VPC 5.02 Update - Improved OS X Performance? (from the 3/6/2002 main news page)
(IMPORTANT: Before you run the 5.02 updater, open and shut down any "Saved State" sessions you have in VPC!)
I'm sure you're already aware of the Virtual PC 5.0.2 release. I just
wanted to let you know that I've installed it on my iMac G4 800, and it
makes a HUGE difference in OS X (10.1.3). Everything is faster and more
responsive. It finally is able to get more processor time when in the
foreground (I'm getting around 85% utilization), and is pretty friendly when
in the background. I'm running Win 2000 on it, and it's finally useable.
It took Connectix a couple of versions to get the OS X / Carbon stuff
figured out, but it looks like they've finally gotten it.
The update is available at http://www.connectix.com/support/vpcm_online.html. The v5.02 vital information file notes (among other things) this comment on using AppleScript to set the CPU usage (background example shown) :
I wonder if you can substitute "foreground" for "background" in the script to set the foreground cpu usage. (Past comments here noted some owners used renice to tweak performance, noted in this page on VPC 5 OS X vs OS 9 performance tests from last December.) I've downloaded it but not installed it yet on my PB G4. I'll see if it shows any noticeable improvements in OS X on that machine.
Here's a clip from the VPC 5.02 Readme file:
CHANGES IN VIRTUAL PC 5.0.2
+ Added significant performance enhancements for Virtual PC running in Mac OS X.
+ Improved performance of Virtual PC in Mac OS X with background applications. Virtual PC will receive more time when it's the foreground application.
+ Virtual PC now yields processor time when a guest OS is in the foreground but idle. (Requires the new Virtual PC Additions.)
+ Increased throttle time in Mac OS X to give more time to video.
+ Faster overall boot time for most OSes (when no other applications are running) in Mac OS X.
+ Greatly improved performance of 16-bit applications running in Windows 2000 in Mac OS X.
+ Improved speed of installing 32-bit operating systems (i.e. Windows 2000) in Mac OS X.
+ Fixed an issue between Virtual PC and sound initialization that caused Virtual PC to crash at launch.
+ Added fix for rare crash when saved states were restored on beige G3s.
+ Added fix to allow PPTP VPN over Shared Networking in Mac OS 9. VPN using Shared Networking in Mac OS X requires the user to be logged in as root. See the "Virtual PC Vital Information" document for more information.
+ Improved diagnostics added to gather more informative troubleshooting details.
+ Fix for printing using Virtual PC emulation to Canon printers in Mac OS X.
+ Added fixes and improvements for serial port emulation including full 9-pin support. This should allow greater compatibility with various serial devices (Mac OS X only).
+ Made serial port preferences window completely interactive. Also detects the attachment of any USB serial adapter while the "Settings..." window is open in Mac OS X. Virtual PC now detects the removal of serial port(s) and defaults back to "none" when this occurs. + Serial port will no longer disappear from the COM Settings panel when the window is opened and USB is enabled.
+ Automatically resume a USB device if packets are detected being sent to a suspended device. Allows the LinkSys WAP11 USB configuration utility to work.
+ Fixed USB connection issues in both Mac OS 9 and Mac OS X that caused crashes on plug-in and un-plug of a device.
+ Added scripting support to allow debugging of blue-screen events in Windows XP. + Fix for MatLab 6 (max [1 2] calculation) so it returns the correct result.
+ Fix for rounding results in CallAtlanta program.
+ Fixed an issue in Mac OS 9 that caused a crash when the hard drive and CD drive were spun down (slept) and a call was made to the drive from Virtual PC.
+ Fixed an issue with reading Joliet CD formats under Mac OS 9.
+ Changed behavior of incompatible or corrupted saved states so that the user is asked before the state is deleted.
+ Fixed an issue with Folder Sharing the Mac's free drive space is now correctly reported by Windows 9x OSes.
+ Allow multiple users in Mac OS X to run Virtual PC and share preferences and drive images. This will also allow the drive image and preferences to be moved into the Virtual PC application folder and placed on a removable drive for increased mobility.
+ New Virtual PC Additions
This page has reader comments/test results with Connectix's Virtual PC version 5. (Note that for OS X's dual processor support it's said the 2nd CPU only handles the video - which helps make up for the slower performance seen in OS X compared to OS 9 in the results below. Also note that only Windows2000, NT and XP professional have dual processor support natively, but I doubt that is used by VPC.)
