Hyperthreading - is it a big deal?

[Wisey]

Currently I am an Intel user and I am thinking of crossing over to AMD. I have read in several instances that AMD's lack of Hyperthreading does have an impact on multitasking and especially for a user like me who like to do encoding, surfing, type some work, check email and etc at the same time.

There is no benchmark to measure the effect of Hyperthreading but I am really keen to hear from users who have both Intel and AMD systems or just cross over to tell me if there is really a big difference when it comes to hyperthreading.

I am not much of a gamer and even if do play, I am not into the highest FPS.

Any comments are welcome ^^

 

[Vee]

How long have you used computers?
HT only exist on Pentium 4, and only on the P4C with 800FSB, 3.06GHz with 533FSB, P4E and P4F. So if you have any other Intel CPU or have ever had any other Intel CPU, you already know what a CPU lacking ht feels like. You also need WindowsXP with HT enabled, so again, if you've ever had any other OS, you already know what no ht feels like.

CPU lightweight things like surfing and editing some text, is IMO not really disturbed by background tasks, even if you lack ht.

There are benchmarks featuring multitasking. Veritest Winstone2004 and SysMark2004. Problem is their scores don't tell the subjective feel of this, and A64 does pretty good on them.

In some very special situations, it can be a big deal. But normally, I don't think it's any deal at all.

 

[aka1nas]

you will probably only notice it if you are doing two or more relatively stressfull tasks at the same time. For example if you are encoding a video in the background while playing a game. A HT cpu will handle this a bit better(but nowhere near as well as a real multi-processor system will). Otherwise, you actually do lose a tiny bit of performance with HT as there is some overhead involved. If you are mainly a gamer and not doing anything exotic, than you probably won't notice a big difference and shouldn't worry about HT. If you are a really really heavy multitasker, then you may want to skip HT and go straight to SMP. It all kinda depends on what you use your computer for mainly.

 

[Wisey]

I have a HT enabled CPU with WinXP, so I know how it feels. I just don't know how it feels like on a AMD64 without it...

 

[Jeff7181]

In some multi-tasking scenarios the Athlon-64 is as fast, or faster than a HT enabled Pentium 4. There are also scenarios where the Pentium 4 is faster than an Athlon-64, and usually when that happens it's by a significant margin, like 10-20%. Xbit Labs did some multi-tasking benchmarks a while back that show this.

*EDIT* The area where HT did the best was when a virus scan was run while doing other tasks. But there were other tests with less CPU intense tasks running simultaneously where the Athlon-64 did quite well.

Basically, if you absolutely MUST encode video while you play Half Life 2, you'll be better off with a HT enabled Pentium 4... and even better off with a dual CPU system... and even better off with two separate computers.

 

[Duvie]

that is the key...cpu usage....

Areas the HT of P4 will destroy Athlon64...

Folding and running and encoding at same time...On athloon64 system the encoding will take precedence and get done relatively as fast as it would as a single task....however virtually no free cycles are ever given to the FH project so virtually nothing is done.....Same thing on a P4 results usually in like 80-90% of encoding speed of normal but during that same time the FH project actually gets significant more work done. Now if you do this using TMPGenc which is multithreaded and probably is the best encoding pp for taking adavnatge of HT you will see a bigger hit in the encoding time because it was using the HT fully before. Now it will relinquish more of that...Still the encoding is like 70-80% of normal with FH app still getting quite a bit done....


Another thing is doing rendering app in a CAD program, Folding and then play a DVD....In A64 that video may stutter...In the P4 I never saw that stutter and once again the rendering was done quickly with little speed loss versus normal single task use and FH got work done!!!!

The A64 reacts a lot Barton core I tested.

The main thing to remember is HT has its advantage only because the disadvanatge of longer pipeline of the P4 has longer idle states. The A64 and Barton cores have much shorter pipelines so those idle states are not near as much. I bet the Athlons could use it but if it garnered more then 5% I would be surprised....

I wont blow smoke up your arse. I miss the P4 for some of the features. I however also enjoy this cooler running chip that at single tasks beats or smokes the p4 I had at 3.5ghz. So there I cant complain.

I may go dual opteron soon for the ultimate in multitasking. I just need to find the equipment to let me OC the piss out of them and get my money's worth...

 

[beatle]

I had the same question about 2 months ago. I had a P4C and I moved to an A64. I actually do several things at once that take up a fair amount of CPU: newsgroup header crunching, video encoding, some webbrowsing (some sites are murder!) and Winamp plugins. These tasks are in addition to the dozen other apps I use that don't require quite so many cycles. I was pleasantly surprised how fast the A64 was able to switch and manage the load with just Windows scheduling the processes.

