Author Topic: What Intel Giveth, Microsoft Taketh Away (Read 24,227 times)

Ralf Maximus · « **on:** November 17, 2007, 03:23 AM »

Fascinating article chronicling what we've all suspected: that each version of Windows + Office consume all available workstation horsepower, no matter how new the hardware.

http://exo-blog.blogspot.com/2007/09/what-intel-giveth-microsoft-taketh-away.html

Included are interactive benchmarks of every conceivable combination of Windows and Office since 2000.

Found via Raymond Chen's superb blog, the current topic as I type is a research paper on "Who would win in a fight between a penguin and a lemur?". No, I'm not kidding.

zridling · « **Reply #1 on:** November 17, 2007, 05:53 AM »

Glad you listed this, Ralf. This was the first thing I noticed with Vista back in early February — it ate everything you threw at it. Office 2007 is the same way, as the post states, requiring 12 times the memory and 3 times the processing power as previous versions of Windows and Office.

Sorta ruins the notion of "optimizing" the codebase for better/faster/newer hardware. I'm sure they're already working on how to slow down Penryn chips.

icekin · « **Reply #2 on:** November 17, 2007, 10:15 AM »

http://blip.tv/file/340692

Okay, I thought this was funny. If you are a VISTA user, please do not get offended.

nontroppo · « **Reply #3 on:** November 17, 2007, 12:46 PM »

F**k, that chart is just sensational; even though I believe the overall premise, I cannot believe the magnitude of the that. I think the sheer scale of regression with Vista can't be so large. I suspect a change in working set allocation is responsible for the memory change (more memory allocated is not necessarily bad), but I cannot explain the overall speed regression. I don't have a clear sense of how that data was aggregated, nor thus its variability. I do realise incredulity is hardly a cogent argument against that data ;-)

Darwin · « **Reply #4 on:** November 17, 2007, 01:19 PM »

After my experience with a Vista notebook (that was the same model with the same spec as an earlier XP MCE notebook that I tried a year ago), I believe the chart. Dual Core processing and 2GB RAM and it was a slug compared to my XP Pro machine with a four year old Centrino chip and 2GB RAM. Loved the hardware, loathed the OS's resource hogging.

Interesting that the analysis also supports another of my suspicions: Win2k on a 7 year old PIII-E machine with 512MB RAM was quicker than my Centrino with a gig of RAM. Makes me want to dualboot the Centrino with Win2k and install Office 2000 on it...

Sidenote: when I first read the "Lemur vs. Penquin" aside in Ralf's first post, my mind managed to convert it to "Leopard vs. Penquin" - imagine my surprise when I found not a shootout between OS X and Linux but a paper speculating about who would win a real fight - a flightless bird or an arboreal primate. Nice.

Ralf Maximus · « **Reply #5 on:** November 17, 2007, 03:39 PM »

What I found interesting was the conclusion of the article, the weird symbiotic nature of hardware and software vendors. When Vista was delayed for so long, Intel was left with a product line of powerful new CPUs that nobody wanted. It really hurt their bottom line, and one might assume was one of the pressures placed on Microsoft to kick Vista out the door when they did.

As soon as Vista hit the streets and people learned that the "minimum requirements" were, shall we say, optimistic... then Intel's fortunes improved. Apply that same economic pressure to all of Intel's customers -- Gateway, Dell, HP/Compaq, etc -- and it's easy to see how our whole IT industry is driven by (cue trumpets) software bloat.

On the up side, I love the idea of a dual-CPU quad-core ~4GHz workstation on my desk, and such wouldn't exist without stuff like Vista + Office2007. So in that, I thank you, mindless marketing morons and greedy software executives.