Although benchmarks like Cinebench do emulate applications performance (Cinebench tests 3D rendering/graphics performance), I'd also like some real world applications test comparisons of VPC 5. If you run actual Applications tests comparing VPC 5 in OS 9 and OS X (or VPC 4 to VPC 5) let me know. (include your mac system details.)
Connectix Comments on OS X Performance:
Connectix has a forum thread at their site that answers the question:
"Q: Why is Virtual PC 5 slower on Mac OS X than on Mac OS 9?". OS X's increased overhead and better (pre-emptive) multi-tasking are cited as reasons but check the thread for the full comments. (Thanks to Gregory Kelly for the link.)
PC Benchmarks: Although I'd prefer real application test comparisons, here are some links to a few PC benchamarks.
(For a copy of the full readme file from VPC 5, click here.)
Virtual PC Performance Tip for OS X: (see comments/updates below.)
In reply to the Virtual PC 5 test results posted earlier today (below) a reader sent a suggestion which I have forwarded to those that sent results.
Try this - "sudo renice -1 VPC-PID". Get the PID with ps or top.
[using the terminal.-Mike]
This should increase the priority of VPC to one less than it normally is
The smaller the number the higher the priority.
I've asked those that sent OS X VPC results today to try this and report back on how it affected test scores.
[Note: A reader suggested using larger numbers than -1 should give much better results. (He noted using -16) ]
Here are some of the replies so far:
Thanks for the tip. I have tested it this morning but I don't see any boost.
This is my results:
with "sudo renice -1 VPC-PID"
Shading (C4D) -> 1.26
Shading (OpenGL) -> 0.75
Raytracing -> 1.49
with "sudo renice -5 VPC-PID"
Shading (C4D) -> 1.24
Shading (OpenGL) -> 0.81
Raytracing -> 1.42
I hope in a future update from Connectix that solve the problem.
Massimo De Carli
Latest Reader Reports: (most recent first)
Comments on VPC5 w/NT and Red Hat Linux
I've been running VPC5 for the last couple of days and have a few
observations on it.
1) As stated by Connectix, WinNT is very responsive. It feels faster than
Win98. The 'additions' installed automatically after booting my VPC4 drive
images with VPC5. The VPC4 additions for WinNT never worked quite right
for me. Both are quite usable though.
2) The main comparison between running VPC5 under both 9.2.2 and OS 10.1.1
that I've made was with Red Hat Linux 7.1. Under VPC4, I couldn't even
install Red Hat, while it installed perfectly under VPC5, including the
creation of the Linux emergency boot disk. This part also failed under
VPC4. I allowed Red Hat to partition the disk image automatically, not by
doing it manually with Disk Druid. Under OS 9.2.2 Linux is showing a CPU
clock speed of roughly 380mhz and 575 bogomips. Under OS 10.1.1 the
bogomips drops to roughly 380. Linux feels a bit sluggish under either
system though, taking a bit of a wait to launch applications but being
somewhat more responsive opening menus and such. Linux is using the
Video mode: 16bit at 1024 x 768 run full-screen vice in a window
All of this is using my B/W G3/450, 448MB RAM, PCI Radeon video card, 2940
and 2930 SCSI cards, and a Kingston KNE1000TX PCI ethernet card. I next
plan on testing it with my G3/300 at work, where I currently run WinNT
under VPC4 and OS 9.2.2.
Ok, I have a theory as to why the sudo renice -1 VPC-PID tips aren't
helping. I have a feeling people are just typing in "VPC-PID" when they
should be entering in a number. For example, after I located the VPC 5
application's ID in my terminal I entered in "sudo renice -16 411" for
the command and got a nice speed boost. It isn't VPC under OS 9, but
it's close to usable.
I used launching Pagemaker 6.5 as my test. My machine is a single
processor 400Mhz G3 with 368 MB of RAM, with 128 MB given over to the
virual PC app. I'm running OS X 10.1.2 and Mac OS 9.2.2.
Anyway, within MacOS X I launched up Virtual PC 5. I then launched up
Pagemaker 6.5 within the app and timed how long it took for the
pallettes to be viewable on screen. Default installation Pagemaker's
launh clocked in at 1 minute 17 seconds. I launched it again at got 1
minute 14. Ick. SLOooooow.
I then went to the terminal and used the grep command to view all the
process ID's. I located VPC 5 which was running as ID 411 and ran the
renice command listed above. I then went back to VPC 5 and relaunched
Pagemaker 6.5 (noting a marked speed up of the Start menus right away).
