So our G5 boxes are still quite competitive for some specialist tasks :cool:
 
Last edited:
  • Like
Reactions: Tratkazir_the_1st
Strangely enough, if those low PowerVM-LX86 and native bogomips weren't bad enough, I noticed that on OpenSUSE Tumbleweed (i586 at least) that even for a Core2 Quad 9650 I get bogomips in the 900 range which should be much higher (e.g. on Ubuntu x86-64 / 2*Xeon X5365 I get scores in the 4000 bogomips range) so it could really be an OpenSUSE issue. I need to try Ubuntu or the x86-64 version of OpenSUSE Tumbleweed on the Q9650.

Re PowerVM-LX86 on Debian merely copying the directories doesn't seem to be enough for powervm-lx86-daemon to launch (although it could be a trivial permissions issue, it complains about permissions in /var/opt/powervm-lx86/daemon, which I did fix but to no avail). However, I was able to start the installer by again spoofing it into thinking it's installing on SLES11, and since my Debian distro has rpm installed, it might work. I need to rebuild the powervm-lx86 rpm to adapt it to Debian packages or apt-install the required deps and remove their requirement from the rpm, let's see how that goes

Cheers,
Turns out that the lower performance of bogomips under i586 Tumbleweed boils down to a mix of both 64- vs 32-bit performance AND Tumbleweed generally resulting in lower CPU performance.

Tumbleweed x86-64 yields 1400 bogomips for my Core2 Quad Q9650 3Ghz (against 900 under i586, same build no), so quite a significant performance drop between x86-64 and i586 BUT also between x86-64 Tumbleweed and x86-64 Ubuntu (1900-2000 bogomips).

In any case, the latter explains the lower scores yielded by hardinfo under ppc64 Tumbleweed vs ppc64 Debian.

So I will definitely be delving into the Debian avenue for PowerVM-LX86.

PS Oddly enough, when running the 32-bit version of bogomips under Tumbleweed x86-64 with my Core2 Quad Q9650 3Ghz, bogomips are in the 550-600 range, so quite a loss of performance with glibc2 32-bit, so maybe the loss of performance under PowerVM-LX86 is also explained by translating 32-bit i586 to ppc64? In that case it would make sense to get a hold of the 32-bit version (QuickTransit) that was infamously used on the PowerBook G4 to run x86 Quake2 at near-native speed. I would love to hear more details about it (what distro, which version and which x86 world in /i386 etc).

Because so far on the G5, we are way off the 80% native performance figure taunted by IBM and obviously of that offered by Rosetta!
 
Last edited:
  • Like
Reactions: dextructor
Going back to OpenSUSE Tumbleweed on the G5, it is possible to install libreoffice 6 ppc64 from Fedora 28 by adding the following repository:


It will complain about some perl dependency but you can either remove the offending package or downgrade it as I did, works well.

I did try to port the libreoffice 7 ppc64 debian packages to rpm, which does work, but there are too many differences in OS / directory structure /packages between Debian and OpenSUSE for it to install successfully. I guess the only way to get libreoffice 7 for opensuse would be to compile it from source, but I think that there are big-endian patches that were used for Debian so we would need those.

Otherwise, LibreOffice 6 is also not bad at all and fully functional!
 

Attachments

  • IMG_20211202_225803.jpg
    IMG_20211202_225803.jpg
    544.3 KB · Views: 91
I did try to port the libreoffice 7 ppc64 debian packages to rpm, which does work, but there are too many differences in OS / directory structure /packages between Debian and OpenSUSE for it to install successfully. I guess the only way to get libreoffice 7 for opensuse would be to compile it from source, but I think that there are big-endian patches that were used for Debian so we would need those.
I don't know if this information serves you, but in Void (32 and 64) you can look at the package status they have 7.4.2.3_2. I installed on my machines, they both have good SSD's and takes less than 30s to load and rendering takes 8s, but after that it's really good.

As for patches, I think that the developer has submitted it upstream, and it was merged. But if for whatever reason it didn't he's very polite and friendly, I never have any trouble contacting him. Maybe he can help

Good luck with you OpenSUSE adventures
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.