Darwin · « **Reply #6 on:** November 17, 2007, 04:01 PM »

I've now read the article and the comments - REALLY interesting stuff. My main concerns reading the piece were voiced by a number of posters to the comments section: it would be interesting to see the various versions of Windows and Office combined and tested as well and it would be more meaningful (to me) if the tests were conducted on actualy hardware, rather than VMWare.

nontroppo · « **Reply #7 on:** November 17, 2007, 04:24 PM »

Quote from: Ralf Maximus on November 17, 2007, 03:39 PM

Apply that same economic pressure to all of Intel's customers -- Gateway, Dell, HP/Compaq, etc -- and it's easy to see how our whole IT industry is driven by (cue trumpets) software bloat.

Except quite a number of PC manufacturers have grumbled pretty loudly over Vista (and pushed for XP retention). Considering the morasses of molasses Vista is, by the bloat argument, Vista must the best OS ever crafted, and PC sales rocketing up like never before. Do I suspect a fracture in the cyclical inevitability of bloat-dharma?

And after just helping install Leopard on an 867Mhz consumer ibook from eons ago, and having it smoothly work (including the eye candy), bloat-dharma is not inevitable. I do fear for what may arrive when MS release the version of Office 2007 for Macs in a month and a half though...

Quote from: Darwin on November 17, 2007, 04:01 PM

it would be interesting to see the various versions of Windows and Office combined and tested as well and it would be more meaningful (to me) if the tests were conducted on actualy hardware, rather than VMWare.

I'd like to see earlier versions of Office on Vista. But as to the virtual machine, I doubt that will make much difference (unless Vista has some kernel bug which triggers problems on a VM, though nothing has been described), and certainly would greatly complicate the process of testing.

Darwin · « **Reply #8 on:** November 17, 2007, 05:24 PM »

My concern WRT the VM is that on my notebook, booting Win2k with 1GB virtual RAM into VirtualPC takes longer than booting Win2k on a PIII-E notebook with 512MB RAM. The author of the piece states that he virutalized some installations but used a P4 machine with 256MB RAM for others. I guess I should re-phrase my concern as being limited to wishing to see the tests run either on the same hardware or all virtualized on the same machine.

Darwin · « **Reply #9 on:** November 17, 2007, 05:30 PM »

Er, the above should read on my 1.4GHz Centrino notebook with 2GB RAM vs the PIII-E which clocks in at 600MHz...

nontroppo · « **Reply #10 on:** November 17, 2007, 05:40 PM »

I think all the main tests in the table claim to be virtualised, which is his first phase analysis, then he will start to use real hardware. At least the table says (virtual) across the board. He is not very clear.

Booting XP into 512MB or 768MB virtual machines on my macbook via VMWare is blazingly fast (faster than any other XP machine I use), I think it depends on the Host capabilities (centrino doesn't support intel virtualization instructions?)

Darwin · « **Reply #11 on:** November 17, 2007, 07:01 PM »

I just timed it - virtualized on the Centrino, Win2k took 80 seconds to load. On my PIII-E it takes about... 80 seconds. So much for seat of the pants impressions... I guess this just reflects how anaemic my current CPU is compared to a modern processor - it's single core and the first generation P-M versus the dual core speed demons available today. Of course, I suppose I could play around with other virtualization options. Frankly, though, I'm tempted to lose the one that I've got as I don't really have a need for it (beyond the tried and true "gee whiz, neato" factor).

Lashiec · « **Reply #12 on:** November 17, 2007, 08:52 PM »

The duopoly between operating systems and hardware makes me laugh. You could say there's some duopoly between hardware manufactures and some software developers, but frankly, nobody is waiting for a new operating system to come out and take advantage of their newly acquired hardware. Hardware constantly advances, independently of the software, and people updates it as they want, no one is going to stay for years with their old systems (the saying "if it works, why change it?" has a limited life IMHO).

I wonder... if there's this "Wintel" thing, why is Intel one of the biggest contributors to the Linux kernel, and why Apple now uses x86 architecture? Or why they're pushing their products in every market they can put a processor in. Intel fortunes now depend on themselves, maybe there was a time that Windows was capital in their strategy, but if Windows goes down, and something else takes its place, Intel will simply chug along. And AMD as well (if they can finally rid of red numbers next year).