It launched in 42 seconds. A 30 second improvement. Better, but still
On a hunch I killed the classic environment and launched Pagemaker 6.5
again. 40 seconds this time. Not much of an improvement to be had from
For my final test I restarted into OS 9.2.2 and launched VPC5. I used
the same disk images and app, this is not a second installation.
Pagemaker 6.5 launched in 14 seconds. So a 3 fold speed increase!
Looks like Connectix has a ways to go yet on making this really viable
under OS X.
He later wrote
Just a followup on my report yesterday with the time tests of launching
PM65. Since yesterday I've checked out Connectix' own forums and
learned quite a bit about this slow emulation problem. It turns out
that the Host OS you're using inside of VPC can make quite an impact on
performance. Oddly enough it's the more modern OS' 32-bit versions
(Windows 2000, XP, etc.) which perform best under OS X! Older OS'
16-bit versions (95, 98, etc.) are more difficult to emulate under OS X.
This is the exact opposite under OS 9 in terms of performance!
The other thing that can kill the VPC performance is other apps hogging
in on it's OS cycles. Under OS 9 it's easy for VPC to essentially "lock
out" other apps from hogging the CPU. OS X they just can't do that,
hence the speed hit. Connectix reps claims in the forum they were
anticipating a 20%-30% average speed hit as oppossed to performance
under OS 9 due to this lack of "lock out" ability under OS X, but many
people are reporting more than that. Could it be the Host OS, or just
other things running in the background? Classic apps are an obvious CPU
hog, but since I've started using the terminal's 'top' command to
monitor CPU usage it's amazing how much of a hog something as innocuous
as iTunes can be! I was seeing CPU grabs of up to 40% at times!
Shutting down iTunes made a big difference in terms of the number of
cycles that VPC can have. iTunes running VPC never seemed to get above
40%. iTunes off I saw VPC edge into the 60-70% range.
I tried it here this morning. I launched into VPC5 and instead of using
my usual Windows 98 installation I instead booted into Windows 2000.
What a difference! The best I managed yesterday was 40 seconds to
launch Pagemaker 6.5. Today it was 12 seconds under Windows 2000!
That's equal to my performance under OS 9 using Windows 98!
Oddly enough the "renice" trick I had tried yesterday that made such a
big impact with Windows 98 didn't make any difference at all with
Windows 2000. This could be the result of NOT running iTunes and
therefore having more CPU cycles to play with, or it could be the
Windows 2000 being easiler for VPC to emulate. Time will tell.
Now just launching Pagemaker isn't the most intensive test in the world
but considering that I was getting 1:17 yesterday to launch the program
just these small changes I made make a world of difference for me. I'd
even go so far as to declare it usable now.
Just to recap I'm running a B&W 400 Mghz G3 with 368 Mb of RAM, 128 of
which is dedicated to VPC.
Beige G3 (upgraded) Report Comparing VPC4 to VPC5:
Here's my VPC 5 test:
Machine Specs: G3 (beige), 489 MHz, 640 MB RAM, VPC 5 installed on
internal 20Gig IDE.
Windows 95, monitor resolution set at 65536 colors with 1024 by 768
pixels; memory set at 256Meg RAM
Mac OS 9.2.2 (VPC5)
Shading (Cinema 4D) 2.03 CB
Shading (OpenGL) 1.18 CB - GL-Factor 0.58x
Raytracing (Single CPU) 2.02 CB
Mac OS 9.2.2 (VPC4)
Shading (Cinema 4D) 1.88 CB
Shading (OpenGL) 1.13 CB - GL-Factor 0.60x
Raytracing (Single CPU) 2.07 CB
Mac OSX (VPC5)
Shading (Cinema 4D) 1.28 CB
Shading (OpenGL) 0.86 CB - GL-Factor 0.67x
Raytracing (Single CPU) 1.47 CB
'Renicing' had no effect.
VPC 5 with NT:
I decided to do a new disk install since I am not convinced my
old Windows 98 C & D drives are optimized. I also decided
to be brave and try a new install of Windows NT Workstation.
I am using an download variation of the VPC 5 upgrade with
no additional OS packs.
I started with the disk set that I had and worked my way up to
Windows NT Workstation Service Pack 6a and IE version 6.
Since I have a fast ethernet connection this didn't take that long
given that the overall feel of VPC 5 running NT Workstation
(dare I say snappy) was much better than I was experiencing using
my VPC 4 Windows 98 disks. Unfortunately, I could not see the
old drive images so I will need to reinstall Windows 97. Below are my
results for CineBench 2000 in OS 9.2.2 and OS X running the same
configuration of VPC 5 installed under OX X.
