Log in

View Full Version : Hyper Threading



RipItUp
08-30-2002, 02:56 AM
I'd like to ask Brian / Howard whether Intels Hyper Threading on forthcoming 3.06GHz cpu's be useful with the DF client?

Regards

Andy

Brian the Fist
08-30-2002, 08:35 AM
First Ill have to find out what you're talking about...

RipItUp
08-30-2002, 10:00 AM
I think this sums it up :-

http://cedar.intel.com/media/training/hyper_threading_intro/tutorial/index.htm

Intel will be rolling this out on upcoming processors as a half way house to having multiple cpu cores on one chip. It requires the software to use it's features efficiently though to really get the most out of it.

Regards

Andy

beefdart
08-30-2002, 12:25 PM
Hold on... So do all Xeon processors currently support this?

"Hyper-Threading Technology is built around Intel® NetBurst™ microarchitecture. Currently, the Intel® Xeon™ processor supports Hyper-Threading Technology. "

Dam Intel and their lack of being able to explain their twisted crappy hardware.

Go AMD running FreeBSD!!

bwkaz
08-30-2002, 01:19 PM
Yeah, Xeon's have hyperthreading. We've got a set of 3 dual-P4 servers at work, that all show up as 4-CPU machines to Windows (of course, that's a BIOS setting -- we could set it to appear as 2 CPU's, but that's no fun ;) ).

Go AMD indeed.

TheOtherPhil
08-30-2002, 02:18 PM
Hyper-threading sounds good on paper, but in practice (http://www.anandtech.com/IT/showdoc.html?i=1606&p=1) doesn't live up to it's claims....that's why it is set to disabled by default on those systems that support it.

bwkaz
08-30-2002, 03:28 PM
Originally posted by TheOtherPhil
Hyper-threading sounds good on paper, but in practice (http://www.anandtech.com/IT/showdoc.html?i=1606&p=1) doesn't live up to it's claims....that's why it is set to disabled by default on those systems that support it. Heh, now why doesn't that surprise me? Intel, do something that only sounds good on paper? Naw, never... :rolleyes:

Jodie
08-31-2002, 09:46 PM
We've seen some pretty significant speed improvements with hyperthreading on our neural network core... (P4 Xeon 2.0ghz)

erk
07-10-2003, 08:42 AM
Also hyper threading lets you split a task over a network of pc's acting like one virtual processor, eg. an openmosix cluster:

http://cedar.intel.com/cgi-bin/ids.dll/content/content.jsp?cntKey=Generic+Editorial%3a%3axeon_openmosix&cntType=IDS_EDITORIAL&catCode=BMB

rsbriggs
07-10-2003, 09:21 AM
There are more than a few folders out here that run single HT enabled chips as two CPUs and dual HT processors as 4 CPUs.

Just make multiple copies of the client, and run them from different directories....

I have one quad HT processor Xeon box that runs 8 copies of the folding client....

djp
07-10-2003, 12:13 PM
Also hyper threading lets you split a task over a network of pc's acting like one virtual processor, eg. an openmosix clusterI think you've misread the article slightly. OpenMosix lets you split a task over a network. HyperThreading just lets you make more thorough use of your PC by tricking the OS into thinking you've got dual processors.

HyperThreading can give some exciting performance gains, but they're mostly going to be seen in real-world end-user scenarios where you're concurrently performing two dissimilar tasks. If you're running two folding clinets on one hyperthreaded chip, both threads are hitting the same several sub-components of the microprocessor. You won't get as much of an acceleration as a user who's running one CPU intensive task and another not-so-very CPU intensive task.

I don't speak for Intel. They don't speak for me. I'm just QA testing some of their network software and hardware.

arcturius
06-24-2004, 10:47 PM
Originally posted by rsbriggs
There are more than a few folders out here that run single HT enabled chips as two CPUs and dual HT processors as 4 CPUs.

Just make multiple copies of the client, and run them from different directories....

I have one quad HT processor Xeon box that runs 8 copies of the folding client....

How much higher is your output when running 2x clients per HT CPU as opposed to only one per CPU? (I've got 2x dual 2.8HT Xeons, but don't feel like doing the testing, if I can get a quick answer from someone who's already done it ;) )

Grumpy
06-25-2004, 02:36 AM
Results have shown 5-10 % increase with 2 HT Clients ;)

tpdooley
06-25-2004, 04:46 AM
And up until recently, if those multiple clients weren't set up with their own temp directories, the clients would stop intermittently..

Digger
06-25-2004, 06:23 AM
Not sure what you mean by "forthcoming" 3.06GHz processor - Pentium IVs of this speed have been available here (UK) for ages and I've been running one for almost 9 months.

I'm no expert on hyperthreading, but I noticed early on that any running process would use a maximum of 50% CPU (I used to run SETI@Home). I assume this is a side-effect of a hyperthreading chip, but it does mean that you can run two copies of the client and get a much higher throughput. The difference is greater than the 5-10% quoted earlier. Although a single version of the client WILL crunch more gens on its own than when run in tandem with another copy, it comes nowhere near the total of the two versions together. From my experience I'd estimate you get something like 180% throughput when running two clients simulateously on a PIV chip.

Welnic
06-25-2004, 01:20 PM
My experience of running on an HT box at work is you get about an extra 10% running two clients with HT on compared to one client with HT off.

bwkaz
06-25-2004, 07:28 PM
Originally posted by Digger
Not sure what you mean by "forthcoming" 3.06GHz processor When that comment was posted, the chip still was forthcoming. The post date is August of 2002, which is 20 or 21 months ago.

:p

(Of course, the thread has been resurrected twice now -- once in July 2003 by erk, and once again just today by arcturius.)