 

[uOpt]

I use HTed P4s quite a bit. I never see more than 10-20% raw speedup on just running several CPU-intensive tasks.

However (and I posted a big thread about it, search for it), I found that for using a 3d game in the foreground while the CPU is loaded with video compression, compilation or the like hyperthreading is a huge advantage. I assume but did not verify that using Mozilla in the foreground while the background is munching might have similar benefits.

So, with Hyperthreading you won't get your CPU-intensive tasks finished much sooner, but they will likely stop harrasing your foreground session.

 

[Jeff7181]

Originally posted by: MartinCracauer
I use HTed P4s quite a bit. I never see more than 10-20% raw speedup on just running several CPU-intensive tasks.

However (and I posted a big thread about it, search for it), I found that for using a 3d game in the foreground while the CPU is loaded with video compression, compilation or the like hyperthreading is a huge advantage. I assume but did not verify that using Mozilla in the foreground while the background is munching might have similar benefits.

So, with Hyperthreading you won't get your CPU-intensive tasks finished much sooner, but they will likely stop harrasing your foreground session.

That's a good way of putting it. :beer:

 

[iversonyin]

Originally posted by: Wisey
Currently I am an Intel user and I am thinking of crossing over to AMD. I have read in several instances that AMD's lack of Hyperthreading does have an impact on multitasking and especially for a user like me who like to do encoding, surfing, type some work, check email and etc at the same time.

Just for encodin, you should get P4

but I think AMD get your best bang for the buck if you are on a budget.

 

[clarkey01]

Even a die hard fan of AMD like me has to hand it hyperthreading, I think if you offered HT in K8 ppl would jump for it. Its a great feature.

I used to say it was a marketing thing, but My 2.8 C can do alot more work (Bridge commander, Msn, CD burning, "file sharring") and i notice no lag. On my AXP 2000+ jesus, it just stops.

 

[AWhackWhiteBoy]

Originally posted by: iversonyin
Just for encodin, you should get P4

but I think AMD get your best bang for the buck if you are on a budget.

Encoding is subjective based on the software and codecs your going to use, why don't people understand this?

 

[AWhackWhiteBoy]

So far this thread is mostly just subjective opinion, "this seems faster than this, OMG ITS SO MUCH BETTER!"

Can we get this gentlemen some background information on the technology, and benchmarks so he can visually see what hes going to get?

 

[Jeff7181]

Originally posted by: clarkey01
Even a die hard fan of AMD like me has to hand it hyperthreading, I think if you offered HT in K8 ppl would jump for it. Its a great feature.

I used to say it was a marketing thing, but My 2.8 C can do alot more work (Bridge commander, Msn, CD burning, "file sharring") and i notice no lag. On my AXP 2000+ jesus, it just stops.

Hyper-Threading in the K8 really wouldn't do much good. It's the HUGE (relatively) pipeline of the P4, especially the Prescott, that makes Hyper-Threading so effective... arstechnica has a really good explanation of pipelining here that will give you the basics of why a stall in the pipeline effects processors with long pipelines so much more than short pipelines.

 

[clarkey01]

yeah I know the 12 stage pipeline wouldt allow HT to perform as well as in the 31 stage prescott pipeline.

I think G5's are 16-17 ? and they use it.

 

[frootbooter]

I've got an idea.

Wait until dual core athlon 64's are out, and then it won't matter because you'll have real SMP anyway.

 

[Jeff7181]

Originally posted by: clarkey01
yeah I know the 12 stage pipeline wouldt allow HT to perform as well as in the 31 stage prescott pipeline.

I think G5's are 16-17 ? and they use it.

IBM has a processor that's 12-14 stages and uses a form of Hyper-Threading... so it's not ALL about pipeline length... arstechnica has an article about that too explaining why that IBM processor with the short pipeline benefits from Hyper-Threading. I think it's safe to say that neither the Athlon-64 or Pentium M have the characteristics that would make them suitable candidates for Hyper-Threading.

 

[BespinReactorShaft]

I've only found it good for SETI@home multi-threading.

 

[uOpt]

Originally posted by: AWhackWhiteBoy
So far this thread is mostly just subjective opinion, "this seems faster than this, OMG ITS SO MUCH BETTER!"

Can we get this gentlemen some background information on the technology, and benchmarks so he can visually see what hes going to get?

Well, I posted a huge article with all kinds of benchmarks here a few months back.

