P4 Hyper Treading in the real world

[moemac8]

How much does this help. I hate printing a large file, decoding a large .rar file or encoding .mp3s and having my entire system slow to a limp. Does HT really help you be able to do other things while such operations are going on?? Was hoping someone who actually has a P4 system will tell me their experiences. If it does help alot then I may skip the AMD thing and go ahead and build a new P4 system instead of waiting for the 64bit AMD.

Thanks

 

[JonnyBlaze]

what speed / type cpu are you running now?

 

[Ogo]

Yes, HyperThreading makes a huge difference in running multiple cpu-heavy apps at once. Things that used to literally freeze my system until they were done now are hardly noticeable.

 

[Hanzou]

How do the Bartons compare to the P4's with hyperthreading in terms of multitasking? I know that the P4 will be better but by how much?

 

[Quackmaster]

Originally posted by: Hanzou
How do the Bartons compare to the P4's with hyperthreading in terms of multitasking? I know that the P4 will be better but by how much?

answer- "a lot". But the Bartons feel "snappier". When I'm running multiple programs (for example: playing a DVD, while using MS office, while running a web browser, at the same time I Encode an mpeg to AVI, while also navigating Windows) the P4 rules the roost by a hugely noticeble margin. Response to my mouse clicks and keyboard strokes is quicker on the Intel. But if I'm not heavily multitasking, or I'm just navigating Windows while running two or 3 less intensive tasks (example: exclude the video encoding from my previous example), then the Barton just owns. It's like it's quicker to the finish line in a 400 meter relay vs the Intel being better in a 5k distance race packed with runners everywhere.

Personally I love this because it's really starting to show the differences in the P4 (GHz and bus speed is everything) architecture vs the Athlon's (IPC is what counts). This will all change as Intel ramps it's clocks up and AMD goes to single-core multiprocessing with the Athlon64, but till then it's quite fun to play with!

 

[OverVolt]

Two Raptors in RAID 0 fixed that "snappines" for me on my P4 !

 

[dakels]

someone in another forum mentioned thattheir machine is a 2.4 P4 OC'd to 3.3ghz ( assume its the 400 fsb model). They also said they had HT enabled. How is that? Does that do anything? Is it possible for me to make my 2.66ghz P4 use HT (Intel 845PE HT capable)? I would assume there must be some huge downsides to this or little effect or I'd see posts all over the place that people are enabling HT with non HT ready cores.

 

[Tom]

Only the 800fsb 2.4c has Hyperthreading. That is probably what you read about. You can't enable HT in an older processor. Maybe Intel will start shipping 533 and 400 fsb processors with HT but I doubt it.

 

[Thor86]

Originally posted by: moemac8
Does HT really help you be able to do other things while such operations are going on??

YES!

 

[Jeff7181]

Look at the benchmarks yourself... link... to see if it's an improvement

Content Creation = no
General Usage = no
Divx encoding = yes
MP3 encoding = not so much, maybe a tad
Archiving performance = no, probably not
Gaming = no, probably not
3D Rendering = yes, most of the time
High end workstation = no, probably not

I say probably not because the charts do show an increase in performance, but it looks about the same from 2.8-2.06 Ghz and it does from 2.53 to 2.8 Ghz... so the increase in performance in most looks like it's just due to the increase in clock speed in that particular test.

The thing that's not tested is multi-tasking. I've read where people say they can leave a divx encode running, and go play Quake 3 Arena and not notice that their CPU is busting it's balls encoding a divx movie at the same time.

 

[dakels]

Originally posted by: dakels
someone in another forum mentioned thattheir machine is a 2.4 P4 OC'd to 3.3ghz ( assume its the 400 fsb model). They also said they had HT enabled. How is that? Does that do anything? Is it possible for me to make my 2.66ghz P4 use HT (Intel 845PE HT capable)? I would assume there must be some huge downsides to this or little effect or I'd see posts all over the place that people are enabling HT with non HT ready cores.

sorry I drank too much. He put a 2.4C and I missed that C. I didnt realize they made a HT processor that low of a ghz. Thought it wsa just the 2.8 and 3.0.

