P4's HT

[slurmsmackenzie]

to make this short, at any given moment, i could find myself doing any one of those things while gaming. of course, there are other things, but, would the the HT really shine against the 939 when multitasking in this fashion? would i see a drop off in game performance or encoding speed if both were running at the same instance? in what instances does hyperthreading really show it's strength?

 

[beatle]

Burning media hardly uses any CPU cycles. Encoding taxes the CPU more and the P4 would come into its own if you decide to game while encoding.

 

[slurmsmackenzie]

so in that respect are we talking a big performance loss in contrast with the 939? is this something that would be sustantially noticable.

 

[Accord99]

Originally posted by: slurmsmackenzie
to make this short, at any given moment, i could find myself doing any one of those things while gaming. of course, there are other things, but, would the the HT really shine against the 939 when multitasking in this fashion? would i see a drop off in game performance or encoding speed if both were running at the same instance? in what instances does hyperthreading really show it's strength?

The main benefit of HT is that with two CPU intensive applications, the total combined throughput exceeds that one just running one applications. So in a normal processor, running two equal priority CPU intensive app will result in each app running at 50% of the performance compared to the app running by itself. By comparison, on a HT processor, each app will run at anywhere from 58%-65%, giving you a combined throughput of 115%-130% (occasionally less, ocassionally more). Then it becomes a question of whether the HT boost is enough to ovecome the A64's typical single threaded advantage.

There is also a question of smoothness that a thread or two has talked about, which may be related to the fact that a non-HT processor can only process one thread at a time and does so for 10-120ms before switching. HT processors can have instructions from two threads in the pipeline.

The one major limitation of HT is that thread priority is not well respected, in the sense that a foreground high priority thread will share the CPU even with a background low priority thread, such as distributing clients. You would have to manually disable the background thread to regain full single-thread performance.

 

[uOpt]

slurmsmackenzie, I have measured exactly what you ask and posted here:
http://forums.anandtech.com/me...8&threadid=1426699

Feel free to bump that thread if you have questions.

 

[slurmsmackenzie]

thanks, marty!

so essentially, what you're saying is that a P4 with hyperthreading will handle video encoding and light gaming and make both run at acceptable rates, whereas the A64, while optimal for running these tasks individually, will see it's productivity decrease substancially in that particular multitasking scenario. correct? so, is the performance difference i'll see with the a64 in single cpu intensive workload negligable compared to the p4? would i be better to get the A64 and be selective about my encoding thereby assuring a level of upgradability (64bit os and encoding proggys... whenever that'll be not to mention sli at this point)? i'm really torn at this point and i'm trying to weed out the fanboys. i like your "numbers" approach. lol thanks for the info.

essentially.. in the long run is a platform difference really gonna matter for any situation, including hyperthreading outside of benchmarks and the absolutely scrutinous (is that a word?) among us? or should i get the a64 save a buck or two and encode while i sleep? thanks again.

 

[beatle]

How often do you upgrade? 64-bit isn't really here yet, IMO, so unless you keep your system for a long time, the upgrade path of the platform shouldn't be a concern.

You might also see my thread on HT. It seems to be a mixed bag of opinions, however.

I think I may break down and buy an A64 next time. I do game and encode at times, but I'll just set the priority of the encoder at idle so I shouldn't have a problem. If the encoding finishes a few minutes later, I'll just have to deal with it.

 

[uOpt]

Originally posted by: slurmsmackenzie
thanks, marty!

so essentially, what you're saying is that a P4 with hyperthreading will handle video encoding and light gaming and make both run at acceptable rates, whereas the A64, while optimal for running these tasks individually, will see it's productivity decrease substancially in that particular multitasking scenario. correct? so, is the performance difference i'll see with the a64 in single cpu intensive workload negligable compared to the p4? would i be better to get the A64 and be selective about my encoding thereby assuring a level of upgradability (64bit os and encoding proggys... whenever that'll be not to mention sli at this point)? i'm really torn at this point and i'm trying to weed out the fanboys. i like your "numbers" approach. lol thanks for the info.

essentially.. in the long run is a platform difference really gonna matter for any situation, including hyperthreading outside of benchmarks and the absolutely scrutinous (is that a word?) among us? or should i get the a64 save a buck or two and encode while i sleep? thanks again.

