What are your impressions of the Vista Beta 2?

[jinydu]

Some of the posts in this thread make me wonder:

Does Prime95 run faster using older versions of Windows?

[mfgoode]

This is all too high brow for a simple math'cian like me.
Nevertheless it might be of interest and help to you.

http://www.macosxhints.com/article.p...06060816033250
1. http://www.macobserver.com/article/2006/04/10.11.shtml
2. http://blogs.msdn.com/hans_vb/archiv...02/614799.aspx

Mally

[James Heinrich]

I have it dual-booting on my AMD system. It seems too locked down for its own good: it has confirmation dialogs for anything you open/change/do (so it seems at least). And I have problems like not being able to share files off it to my other Windows machines, and Prime95-specific I can't seem to get it to work properly. Notably, I get errors about not being able to write to file (forget exact wording) -- after factoring a number (for LMH) it doesn't write to results.txt (but running a TF assignment from Primenet works fine (the second time at least, at first it wouldn't connect to Primenet at all)). Strange things like that (beyond the expected things like not having drivers for half my hardware). Maybe playing with it a little more I'll get it working properly, but I'm not happy with it yet. Oh, and P-1 is basically impossible with Vista, since it takes 95% (literally) of physical RAM for its own purposes (pre-caching of some kind, the benefit of which I haven't seen yet). So on a 1GB machine Vista uses 946MB sitting at the desktop...

Oh, and it looks pretty

[Xyzzy]

You can turn off those confirmation dialog boxes. I agree that they were incredibly annoying.

It got to the point I just blindly clicked "Yes" to any dialog box. That can't be a safe security practice.

[cheesehead]

Originally Posted by jinydu

Does Prime95 run faster using older versions of Windows?

When I migrated from Win3.11 to Win98, and later from Win98 to WinXP, I found that Prime95 (same version) iteration times were consistently longer, more than 1% in the first case, a little less in the second. (Of course, it was not that Prime95 was running slower, but instead that the newer OS imposed more overhead. C'est la guerre. )

[James Heinrich]

Moving from Win3.1 to Win98 and using the same Prime95 version (presumably 16-bit) would impose an emulation layer. Conversely, while I can't speak about 16/32-bit changes, the 64-bit Prime95 under Vista vs the 32-bit version under XP has slightly faster iteration times (~0.5%) but is hugely faster for trial factoring:

Best time for 58 bit trial factors: 2.920 ms vs 4.657 ms.  (37%)
Best time for 59 bit trial factors: 2.924 ms vs 4.671 ms.  (37%)
Best time for 60 bit trial factors: 3.105 ms vs 4.654 ms.  (33%)
Best time for 61 bit trial factors: 3.200 ms vs 4.674 ms.  (32%)
Best time for 62 bit trial factors: 3.796 ms vs 8.366 ms.  (55%)
Best time for 63 bit trial factors: 4.087 ms vs 8.403 ms.  (51%)
Best time for 64 bit trial factors: 5.268 ms vs 10.847 ms. (51%)
Best time for 65 bit trial factors: 6.240 ms vs 10.774 ms. (42%)
Best time for 66 bit trial factors: 6.201 ms vs 10.776 ms. (42%)
Best time for 67 bit trial factors: 6.183 ms vs 10.775 ms. (43%)

[jinydu]

Originally Posted by cheesehead

When I migrated from Win3.11 to Win98, and later from Win98 to WinXP, I found that Prime95 (same version) iteration times were consistently longer, more than 1% in the first case, a little less in the second. (Of course, it was not that Prime95 was running slower, but instead that the newer OS imposed more overhead. C'est la guerre. )

So I suppose it is probably safe to say that Prime95 on Vista (when the public release finally comes) will be even slower...

[James Heinrich]

I say for certain on the final release of Vista, nor for the 32-bit Prime95 on Vista, but my (admittedly limited) testing shows Prime95 in Vista ~0.5% faster for LL and ~40% faster for TF...

[Mystwalker]

Originally Posted by James Heinrich

64-bit Prime95 under Vista vs the 32-bit version [...] is hugely faster for trial factoring

The important factor is most likely the 64 vs. 32 bit aspect.
You should see a similar result when comparing both versions of WinXP only. TF is already ported to 64 bit, whereas porting LL is much more complex...

Differences of <1% between OSes are more or less negligible, as a lot of things (background processes, drivers, coincidence) can influence the results.

Vista could be better suited for DCing on work computers, depending on the amount of work the GUI offloads to the graphics card, the number and load of background processes etc.
But I won't expect massive changes. After all, the OS itself not that important when it comes to the performance-critical parts of the code.

[James Heinrich]

Originally Posted by Mystwalker

The important factor is most likely the 64 vs. 32 bit aspect.

Absolutely

I've attached some numbers to quantify things a little, comparing Prime95[32] on XP vs. Prime95[32] and Prime95[64] under Vista64:

• LL seems to be nearly dead even under XP and Vista for Prime95[32], insignificantly faster for Prime95[64] on Vista
• TF seems marginally faster for Prime95[32] (~1.5%) for <=62-bit, and hugely faster (30-50%) for Prime95[64] across the board.

(numbers are from an Opteron165@2601MHz, running single Prime95 instance).

[jinydu]

This may seem like a stupid question, but:

Is "64-bit" vs "32-bit" (as used earlier in this thread) a property of the operation system or the hardware?