 

[dakels]

Originally posted by: Jeff7181
Look at the benchmarks yourself... link...">http://www.anandtech.com/cpu/showdoc.html?i=1746[/S</a> to see if it's an improvement

Content Creation = no
General Usage = no
Divx encoding = yes
MP3 encoding = not so much, maybe a tad
Archiving performance = no, probably not
Gaming = no, probably not
3D Rendering = yes, most of the time
High end workstation = no, probably not

I say probably not because the charts do show an increase in performance, but it looks about the same from 2.8-2.06 Ghz and it does from 2.53 to 2.8 Ghz... so the increase in performance in most looks like it's just due to the increase in clock speed in that particular test.

The thing that's not tested is multi-tasking. I've read where people say they can leave a divx encode running, and go play Quake 3 Arena and not notice that their CPU is busting it's balls encoding a divx movie at the same time.

Now this is just for single processor HT not real dual processer. How does HT decide where the lion's share of CPU resources go to? Like if you are running Quake and decoding a movie in the background? Say Quake needs about 75% of the CPU resources for playing the game, then does the HT technology spread the other remaining 25% to the background video decoding? Or does it just unintelligently chop the processes in half giving quake only 50% and the video 50% starving quake's performance? Also, what controls the multithreading? The OS? I would assume then that a OS thats more attuned to multiprocessing/multitasking would work better then one that is not.

I could only see single processer HT to be valuable for tasks that don't require 100% CPU usage from that single task at that moment. Otherwise, you are starving one or the other. You only have the same amount of clock cycles as a non-HT single processor, if a process you do maxes it, then HT or not, it's not going to help. It's not DP. Just looks like better multitasking to me. So then, why is this a hardware feature and not a software/OS feature? From what I know they've been fine tuning this with Mac OS X, based on Unix/Linux who've been doing OS controlled multitasking.

 

[Jeff7181]

It uses different parts of the processor... in an over simplified way... the right half is running quake 3, and the left have is encoding the divx movie.

 

[sonoran]

Originally posted by: dakels
Now this is just for single processor HT not real dual processer. How does HT decide where the lion's share of CPU resources go to?

You could set the lower priority process to a lower priority using task manager. The OS would then ensure it gets fewer time slices.

You only have the same amount of clock cycles as a non-HT single processor, if a process you do maxes it, then HT or not, it's not going to help. It's not DP.

And this is where you fail to understand hyperthreading. Imagine you're in the checkout line at the grocery. There's one cashier, and the old lady right in front of you being rung up just decided to spend the next 5 minutes looking through her book of coupons. You're stuck waiting on her. Now what if you could just shove her aside, step right up to the cashier, and get checked out? There's still one cashier working - for the same amount of time, but there is more work getting done. THAT is the nature of hyperthreading. The CPU is actually prepping instructions from more than one process at the same time. When one process stalls, it can immediately start executing the other.

 

[sonoran]

Originally posted by: dakels
I could only see single processer HT to be valuable for tasks that don't require 100% CPU usage from that single task at that moment.

BTW, your observation about a single task running at 100% is probably pretty accurate. That's why you don't see a huge boost in most benchmarks - they're maxing out the CPU with a single task that's not stalling out, thus they don't show much improvement. What I'm reading from people out in the real world though is that it makes for a much more responsive machine whenever they want to do more than one thing at a time.

 

[moemac8]

JohnnyBlaze, I have this

AthlonXP 1800
GF4 ti4600
512 MB DDR ram

 

[Duvie]

Originally posted by: Jeff7181
Look at the benchmarks yourself... link...">http://www.anandtech.com/cpu/showdoc.html?i=1746[/S</a> to see if it's an improvement

Content Creation = no
General Usage = no
Divx encoding = yes
MP3 encoding = not so much, maybe a tad
Archiving performance = no, probably not
Gaming = no, probably not
3D Rendering = yes, most of the time
High end workstation = no, probably not

I say probably not because the charts do show an increase in performance, but it looks about the same from 2.8-2.06 Ghz and it does from 2.53 to 2.8 Ghz... so the increase in performance in most looks like it's just due to the increase in clock speed in that particular test.

