Vlite Vista RAM Footprint & System Latency ... still huge ?!

[benifin]

Howdy nuhi and everyone else - keep up the great work.

nuhi - even with the latest [ excellent ] beta, the RAM footprint of a fully-stripped Vlite Vista 32 install - with a video driver installed - is still around the 240 <-> 280 meg.

My fully-stripped XP X32 SP2 NLite install - with video driver installed - has a RAM footprint of ~ 70meg

My fully-stripped XP X64 SP2 NLite install - with video driver installed - has a RAM footprint of ~ 104meg

In the XP X32 SP2 NLite install, the DPC Latency Checker runs at a consistent 1ms with occasional peaks up to 5ms - never higher.

In the XP X64 SP2 NLite install, the DPC Latency Checker runs at a consistent 8ms with occasional peaks up to 16ms - never higher.

In the VLite Vista 1.0B install, the DPC Latency Checker runs at a consistent ~30ms with regular peaks of ~60ms and semi-regular peaks up to 1000ms.

nuhi

Do you think it is probable / possible that VLite will eventually have a RAM footprint of around 100meg [with Video Driver installed] and a DPC Latency reading similar to the XP x32 and XP x64 readings ?

Thanks and keep up the great work.

Ben

Edited June 9, 2007 by benifin

[nuhi]

Fully stripped XP vs fully stripped Vista in the future...well I think that XP will always be lighter, same as the Windows 2000 is when compared to the XP.

Thing is that you always want that something extra from the new core, be it the looks or something like DX10. And your machine is much faster then before so you switch.

Reason why I remove all these components is mainly to have a clean slate, better organized OS, then comes the optimization of hdd, ram and cpu usage as a bonus. While the OS is new like Vista is then it's reasonable to shift priorities towards the performance first, especially when you compare how much faster the Lite XP is.

[aviv00]

hey all

i try runing "DPC Latency Checker" and get 1ms with vista

with what u made that latency test ?

Edited June 9, 2007 by aviv00

[benifin]

Hi aviv00

Ge this download and then report back.

http://www.thesycon.de/dpclat/dpclat.exe

Ope it, maximise it and let it run for, say, 5 mins - then advie the maximum peak and the regular peak

There is NO WAY you will get, or are getting a stable 1ms reading with VLite Vista 32.

Ben

[aviv00]

http://img526.imageshack.us/img526/4673/bizilatencyip5.jpg

http://img123.imageshack.us/img123/9045/idlelatencyei1.jpg

not 5 min but do the thing

[benifin]

Hi aviv00

As I said, you are getting readings that are VERY high.

You are miss-reading graph.

Your system is peaking at 1000ms [ 1 full sec ] NOT 1ms [ millisecond ] .

Ben

[impulse686]

There are 1,000,000 µs and 1,000 ms in a second. The graph measures latency in µs, so aviv00 is in fact getting around 1ms...unless the units in the program are off by a factor of 1000.

Edited June 11, 2007 by impulse686

[benifin]

Hi impulse686

Its a misuse of symbols of the DPC Latency graph.

You will see it reports at 1 second intervals, but they list that as 1000µs - its a mis-type in the graphical display part of the program.

Ben

[jamieo]

Its a misuse of symbols of the DPC Latency graph.

You will see it reports at 1 second intervals, but they list that as 1000µs - its a mis-type in the graphical display part of the program.

It's not a mis-type or a misuse, your assumptions are incorrect. It reports once per second AND the test interval is 1000µs. ie. it takes around 1000 readings per second (the interval is 976µs on my machines) and reports the highest each second. The page where that tool is downloaded from explains:

DPC Latency Checker periodically updates its internal statistical data at an interval displayed as Test Interval. This statistical data is queried and displayed on screen once per second. <snip> Each bar represents the maximum DPC latency occurred within one second.

Btw, I get similar readings to aviv00. Disabling all possible drivers (inc usb, firewire, sound and network) makes little difference. Also tried disabling some services with no improvement. I suspect there are components/drivers within the core of Vista (not oem drivers) continuously polling and causing the peaks.

Bear in mind the peak may only occur once within the reported interval. Therefore, a reading of 1ms every 5sec is likely a single peak of 1ms for one test interval out of 5000. Also, the test is only a guide as to how deterministic your system is. That is, how close it is to responding to requests in real time, not something windows is intended for.

That said, it's obvious that Vista performs poorly in this area - my old Athlon XP1800 with XP Pro SP2 is rock solid at around 35µs (fluctuating between 5-80µs) with an absolute maximum of 188µs and I'm sure my Vista system will perform at least as well with XP on it. However, vLite is unlikely to improve matters - if it does it will be a side effect of component removal but the culprit is likely to be something you wouldn't want to remove - or something you can't. Hopefully someone will find the cause and a way to tweak it, then vLite could possibly include that tweak as an option...

Edited June 11, 2007 by jamieo

[MFCK]

Hi all !

After one year of Vista life, i can say that latency is always higher on Vista x86 than x64.

I see always Latency equal of 1000µs on Vista x86 equiped PC (values change depending on hardware).

And on my PC with x64 installed, i have a latency value equal of 17µs/20µs !! (900µs on my laptop with Vista x86) !

This is the case with full genuine Vista or Vlited Vista.

Has someone constated this fact like me??

Can someone explain me why this difference ??