CPU benchmark Vista x86 versus Vista x64?

[basilisk561]

Same hardware, ran PerformanceTest on Windows Vista Ultimate x86, then installed the 64-bit version and compared results. Both installs were as clean as possible without omitting driver installs and Windows Updates. Both installs were as close to being "identical" as you could get. I did run the 64-bit PerformanceTest on the x64 Vista, and obviously PT 32-bit on x86.

I was surprised to see that the CPU test results were *significantly* lower on Vista x64, with the exception of integer math and image rotation. I'm hoping there is an architectual explanation to this and I'm comparing apples to oranges. All other test results were favorable to Vista 64 (as expected).

Vista x64 (32-bit in BOLD)
CPU - Integer Math: 307.6 222.6 (-27.6%)
CPU - Floating Point Math: 413.4 528.7 (27.9%)
CPU - Find Prime Numbers: 366.5 589.4 (60.8%)
CPU - SSE: 2491.1 4638.5 (86.2%)
CPU - Compression: 3571.3 5083.6 (42.3%)
CPU - Encryption: 18.9 31.9 (69.0%)
CPU - Image Rotation: 1284.8 1029.0 (-19.9%)
CPU - String Sorting: 2284.5 3321.8 (45.4%)



Obviously these results are counter-intuitive. What gives?

Thanks in advance.

[basilisk561]

Hi, just to clarify, my post is not solicit advice on how to improve the performance ofmy system in 64-bit Vista, but to inquire as to if it is possible and why Performance Test would potentially give the impression that CPU performance was less on 64-bit Vista when compared to 32-bit CPU benchmarks on the same hardware.

But, if anyone could suggest why my 64-bit Vista would actually be slower and things I could check, that would be appreciated.

I just can't imagine the results I got as being true...

[David (PassMark)]

It would have helped if you detailed the hardware and the versions of PT being used.

We did some similar tests a couple of years ago and got different results.

Check you you using the latest versions of PerformanceTest 32bit and 64bit.

If you have a multi-core / multi-CPU machine then check the number of processed in the Edit / Processes window is at least equal to the number of CPU multipled by the the number of cores in each CPU.

[basilisk561]

It would have helped if you detailed the hardware and the versions of PT being used.

We did some similar tests a couple of years ago and got different results.

Check you you using the latest versions of PerformanceTest 32bit and 64bit.

If you have a multi-core / multi-CPU machine then check the number of processed in the Edit / Processes window is at least equal to the number of CPU multipled by the the number of cores in each CPU.

Thanks. These were the latest versions of PT.

So it turns out it was the Processes setting was set to 1. I upped it to 2 and the results were favorable to the 64bit system as expected.

My question now is this -- is there a chance that the benchmarks I ran when 32-bit Vista was installed were with Processes set to 1? What is the default in the 32-bit version? Would the PT application try to detect the number of cores and autoset the default?

[David (PassMark)]

The software should auto-detect and set the correct value. So it would be interesting to know what hardware you have. And if you had been playing with the hardware. e.g. upgrading your CPU from an old single core CPU to a new dual core. Or playing with BIOS settings.

Or maybe you manually set this value without realising the significance?

[Ian (PassMark)]

This behavior is most likely caused by a bug introduced in v6.1.1013 where the default number of CPU test processes was 1, rather than the number of CPU's. This resulted in much lower CPU scores if this default was not changed.

We have now corrected this in the 32-bit version of PerformanceTest. The correction for the 64-bit version of PerformanceTest should be available shortly - in the meantime if you are using the 64-bit version of PerformanceTest v6.1.1013, you should manually set:
Edit->Preferences, CPU Tests, Number of Processes to the total number of CPUs (CPU packages x CPU cores).

Sorry for any inconvenience caused.

Regards,
Ian

[Ian (PassMark)]

This issue has now been resolved in the 64-bit version of PerformanceTest, v6.1.1014.

Regards,
Ian

[dava04]

32bit vs 64 bit

Same kind of issue, some CPU activities are slower under Vista 64 than XP32.
First number is unders xp32 second is under vista 64:

CPU inerger math: 207.9 vs 525.4
floating point: 566.3 vs 711.8
prime numbers: 601.4 vs 473.2
sse: 4826.9 vs 4243.5
compression: 5308.1 vs 5948.3
encryption: 32.4 vs 31.6
image rotation: 1078.3 vs 2183
string sorting: 3436.2 vs 3421

I know they are different OS's but maybe there is a better explanation than that.

The system specs were exactly the same for both tests, see below: (overkill? )
Item Value
OS Name Microsoft® Windows Vista™ Ultimate
Version 6.0.6001 Service Pack 1 Build 6001
Other OS Description Not Available
OS Manufacturer Microsoft Corporation
System Name HOME-PC
System Manufacturer OEM
System Model OEM
System Type x64-based PC
Processor Intel(R) Core(TM)2 Duo CPU E6550 @ 2.33GHz, 2331 Mhz, 2 Core(s), 2 Logical Processor(s)
BIOS Version/Date Phoenix Technologies, LTD 6.00 PG, 24/10/2007
SMBIOS Version 2.5
Installed Physical Memory (RAM) 4.00 GB
Total Physical Memory 3.25 GB
Available Physical Memory 2.30 GB
Total Virtual Memory 6.69 GB
Available Virtual Memory 5.46 GB
Page File Space 3.54 GB
Page File C:\pagefile.sys


[Display]
Name ATI Radeon HD 2600 PRO
Adapter Type ATI Radeon Graphics Processor (0x9589), ATI Technologies Inc. compatible
Adapter Description ATI Radeon HD 2600 PRO
Adapter RAM 256.00 MB (268,435,456 bytes)
Installed Drivers atidxx32,atidxx64.dll,atiumdag,atiumdva,atiumd64.d ll,atiumd6a.dat,atitmm64.dll
Driver Version 7.01.01.747
Resolution 1680 x 1050 x 59 hertz
Driver c:\windows\system32\drivers\atikmdag.sys (7.1.1.747, 4.05 MB (4,249,600 bytes), 29/03/2008 14:24)


Any thoughts on this, also 2D lines took at hit of a 90% drop, but my 2D perfomance in comparision other systems was never great and usually boardered on WTF!

Thanks.

[David (PassMark)]

I edited your post down as most of the detail about device memory maps is not relevant (overkill, as you mentioned).

You didn't mention what version of PerformanceTest you are using nor if you are using the 32bit or 64bit releases. Nor what test methodology you followed.

The explanation is probably something like background activity on Vista using up some CPU time during a couple of the tests. But without a detailed look at the machine it is hard to be sure.

2D results are worse in 64bit Vista due to the extra eye candy and often poor 64bit device drivers.

[dava04]

The software I used was your Passmark program, I used 64 bit on the 64bit system, downloaded 2days ago so i assume the most recent, and the 32bit one on the 32bit system, also the latest version.

My 'methodology' was to turn everything off, so basic running and the press 'run all tests'!

Abracadabra: passmark spits out numbers.

Thank you.

[David (PassMark)]

Make sure the machine is idle before running any tests. Turn of the Vista search indexing service, turn off any antivirus programs. For the tests that seem out of whack, like the prime number test, re-run them a few times, ignoring the low values. And / or increase the test duration from the edit / preferences window.

Longer tests are more likely to be interfered with by background tasks, but the average should be more stable.