Vista with AMD. XP with Intel. Huge speed difference.

[Mike Doty]

Should XP running server software on a 2.80 GHZ Pentium 4
run 3 times faster than the same thing running on a
AMD 64x2 Dual Core 6000+ 3.00 GHZ using Vista Home Premium
Service Pack 2?

I've turned off every service I can think of and virus checking.
Is Vista slow or the AMD much slower? Is there some
configurations to look at with Vista. These are both 32-bit.

http://cpuinfo.visualware.com/

Vista machine with AMD reports 3013.2 MHz 1094 MB/sec transfer rate.
XP machine with Intel reports 2799.6 MHz 1007MB/sec transfer rate.

Just made application single threaded and AMD machine running Vista is still 3-times slower than Intel with XP.

[Michael Mattias]

The same code?

The same code will execute the instructions at a speed directly proportionate to the CPU speed... laws of physics and all that.

But "getting your turn to execute those instructions" can vary wildly (yes, wildly!) depending on operating system, current load (including number and type of service processes), priorities of current threads and processes, number of CPUs, operating system, page file, caching and other system configuration options... and on and on and on.

I just don't know you can make any kind of general rule regarding "this operating system versus that operating system" at all.

MCM

[Mike Doty]

Yes, the same code I wrote.
300% slower on the AMD machine using Vista over Intel using XP.
Is there anything in Vista that might make such a difference?
1-user with both machines showing about 5% CPU before
running my code and about 10% running it. Same network cards.


Sending or receiving takes 3 times longer when running on the Vista machine.

[Michael Mattias]

Well, I'm sure the Vista O/S does more error checking... every other new version of Windows has done more than the previous version.

Plus, I would not be surprised if there were....
A. Additional "security" checks being run more often
B. System settings which differ from prior O/S defaults.
C. More code to support new features... if nothing else, the equivalent of an additional "Select case" for each function to 'do something different if this new feature is enabled."

But I would think still the greatest additional factor would be the behavior of the 'normally running' system services... probably sucking up more resources than they did under a prior O/S, resulting in your application getting a lower percentage of the total CPU cycles available.

You know what might be interesting? a PROFILE of the same application running on both systems. I'm not sure it WILL be interesting, but until ya runs it, you just don't know. But it might tell you that "function X requires a greater percentage of the time under Vista than under XP ".. which would tell you it is doing MORE than it used to.

The PROFILE is probably a good idea anyway: it sounds like you are 'not happy' with the performance under Vista, and the PROFILE is where you would have to start the optimization journey anyway.

MCM

[Chris Boss]

One area to check first is your use of SLEEP in any worker threads.

Are you using Threads ?

Vista handles secondary thread timing differently than does XP.
I don't know the exact details of why or the difference but I had an experience with some sample code using threads.

I had an app with two worker threads.
Each thread calculated a new frame for a simple animation and then sent a message to the primary thread to redraw the next frame.

I used SLEEP to control the timing of the threads, since I wanted one animation to run faster than the other. It worked fine on XP.

When I ran it on Vista, one thread almost never executed (just once in awhile). Vista obviously handled the timing of the threads differently and the use of SLEEP, while fine on XP, caused serious problems on Vista. By removing the sleep command, both thread ran equally, but I lost the ability to slow down one thread liked I wanted to.

When SLEEP is used in threads it can cause problems when the operating system does context switches between threads. Vista is obviously different than XP in how it handles this.

[Mike Doty]

Thanks much for that suggestion. I actually removed a SLEEP statement.

I just noticed that transmitting just one 39-megabyte file takes about 4 seconds on either box.

I've been testing new code where I send/receive multiple files that are read from a DDT textbox.

At least this is getting narrowed down to the bottleneck.
It might be the TCP OPEN or TCP CLOSE.
It might be the getting of the file name from the DDT textbox.
It might be the time to create each file on disk is much slower with Vista.
Maybe this is a disk caching issue.
I've been transmitting about 40 files that comprise about 2.6 megabytes that is much slower on the Vista box than the XP box.

[John Petty]

The same code?

The same code will execute the instructions at a speed directly proportionate to the CPU speed... laws of physics and all that.
MCM

Unfortunately that statement is totally incorrect in a modern processor. Comparing different brands by clock speed is fairly meaningless. Both use different cache reading methods, pipeline lengths, predictave branch lengths, numbers of out of order processors both for normal and floating point etc etc.
The differences can make speed differences quite large though I wouldn't anything like the amount quoted so I guess most of it is Vista vs XP.

[Mike Doty]

CONTROL SET TEXT dramatically slow down performance?

CONTROL SET TEXT might dramatically decrease performance.

Look at the difference of 11859 ticks calling DisplayEvent on Vista server
compared to 1903 ticks without using the CALL DisplayEvent on the XP server.

This test called DisplayEvent 192 times uploading 40 files.
Vista (from this test) is almost 400% slower using CONTROL SET TEXT than XP.

Transmit 40 files to 32-bit Vista server
With CALL DisplayEvent 11859 ticks
Without CALL DisplayEvent 3234 ticks

Transmit 40 files to XP PRO SP3 server
With CALL DisplayEvent 3011 ticks
Without CALL DisplayEvent 1903 ticks

SUB DisplayEvent(BYVAL buffer AS STRING)
  gHits = gHits + 1
  IF gHits > 2000000000 THEN
    gHits = 1
  END IF
  CONTROL SET TEXT gDlgMain, %labAction, TIME$ + "  " + Buffer
  CONTROL SET TEXT gDlgMain, %labActive, FORMAT$(gActive)
  CONTROL SET TEXT gDlgMain, %labHits, FORMAT$(gHits)
END SUB

I have submitted this to support.

[Scott Slater]

XP is much faster that Vista (when compared on the same machine). The good news is that Windows 7 is actually faster than XP. MS completely reworked the kernel and I am definitely impressed with Windows 7. I have never said that about any of the previous Windows OS's... I did side-by-side comparisons of all 3 OSs on a HP NetBook Mini as well as on several desktops and laptops.

So I would agree that it is more about the OSs in your case as well.

[Kev Peel]

Try disabling Aero, or better, POST a custom message to update the screen.

[Mike Doty]

Thanks, guys.
I've optimized everything on Vista.
I like the POST suggestion.

[Scott Turchin]

I'm anxious to see Windows 7 myself, have heard good about it.
I will probably update my laptop first as it's on Vista.

There's no forgiving with AMD's, if you lose a fan or it slows down, things heat up......and when they heat up they slow down....

Not to slam AMD, they just run too dang hot...my mom's laptop smoked 2 weeks after warranty expiration, heat sink foam they put on just smoldered and took out the video card, bad design. (HP laptop too, was a nice unit)....
Have probably fixed 7 computers in the past year with heat related issues, dirty fans etc.

Always AMD's tho, and I know Intels have to be cool too but they tend to be more forgiving.

[Brice Manuel]

Writing, copying, moving files on Vista is slower than XP. At least it is not as bad as when Vista first came out. Windows 7 is also slower than XP when it comes to writing, copying, moving files.

Vista (from this test) is almost 400% slower using CONTROL SET TEXT than XP.

I do not think this is a bug with PB or with your program. It will be fine in XP. Under Vista, it is slower because the GUI is hardware accelerated. Using text commands is a major no-no in DX based 3D games and a great bottle-neck. Since the GUI in Vista is using DX based 3D, this would be my guess for the problem.

In DX based 3D games the solution is to use old-fashoined bitmap fonts. That is not a solution for apps, but the speed hit from text commands really isn't that big of an issue for apps, either. It is not like your app needs to be running at 200fps and moving 500+ sprites around with no tearing