Performance Comparison of 3 Thinkpads in C++ compilation

[mikesugar]

Hi. In case anyone is interested I just did an unscientific performance comparison of the three machines below by doing a "rebuild all" on a large C++ project using Visual Studio 6.

#1 2379DXU
T42, 1.7GHz Pentium M, 1GB, 60GB@7200

#2 2623DDU
T60P, 2GHz T2500, 1GB, 100GB@7200

#3 87414BU
T60 Widescreen, 2GHz T7200, 2GB, 100GB@7200

Rebuild all times in Seconds:
#1: 127 sec
#2: 44 sec
#3: 34 sec

Mike

[own6volvos]

Any chance you could email me that to test it out on my desktop core 2 duo, and T7400 T60? I could also try it out on my C100 with a 1.6 celeron to add to the mix. I have visual studios installed on almost everything since I am a tutor/grader for a C# class.

[RonS]

I'm sorry, but this means ABSOLUTELY NOTHING.

1. File fragmentation can wildly effect build times on large projects, particularly on precompiled header (PCH) files. This can and will vary grately between machines.

2. The first time you compile a project on a target machine, the binaries (PCH, OBJ, etc) will be dirty and need to be rebuilt because they are coded with a specific DevStudio install. With a new install, they are all rebuilt. Subsequent builds may or may not need to rebuild those files, depending on LOTS of factors. This is especially true on Visual Studio 6.0. I'd bet that the T42 machine had a set of "dirty" files. If you do a full rebuild two more times, you'll probably see better numbers for the second build.

3. RAM for cached files (compiler binary executable and headers) will make a massive difference. This is probably why the machine with 2GB of memory did better than the others.

There's a lot more, but this is just total BS. Sorry to dump on your thread.

[own6volvos]

Well, by rough comparison, the performance jumps between each processor is in line with the other accepted comparisons. Perhaps some dhrystone tests on each machine to compare the gains again?

[RonS]

I have about two solid years of software development time on a T42p, using both VC6 (C/C++) and VS 2003/2005 (C#). I have about 9 months on a T60p, using the same source projects and environments.

I can barely tell the difference in build times. There no way that real-world performance of the T60 is twice that of the T42 with everything else being equal.

The T60p is certainly faster than any T42 in the graphics world. And with apps that are CPU bound (as opposed to I/O bound), the T60 is faster especially with CPU-bound multi-threaded apps.

Real-world performance differences are determined by many factors, some of which are:

- System RAM (1GB-2GB is best)
- Disk speed (7200rpm is best)
- Video (v5200/v5250 machines perform faster than x1300 or GMA950)
- Anti-virus programs, particularly Symmantec Anti-virus. Dump that program, install a free copy of Kapersky AV, and watch a smile return to your face.
- Running on battery. Make sure your power options are configured for the CPU to run in adaptive mode. I use adaptive mode all the time, even when plugged in. CPU performance times are the same as on battery, and the system runs cooler.
- Services and start-up programs. Get to know your task list. Audit your services.
- Disk fragmentation, bloated registry, etc. Disk fragmentation particularly effects the system pagefile.sys.

[christopher_wolf]

Having used both the T4X and the T6X to develop code in a scientific setting, I can pretty much agree with Ron's analysis of it. If you just look at the build times for a given program, you are going to introduce quite a few unknowns that happen to be very difficult to pin down with an acceptable level of accuracy. Even with the Intel compilers, the difference might be on the order of a few seconds, again depending on where the resources for it are, fragmentation, how the memory is handled, etc.

[mikesugar]

Ron, let me put your mind at ease. I am an engineer (MS Mech E '92) and know a thing or two about proper experiments. Although this was certainly "unscientific", I took steps to keep the field level.

The 2 T60s were brand new restores of factory settings. With VS 6 and VS.NET 2003 recently installed.

The T42 was not a fresh install, but recently defragged.

The times were from the third consecutive "Rebuild All", not "Rebuild", so I think that eliminates a lot of your concern about how much work was really done in each test. Incidentally, the T60P (T2500) got 2 sec faster (from 46 to 44 sec) from the first to the second rebuild and stayed there for the third. The other two machines turned in the same times on all rebuilds.

All machines were running on AC power.

I wasn't running any other apps during the builds, but any background services were left as is. (None of the machines was running Antivirus)

Maybe the T42 was at a little disadvantage because it's been in long term use. but I've had this big C++ project for several years. It compiles noticeably, but not substantially faster on the T42 than it did on my A20p, and it hasn't gotten any slower over time, so I think the T42 performance is fair to use.

The only difference that could unlevel the field is that the T60s were running XP Pro and the T42 was running Win2K Pro.

But bottom line, I'm talking about machine performance, not CPU performance. I think the results give pretty accurate representations of realtive performance for developers doing large compiles. I have to say, I was shocked to see how fast the compile log messages flew by on the T60 machines.

Should you use the data to indicate how the machines will work with 3DS max? No. DVD rendering? No. Transcoding? No. Small C++ projects? .... No. But for developers contemplating an upgrade to a Duo thinkpad, the T60 will blow the doors off the T42, and I'm pretty sure you will see relative compile performance within a few % of what I reported. I'd bet money that anyone "rebuilding all" on a large C++ project with visual studio 6 will find that a T60 with 2GB and a T7200 is about 4X as fast as a T42 with 1GB and a pentium M @1.7GHz

I'm sorry, but this means ABSOLUTELY NOTHING.

1. File fragmentation can wildly effect build times on large projects, particularly on precompiled header (PCH) files. This can and will vary grately between machines.

2. The first time you compile a project on a target machine, the binaries (PCH, OBJ, etc) will be dirty and need to be rebuilt because they are coded with a specific DevStudio install. With a new install, they are all rebuilt. Subsequent builds may or may not need to rebuild those files, depending on LOTS of factors. This is especially true on Visual Studio 6.0. I'd bet that the T42 machine had a set of "dirty" files. If you do a full rebuild two more times, you'll probably see better numbers for the second build.

3. RAM for cached files (compiler binary executable and headers) will make a massive difference. This is probably why the machine with 2GB of memory did better than the others.

There's a lot more, but this is just total BS. Sorry to dump on your thread.