Cannot find it anymore, is it possible that these things get pruned?

In any case, the hard numbers in there were the foundation of what I said above.

The longer pipeline in the prescott doesn't make HT more effective, benchmarks show about the same speedup as for Northwood.

 

[Jeff7181]

Originally posted by: MartinCracauer

Originally posted by: AWhackWhiteBoy
So far this thread is mostly just subjective opinion, "this seems faster than this, OMG ITS SO MUCH BETTER!"

Can we get this gentlemen some background information on the technology, and benchmarks so he can visually see what hes going to get?

Well, I posted a huge article with all kinds of benchmarks here a few months back.

Cannot find it anymore, is it possible that these things get pruned?

In any case, the hard numbers in there were the foundation of what I said above.

The longer pipeline in the prescott doesn't make HT more effective, benchmarks show about the same speedup as for Northwood.

You're aware that the cache size and transistor size wasn't the only thing that changed, right? There were many other changes to the core. That's one reason people are so impressed by the Prescott... it has a 50% longer pipeline that the Northwood, and shows almost no difference in performance. That says quite a bit about the work Intel did to lengthen the pipeline without destroying performance like with the P3 to P4 change.

 

[whorush]

i just posted this in another topic...

HT isnt the end all be all. it helps the P4 because its pipeline is like 34 stages or something. one branch misprediction and the whole thing is empty. big bubble. so since the whole huge pipeline was empty all the time they tricked the OS into giving it 2x as much data. if HT is the end all be all, why isn't it in the pentium M or the ITANIC? (not sure if i'm imagining this, but i think they may have talked about putting it in the itanic, but anyway.)

anyway, the hammer has a short pipeline, around 12 stages, so it wouldnt benefit from HT, in fact, it could hurt preformance.

that being said, A64s are very very fast, fast enough to win the vast majority of benchmarks and plenty fast to run a word processor play a cd surf the web and check your email.

 

[uOpt]

Again, Northwood and Prescott get about the same benefit out of Hyperthreading. Since the Prescott pipeline is much longer but doesn't make HT more effective you can rule out the long pipeline as the main factor for hyperthreading speedup.

Also, HT performance gets completely trashed, way below non-HT performance if you overload the L1 cache.

Both combined indicated that the bottlenack that HT works around is "feeding issues", i.e. caches and memory.

 

[Jeff7181]

Originally posted by: MartinCracauer
Again, Northwood and Prescott get about the same benefit out of Hyperthreading. Since the Prescott pipeline is much longer but doesn't make HT more effective you can rule out the long pipeline as the main factor for hyperthreading speedup.

Also, HT performance gets completely trashed, way below non-HT performance if you overload the L1 cache.

Both combined indicated that the bottlenack that HT works around is "feeding issues", i.e. caches and memory.

That article I linked to talks about that... HT in the P4 is more of a method of hiding latency to keep the pipeline full. But, the reason it's so important to keep the pipeline full is because it's so long.

 

[orion7144]

I had my AMD64 3000+ for 3-4 weeks before I had to go back to the P4. For me the HT enabled P4's make my life much easier (for what I do with them) so I had to give up the ooooooh factor of having a 64bit CPU. There virtually no cost difference in the two and for a few FPS I could not see any benefit in keeping the A64. I chalk the loss up to reading all the hype here and other places.

 

[Duvie]

Originally posted by: MartinCracauer
Again, Northwood and Prescott get about the same benefit out of Hyperthreading. Since the Prescott pipeline is much longer but doesn't make HT more effective you can rule out the long pipeline as the main factor for hyperthreading speedup.

Also, HT performance gets completely trashed, way below non-HT performance if you overload the L1 cache.

Both combined indicated that the bottlenack that HT works around is "feeding issues", i.e. caches and memory.

Actually I had always heard the prescott was doing a bit better in HT and was touted as being so.....

As Jeff says and I will agree the P4's HT is just a way to make the architecture mpre eefificent to overcome the huge latency penalties it takes from an enlarged pipeline and branch mispredictions. That aside it comes through and has a side benefit that it can help when 2 or more heavily cpu intensive apps work.

I have repeated have said the p4 wasa better multitasker at things I did as well. I agree with Orion, but point out I do less encoding these days and there is simply nothing faster in single task use in myu encoding and CAD applications then this A64 I have now. I was in limbo cause I di not want to go to the prescott which IMO was 1 step forward 2 step back architecture. I got the sckt 939 and A64 for more of the future. If Intel' srumors of its dual core cpus being power hungry prescott cores..I MADE THE RIGHT CHOICE!!!