Dual G4, 450 MHz, 1 Gig RAM, VPC 5 installed on internal 36Gig SCSI drive with
VPC 5 settings:
Windows NT Workstation Service Pack 6a (no virus protection software installed)
networking active through DHCP, monitor resolution set at 65536 colors with
800 by 600 pixels; memory set at 256Meg RAM
Mac OS 9.2.2
DAVE 3.1 running, file sharing off, Norton antivirus autoprotection engaged
Shading (Cinema 4D) 1.73 CD
Shading (OpenGL) 0.95 CB - GL-Factor 0.55x
Raytracing (Single CPU) 1.55 CB
Mac OS X 10.1.1
DAVE not running, file sharing off, no antivirus auto protection engaged
Shading (Cinema 4D) 1.02 CD
Shading (OpenGL) 0.64 CB - GL-Factor 0.63x
Raytracing (Single CPU) 0.74 CB
As I said above, the overall feel and response of VPC 5 is much better than
these numbers indicate and makes me feel much better about the software
than my initial impressions running my original VPC 4 disks.
(MP note: Only Windows 2000, NT and XP Pro (natively) support dual processors
(and VPC only uses the 2nd CPU for video I believe).
Dual G4/800 Owner Tests: (updated 12/10 with Cinebench scores)
This is on a Dual 800MHz G4 QuickSilver with Mac OS X 10.1.1. VPC was
running Win2k with all the patches and nothing else running.
Whetstone FPU 189 MWIPS
Dhrystone ALU 848 MDIPS
Benchmark index 217
Write 4.01 MB/sec Read 5.7 MB/sec
...here are the [Cinebench 2000] results:
Mac OS X 10.1.1:
Shading C4D: 1.76 CB
Shading OpenGL: 1.16 CB
Raytracing: 2.25 CB
Mac OS 9.2.2:
Shading C4D: 2.46 CB
Shading OpenGL: 1.94 CB
Raytracing: 3.03 CB
I thought that some benchmarks from FreshDiagnose on my PC laptop may put
the VPC 5 benchmarks into better perspective.
My machine is a Toshiba Satellite 1625, 475MHz K6-2 processor, 128 MB RAM,
WhetStone FPU 256 MWIPS
DhryStone ALU 1,191 MDIPS
FreshDiagnose also provides some system comparisons. The list the
following for a "Intel Celeron 700 MHz"
WhetStone FPU 315 MWIPS
DhryStone ALU 1,943 MDIPS
Interesting to note, the Floating Point performance of the 700 MHz Celeron
is only slightly faster than my K6-2 and only about twice as fast as the
VPC. Not bad for emulation, especially on a system to cooperatively
multitasks so well.
All in all, if these benchmarks are indicative of overall performance VPC
is doing pretty well. I'd be happy to get roughly Pentium 200MHz
performance on my Dual 450MHz. That's plenty for most Office apps, like
my ordering software.
As for the tip about renicing VPC, that should help. It seems most likely
that the great multi-tasking in OSX is hurting VPC performance. I did
renice the test drive but I didn't notice too much of a difference... I
didn't run benchmarks however. I'll see if I can bench the test drive
tomorrow at different run levels. I'd like to whip up an applescript to
automate the renicing process... I'll send you a copy when I get it done.
Virtual PC 5 vs V4 in OS 9:
OK Mike check this out. (G4 733, 1.2 GB ram)
Shading (Cinema 4D) 1.71 CB
Shading (Open GL) 1.04 CB
GL Factor 0.60x
Raytracing (Single CPU) 1.97 CB
Shading (Cinema 4D) 1.84 CB
Shading (Open GL) 1.43 CB
GL Factor 0.77x
Raytracing (Single CPU) 2.30 CB
Now look at this!
Shading (Cinema 4D) 2.16 CB
Shading (Open GL) 1.58 CB
GL Factor 0.73x
Raytracing (Single CPU) 2.76 CB
It looks like VPC 5 really works better in OS 9.2.2 than VPC 4
VPC 4 in OS 9 is better than VPC 5 [in OS X]
VPC 5 in OS 9 is best
Virtual PC v5 Tests with Dual G4 vs. Single G4 in OS X:
Tried VPC 5 on both a Titanium Powerbook 500 w/OS 9.2.2 and a Dual
G4/500. Both had 256 megs of RAM (160 devoted to the virtual machine,
which was Windows XP Professional). Neither had any apps running
except the finder, and it was running full screen.