The thing that's not tested is multi-tasking. I've read where people say they can leave a divx encode running, and go play Quake 3 Arena and not notice that their CPU is busting it's balls encoding a divx movie at the same time.

I totally agree Jeff....HT being still relatively new since until the launch of the p4c models it was relegated to INtels flagship and expensive 3.06ghz cpu...I think we just aren't seeing the type of reviews doing the things HT will take advantage of...IE multitasking!!! Only programs that are coded for Dual cpus may even show an increase with HT enabled and no multitasking. Dual cpus are better for that scenario. I think the other issue may be..HOw do we show this advantage?? How do we eqaute numbers to it ppl can get a sense of?? You almost got to see it to believe it type thing, as well as the feel will be variable to all users based on more then just hardware but software (apps running and OS) and it ability to take advantage of the HT...

 

[Jeff7181]

Bout the only thing I can think of is to maybe, encode some MP3's while you run benchmarks.

 

[Duvie]

I guess I could run a divx encode of a movie I have ran when it was the only thing utilizing the cpu and see what the precentage hit is on my FPS....Then each time bombard it with another app...

Apps I can think of that really like to use as much cpu as possible for as long:

Divx encoding
Mp3 endoding
MPeg2 encoding
Winzip or winrar files (archiving)
Cadd Rendering

I would say capturing from my AIW card but these systems are so fast now a days I can cpture 720x480 at DVD settings and not use more then 50% cpu. Palying DVDs takes less then 10% so it would need to be coupled with a lot of things but then just ends up sucking the ram down and then that has an effect on the paas I may be testing.

So maybe it is a question of numbers of cpu intensive programs and not just shear number of apps....

 

[Quackmaster]

Originally posted by: Duvie

these systems are so fast now a days I can cpture 720x480 at DVD settings and not use more then 50% cpu. Palying DVDs takes less then 10% so it would need to be coupled with a lot of things but then just ends up sucking the ram down and then that has an effect on the paas I may be testing.

So maybe it is a question of numbers of cpu intensive programs and not just shear number of apps....

I would agree with that absolutely. I tested my p4.2.4c @ 3.0GHz on an Asus p4p800 deluxe using ONE stick of Kingston Hyper X 3500. When limited by RAM, obviously a lot of hdd disk caching and swapping is going on (not to mention WinXP itself wanting the whole 256MB), but when I loaded both sticks and downclocked (underclocked, whatever ya wanna call it) the RAM to see the effect it would have on multitasking, the differences was not as great as you might think. The response time was slower, but not too bad because of MAM (PAT) and tight timings on RAM latencies, etc.

Also, hey overvolt. I too am running RAID. I have a pair of 200GB WD PATA (w/ 8MB cache each) in RAID 1 for safety, and another pair of 10GB stipped into a 20GB array for speed. But I still think the Barton/Nforce2 combos are "snappier" at lower speeds. But I fixed that by running my 2.4c @ 3.4GHz...rock stable!

 

[render]

It's nice to see 4 processors in windows task manager.(dual xeon setup)

 

[DSE]

Now this is just for single processor HT not real dual processer. How does HT decide where the lion's share of CPU resources go to? Like if you are running Quake and decoding a movie in the background? Say Quake needs about 75% of the CPU resources for playing the game, then does the HT technology spread the other remaining 25% to the background video decoding? Or does it just unintelligently chop the processes in half giving quake only 50% and the video 50% starving quake's performance? Also, what controls the multithreading? The OS? I would assume then that a OS thats more attuned to multiprocessing/multitasking would work better then one that is not.

Now, from my understanding of HT, the processor load is not the governing factor of how much improvement you get from HT. In fact, if you are running two processes that each only take up 50% of proc load, you probably won't see much, if at all, improvement from HT. The big idea behind HT is that a command/thread does not use all the resources of a CPU in a given clock cycle. HT makes use of the unused resources in the same cycle for the same clock cycle. So I would think the biggest determining factor on how much improvement you see might be the combination of processes that you run at the same time, like if you're running one process that does not (or rarely) use FPU while another that consistently hammers the FPU, then you would see great improvements, while if you tend to run programs that use the same resources, regardless of if you have HT or not, you won't see much of an improvement. This is all assuming limitations of memory bus, hdd performance, etc..