Researchers Sour on Vista Service Pack 1 Performance

Stony Stevenson passed us a link indicating that a group of researchers has described Microsoft's upcoming Windows Vista Service Pack 1 as basically a performance dud. Researchers from the Devil Mountain Software group is claiming that a series of in-house benchmark tests showed that users hoping to receive a speed boost from the update will be disappointed. "Devil Mountain ran its DMS Clarity Studio framework on a laptop Barth described as a "barn burner" -- dual-core processor, dedicated graphics, and either 1GB or 2GB of memory -- to compare performance of the SP1 release candidate that Microsoft released last week with the RTM version that hit general distribution last January. The Vista RTM was not updated with any of the bug fixes, patches or performance packs that Microsoft has pushed through Windows Update since the operating system's debut. 'One gigabyte, 2GB [of memory], it didn't make a difference,' said [CTO Craig] Barth. 'SP1 was never more than 1% or 2% faster.'"

Comments

Has it ever improved efficiency?

[Ckwop]

Without wishing to troll, when has a Window service pack ever improved the speed of a Windows OS?

In fact, and I'm sure someone on Slashdot has raw data on this (that perhaps even shows I'm wrong), Apple are the only company who has ever achieved this on a regular basis.

I've found in my rather short development career is something scarily similar to the first law of thermodynamics: "Bad code once created can never be destroyed." In most commercial situations, the risk of breaking a routine far outweighs the benefit the change brings.

We've built an entire area of study, refactoring, on trying to sell the importance of keeping code clean. I'm still not 100% convinced that the case for refactoring has been made. If you spend three months refactoring, is the simpler overall structure really going to speed up development sufficiently to justify the capital outlay? In all but the very worst code-bases, the answer is unclear.Bear in mind, refactoring my cause you to notice bugs that you can't fix because it would break an interface. Now your code has to be badly structured to support this bad business logic. This can be enough to render the effort useless.

This is why service packs rarely improve functionality or performance. Windows XP SP2 is a notable exception. The risk is simply too great.

Simon

Why I even care one bit

[Mascot]

Vista has one great selling point as far as I'm concerned: DX10. It's inevitable that games will eventually require it, though so far it's not exactly a big deal.

So I notice Crysis has a "Very High" setting that's disabled for me in XP. Ok, I think, the first half or so of the game runs ok with High settings, so maybe it might just barely be playable on Very High. Just to be able to see what it looks like.

I boot into Vista and install the game there. Lo and behold, it runs at almost exactly half the FPS on High compared to in XP. Had to drop it to Medium to be even remotely playable. Needless to say, Very High is what I'd need to be to enjoy it with everything at max.

Is the culprit crap drivers for my hardware, general performance drain by Vista, or DRM using everything it can to make sure I'm actually allowed to use the computer today? I don't know, but I do know Vista has made me seriously try a Linux on a desktop for the first time (only used it for servers until now). If only more games supported it, or ran under Wine, I'd be happy as can be.

Re:Optimization

[TheRaven64]

Most worthwhile optimisation is done by rethinking the design, and to a lesser degree hand-coding parts where you know the realities better than the compiler can guess, and just how to exploit that. Micro-optimisations in the right place (not even at the assembly level, just tweaking a few algorithms or data structures, or even the code layout) can give huge benefits. I got a 25% speed gain from some code I was working on a few years ago just be moving a couple of functions into a header and marking them as static inline so the compiler could inline them. Memoisation of frequently-called functions can also give some benefits.

The hard part is usually not the optimisation, it's working out where the optimisations need to go. This typically involves wading through huge amounts of data from profiling runs.

Re:Are we shocked?

[jeffasselin]

The RAM usage at startup for a newly-installed system is simply absurd. 600-700MB is not an exaggeration. The graphics card needs for the new environment (without which it's mostly XP right? It's not like there's a new object-oriented file system in there right?) are quite hgh for most business needs.

The slow file copy isn't a joke. We're talking 1hr+ to copy 2.5GB to a FW hard drive from internal SATA. That's about 25MB/min, 120 times slower than the peak speed of FW. I think you can get more out of a parallel port.

There are some nice additions. But it's not worth the trouble, as some of the flaws totally override those.