Nobody wanted a Core 2 Duo?

Of course, all of this can't hide the fact that Microsoft went a little overboard this time with Vista requirements, but in the case of Office, OpenOffice can't really be an example of how to make efficient office suites. And like the recent blog entry that Jeff Atwood published, software tends to spoil after some versions. Ask Adobe, Symantec, Corel, etc. Hey, a theory: Big company -> Inefficient software

f0dder · « **Reply #13 on:** November 17, 2007, 10:30 PM »

Quote

but in the case of Office, OpenOffice can't really be an example of how to make efficient office suites.

Amen to that!

Although looking aside from the bloat and pathetic slow startup times, it does seem (first impression, not benchmarked) that their scripting runs faster than the VBA in office...

Darwin · « **Reply #14 on:** November 17, 2007, 10:34 PM »

Yes - Open/Star Office can hardly be offered up as examples of speed demons. Having said that, it is worth noting that from my understanding this is actually due to their ongoing decision to preload all of the components whenever a user fires up any part of the suite...

Lashiec · « **Reply #15 on:** November 18, 2007, 12:43 AM »

Yes, that's the main problem. Supposedly they were going to fix it in OpenOffice 3.0, but for what I saw in one of the slides showed at the last OpenOffice.org conference, they were going to include some performance improvements, but not major, so that may mean they're not going to solve the issue. To me, it seems that it would require significant work, and if the suite gets more and more complex, things could get worse from a coding perspective.

OK, let's go back to schedule.

nontroppo · « **Reply #16 on:** November 18, 2007, 01:08 AM »

Well, it would be interesting to see if an analogous script benchmark could be made for OpenOffice *and* Office - to compare OO+Linux against O2007+Vista on the same hardware.

f0dder · « **Reply #17 on:** November 18, 2007, 01:19 AM »

You might also want to run OO/Win32, OO/Win64, O2007/WhINE, just for the fun of it

Deozaan · « **Reply #18 on:** November 18, 2007, 03:50 AM »

I wonder how well Vista and Office 2007 would run on a PC with a Gogol-core CPU with a Gogolbyte of RAM.

One can always dream...

zridling · « **Reply #19 on:** November 19, 2007, 10:00 AM »

And here I was under the mistaken assumption that computers would get faster, better, and save me lots-o-time!

nontroppo · « **Reply #20 on:** November 19, 2007, 11:35 AM »

Zaine: it just depends on the OS manufacturer

I actually do think hardware has advanced and provides significant advantage. The core2duo, considering the fact that the future is virtualisation and hardware abstraction (did you see the new Ubuntu JeOS?), has been a godsend (far more than just lumping two processors together). I am working in ways now that really were unthinkable a few years back, and my productivity boost is tangible.

nontroppo · « **Reply #21 on:** November 19, 2007, 12:31 PM »

Hm, it appears SP1 will not do much for performance according to current beta versions (but perhaps they still have debug code and whatnot riding along):

http://blog.scotsnewsletter.com/2007/09/29/hands-on-windows-vista-service-pack-1-beta/
http://www.pcworld.ca/news/column/3eef651f0a010408008b33e8065121c5/pg1.htm

nontroppo · « **Reply #22 on:** November 19, 2007, 03:03 PM »

A follow-up post from the same guys that did the initial testing:

http://exo-blog.blogspot.com/2007/11/vista-sp1-performance-dud.html

Ralf Maximus · « **Reply #23 on:** November 19, 2007, 03:07 PM »

Quote from: nontroppo on November 19, 2007, 03:03 PM

A follow-up post from the same guys that did the initial testing:

http://exo-blog.blogspot.com/2007/11/vista-sp1-performance-dud.html

The comments are... ah, a bit acidic.

lanux128 · « **Reply #24 on:** November 20, 2007, 12:24 AM »

yes, i think this is obvious.. every conceivable breakthroughs in processor technology are consumed immediately by Microsoft's software.. the modern day equivalent would be Vista as a way of getting back at dual-core processors..