I think you got the wrong conclusion.

My conclusion was that HT only pays off when you do a bus-intensie application (read: game) in the foreground while there is a CPU hog like video encoding or compilation in the background.

I found HT to be pretty much useless for just ding several compilations or several video encoding at once, it is very hard to even get 20% speedup out of it.

There are some isolated things that improve, e.g. I have a bzip2 decoder which is threaded and that one can decode two bzip2 files at the same time which gets up to 38% improvement, but this is so rare.

The AMD64 is slight worse than non-hyperthreaded P4

Playign with the priorities for the processes in the scheduler won't help much with the game+encoding situation. It would help with two similar things, but the bus-bound game won't be under normal control of the scheduler.

 

[slurmsmackenzie]

Originally posted by: MartinCracauer

Originally posted by: slurmsmackenzie
thanks, marty!

so essentially, what you're saying is that a P4 with hyperthreading will handle video encoding and light gaming and make both run at acceptable rates, whereas the A64, while optimal for running these tasks individually, will see it's productivity decrease substancially in that particular multitasking scenario. correct? so, is the performance difference i'll see with the a64 in single cpu intensive workload negligable compared to the p4? would i be better to get the A64 and be selective about my encoding thereby assuring a level of upgradability (64bit os and encoding proggys... whenever that'll be not to mention sli at this point)? i'm really torn at this point and i'm trying to weed out the fanboys. i like your "numbers" approach. lol thanks for the info.

essentially.. in the long run is a platform difference really gonna matter for any situation, including hyperthreading outside of benchmarks and the absolutely scrutinous (is that a word?) among us? or should i get the a64 save a buck or two and encode while i sleep? thanks again.

I think you got the wrong conclusion.

My conclusion was that HT only pays off when you do a bus-intensie application (read: game) in the foreground while there is a CPU hog like video encoding or compilation in the background.

I found HT to be pretty much useless for just ding several compilations or several video encoding at once, it is very hard to even get 20% speedup out of it.

The AMD64 is slight worse than non-hyperthreaded P4.

______________________________________________________

that's exactly what i'm talking about. i'm not particularly interested in encoding two full movies at once, but 1 encode and gaming is something i could definitely see myself doing. i could get alot more done. so was i off? you need to dumb stuff down a bit for me, if you would be so kind.

when you say A64 is worse (and give it it's own paragraph), am i to assume that you mean on the whole? could you elaborate on that, please?

 

[LTC8K6]

Duvie's HT testing thread might interest you.

 

[uOpt]

Originally posted by: slurmsmackenzie
when you say A64 is worse (and give it it's own paragraph), am i to assume that you mean on the whole? could you elaborate on that, please?

I found that the AMD64 does the two-cpu-hogs things as good as the Pentium-4 when not using HT.

However, if you do game-and-cpu-hog then the P4 was slightly better even when not using HT. But keep in mind that this setup was only usable with HT anyway. Neither the P4 without HT nor the AMD64 would run the game well enough while there is a CPU hog in the background. I don't think scheduler fiddling will fix this. This is for Doom3, maybe other games are better, but I kind of doubt it, the moment you have a real 3D game with lots of textures things probably degrade quickly without HT even with lighter games.

Also keep in mind that turning on HT lead to a slight slowdown for some of the tasks that were not using multiple parallel threads or processes but lots of new threads and disk/io, such as C compilation.

A quick test indicated that turning on the preemptive scheduling option in the Linux kernel solved this, having quick sequences of single tasks being equally fast with HT than without. However, I didn't take these numbers in a controlled enough environment so I need to re-run the tests to be sure.