Overall, pretty slow (what i expected). However, what surprised me
was that the TiBook performance seems better than the Dual G4/500- So
here, even having dual processors does not seem to overcome the speed
penalty of running VPC in OS X :(
I grabbed a benchmark utility called "FreshDiagnose" from
versiontracker (never heard of it, but i figured it would be fine for
relative testing). Sure enough, the Tibook scored very similar to and
in some cases outperformed the Dual G4/500 (CPU, "Multimedia",
(Higher numbers are better)
Tibook Whetstone FPU 155 MWIPS, Dhrystone ALU 569 MDIPS
Dual G4 Whetsone FPU 153 MWIPS, Dhrystone ALU 552 MDIPS
Tibook Benchmark index 173
Dual G4 Benchmark index 178
Tibook Write 1.67 MB/sec Read 6.23 MB/sec
Dual G4 Write 3.78 MB/sec Read 5.55 MB/sec
[Notebook drives are slower than desktop models-Mike]
I'm pleased Connectix has brought Virtual PC to OS X- it is a shame
that it did not result in larger (or any) gains over OS 9 in
performance- even with an extra CPU.
I welcome other Dual G4 owner test results with applications also. And if anyone has tested VPC 4 vs VPC 5 in OS 9.x. Send me your test results and system details.
Virtual PC v5 Tests on PowerBook G4/667 in OS 9 and OS X:
Hi! I have my Powerbook 667 with 512 Mb with MacOS 9.2.1 (Italian) and MacOS
X 10.1.1. Well, I have tested VPC5 with Cinebench 2000 for windows:
Shading (Cinema4D) = 2.12
Shading (OpenGL) = 1.24
Raytracing = 2.41
MacOS X 10.1.1
Shading (Cinema4D) = 1.29
Shading (OpenGL) = 0.83
Raytracing = 1.38
Good performance on old MacOS. Slooow on new MacOS.
Massimo De Carli
I noted yesterday I was disappointed in the OS X test drive performance compared to OS 9 running VPC 4. I'd hoped v5 might address this. Perhaps for dual G4 owners the support version 5 has for dual processors will help. (It uses the 2nd CPU for video tasks I believe.)
If any readers test Virtual PC5 with a dual G4, send me the results of any tests you run comparing OS 9 to OS X.
Another VPC 5 owner wrote with comments:
"Mike, I upgraded to VPC5 today and can now login to my Novell server
using OSX 10.1.1 and VPC 5. I could not login using VPC test drive for
OSX. I logged in fine under OS 9.2.1. VPC 5 is not as fast under OSX
as VPC 4 under OS 9.2.1.
I use it for accessing my accounting system at
my office, a vehicle tracking system, a pricing program, and estimating
software. The vehicle tracking worked under VPC test drive because it
uses IP to communicate with it's server. The other 3 programs all use
Novell 5.1 and IPX to communicate with the Novell server. I had to turn
on something called "Virtual Switch" in VPC 5 to make the Novell server
work with VPC 5. I hope they can speed it up to where VPC 4 under OS
9.2.1 was because it was great.
OSX 10.1.1, G4 733, 1.2 GB ram
Notes on Vitual PC and Pre-G3 Macs/CPU Upgrades:
A reader sent a reminder that Connectix does not officially support Macs other than G3 or G4 models (Mac that shipped with a G3 or G4 CPU) per their latest FAQ (.PDF file).
VPC has always been very sensitive with upgrades as far as
speeds (CPU overclocking and even cache speeds). In the past, it has been one of the best tests for marginal settings/speeds with CPU upgrades. (I ran VPC 2 and 3 with several CPU upgrades in older macs - it would often lock up at startup if cache speeds or CPU speeds set too high, even though other apps may have ran OK.
The Misc. articles page has several VPC articles on previous versions, including
the full readme file from v4 and v5, notes on 3d support, performance tests, etc.
I hope to get a copy to test here.
Even VPC 4 may or may not work with CPU upgraded older Macs. One Bluechip owner noted it doesn't on his upgraded Wallstreet and that Connectix offered to refund his money.
I think they officially don't support upgrades just to avoid the
tech support costs since there's an infinite number of variables
with older macs and cpu upgrades literally.
And note in the past here there were cases where upgrade owners fixed issues with VPC by just switching the cache control software to another
brand (I think some sonnet owners had this problem that was
solved by using PL or xlr8's cache control).