Hyperthreading with Creamware

Immanuel · Post by **Immanuel** » Fri Jun 13, 2003 8:32 pm

Just don't wet your diaper ok

sandrob · Post by **sandrob** » Sat Jun 14, 2003 2:29 am

hey, hurbid! i heard cw almost bamkuprt?

borg · Post by **borg** » Sat Jun 14, 2003 3:37 am

On 2003-06-13 21:18, hubird wrote:
snif snif, my name is Hubird, not Hubrid, why are so many of us calling me hubrid, I'm not hybrid at all, I love just my little big mac, woeahhh

yeah everybody... it shows a big disrespect for huub

you all better learn how to spell properly... here's what happened to DXL: http://www.thespark.com/content/122299/satan.html

marcuspocus · Post by **marcuspocus** » Sat Jun 14, 2003 5:52 am

Post by **hubird** » Sat Jun 14, 2003 8:50 am

paulrmartin · Post by **paulrmartin** » Sat Jun 14, 2003 1:24 pm

On 2003-06-13 16:05, paulrmartin wrote:
Question:
Isn't hyper-threading like having 2 CPU's?
If so, could I use my computer to do audio on one side and video on the other?

Paul

Answers? Comments? Am I off-track?

w_ellis · Post by **w_ellis** » Sat Jun 14, 2003 1:38 pm

It's fairly complicated, but these might help...
http://www.anandtech.com/cpu/showdoc.html?i=1746
http://www6.tomshardware.com/cpu/20021227/index.html
http://www6.tomshardware.com/cpu/20021202/index.html

Cheers,
Will

paulrmartin · Post by **paulrmartin** » Sat Jun 14, 2003 4:04 pm

From the second link:
"...Threading an existing serial application increases the complexity of the application, Intel says. Sharing of resources, such as global data, can introduce common parallel programming errors such as storage conflicts and other race conditions. Debugging such problems is difficult as they are non-deterministic, and introducing debugging probes, such as print statements, can mask these errors."

Sharing of resources, such as gobal data...: This caught my eye.

Do you think this would be a cause for concern for us?

[ This Message was edited by: paulrmartin on 2003-06-14 17:04 ]

paulrmartin · Post by **paulrmartin** » Sat Jun 14, 2003 4:21 pm

This is a VERY encouraging thread:

http://www.intel.com/home/do_more/music ... rpheus.htm

Very encouraging indeed!

Immanuel · Post by **Immanuel** » Sat Jun 14, 2003 11:15 pm

He's got a 2GHz P4. Those are not made in HT versions.

sandrob · Post by **sandrob** » Mon Jun 16, 2003 7:52 am

it looks like "never ending story". so...
i just did recording from cubase to wavelab (asio 24bit source -> stm1632 -> 24bit wave dest.) and i had a lot of clicks and distortions.
then i update computer from "acpi multyprocessors fucken pc" to...

... and now the recorded tracks are clear - problem is solved.
fynaly, "acpi pc" mode works without problems to me instead of all other modes.

i will like if someone with "acpi multyprocessors pc" can try to do the same: try to do recordings from cubase to wavelab (asio 24bit source -> stm1632 -> 24bit wave dest.)????

paulrmartin · Post by **paulrmartin** » Mon Jun 16, 2003 8:22 am

Sandro, Can't you bounce within Cubase itself and then process the file in wavelab, if need be? Just curious

sandrob · Post by **sandrob** » Mon Jun 16, 2003 11:58 am

you mean that i do recording in the stereo cubase's chanell?!
yes! if i use asio only, than works perfect in every case. so i can connect asio to asio and do record in cubase.
something is wrong with wdm drivers if computer is in "acpi multyprocessors pc" mode - something with recordings by useing "wave dest". playback is correct.

i will like if someone can try the same?

btw: i found that pulsar works better in modes with 16 irq's instead of 22.

_________________
got my mojo working, but it just won't work on you

[ This Message was edited by: sandrob on 2003-06-16 12:59 ]

[ This Message was edited by: sandrob on 2003-06-16 13:05 ]

Micha · Post by **Micha** » Tue Jun 17, 2003 9:09 am

Hi Sandrob,
great! Your config makes very much sense. The 2nd CPU is not a real CPU, so I can imagine that WDM drivers could get confused when they want to do parallel processing and then, because in reality it is only one CPU that can do the trick only with a little latency -however small the amount may be-, they find out that it is not really parallel. A normal WDM is not prepared to do latency correction in this case, so the result may be nasty.
The trick with HT is that the CPU is divided in very small time slices with TWO input units instead of one, one for the even numbered slices, one for the odd. So the capacity of the CPU can be used more efficiently in normal use. In continous use (arithmetic!, MPEG4 etc.) there is no advantage.
But usually we have more programs running, not only SFP.
So your solution seems to become our standard here, I suppose.
Thanks for your excellent investigation!

sandrob · Post by **sandrob** » Tue Jun 17, 2003 11:13 am

i like to share my new experiences but i have no enough knowledge to know what's realy happends here.

so, in the "acpi mode" (note: this is not "acpi uniprocessor mode") i still have 2 virtual cpu's in managger and i don't have any problem with wdm drivers!!
so, i guess that pulsar can handle with 16 irq's instead of 22?!