hyperthreading or MCH?

[SiiRuHs]

I've notice Kyle, on several occasions, notes how smooth Intel systems are with hyperthreading and I'm curious to know if this might not be due to Intel's chipsets? The MCH's seem to lay pretty good beat down on everything else out there, including nVidia (yes I'm ignoring video performance). Anyone have any opinions on this? Maybe try the system without HT on or use another chipset which does support HT?

 

[Dan_D]

I don't know how the other chipsets compare. HT is a hard thing to benchmark.

 

[Merlin45]

it is easy to compare, because you can just disable HT and you have an identical machine w/o ht. I can say from experience that with HT disabled the machine benchmarks a tiny bit slower, but in multitasking (which I do alot) the machine is alot more responsive.

 

[emorphien]

I agree merlin. A lot of people (who don't use Intel) talk about how HT is a waste, or a solution to a failure. In some senses that may be correct (long pipeline) but it does help improve responsiveness. Overall functionality and speed isn't really much faster with it on, but responsiveness during multitasking improves considerably at times.

 

[pxc]

I can say from experience that with HT disabled the machine benchmarks a tiny bit slower, but in multitasking (which I do alot) the machine is alot more responsive.

It's not just multitasking, but many demanding apps are multi-threaded, like encoding apps. But I agree that with single threaded benchmarks (especially games), HT makes very little difference overall. And if it helps out with multitasking, that's also great.

Since it looks like the only EM64T (x86-64) processors Intel is making are Xeons and Prescotts in the next year or more, I'll probably be moving to an Athlon 64 later this year. I'll miss you, HT!!!!

 

[emorphien]

It's not just multitasking, but many demanding apps are multi-threaded, like encoding apps. But I agree that with single threaded benchmarks (especially games), HT makes very little difference overall. And if it helps out with multitasking, that's also great.

Since it looks like the only EM64T (x86-64) processors Intel is making are Xeons and Prescotts in the next year or more, I'll probably be moving to an Athlon 64 later this year. I'll miss you, HT!!!!

Perhaps you'll have dual core by the time you're buying. Or just go dual CPU. When I build my workstation I want dual CPUs with dual cores! woopeee!

 

[mc_P]

then whats that hyper transport crap on the athlon 64's do?

 

[corrosive23]

then whats that hyper transport crap on the athlon 64's do?

Its a direct link between chips on the board.

 

[ShePearl]

It's not just multitasking, but many demanding apps are multi-threaded, like encoding apps. But I agree that with single threaded benchmarks (especially games), HT makes very little difference overall. And if it helps out with multitasking, that's also great.

I'd like to know why game developers don't take advantage of Hyperthreading ?

I don't know, I maybe misinterpreting how Hyperthreading works. But can't developers make their engine to take advantage of Htperthreading ? For example, like run phycisc calculations in one thread, while running AI in another thread ? This way, I think gaming performance of Pentium4 will increase, although Athlon64 may still beat P4.

Seriously, currently AMD kills Intel when it comes to gaming. AMD is in just different league, gaming performance-wise. And I think this is where Intel is losing enthusiasts. Just about every PC gamer prefers AMD over Intel. I think Intel needs to do something about this.

 

[Merlin45]

most games aren't designed to take advantage of SMP, hence SMT doesn't help much. It takes more time to design an engine for what was not too long ago, a very small market. (before HT simultaneous multithreading was only possible on SMP systems, which aren't too common) the exceptions are quake3, but SMP support was broken in patch 1.28 IIRC, and not fixed. Doom3 will have SMP support, so it will get a boost from HT.

 

[SiiRuHs]

Problem with multithreading games is many aspects of the game make use of the same components of the CPU. Hyperthreading as I know it allows the CPU to use unutilized resources (i.e. pipes). If you're doing 2 stressful and dissimilar processes you'll see a boost but if the processes are similar (e.g. FPU intensive) HT won't help.
Thanks for the response. I was wondering if dualies or HT would be worthwhile or if I should just look for an intel based mobo. I'm in no rush to buy, it was just a curiosity.

 

[SLee]

Problem with multithreading games is many aspects of the game make use of the same components of the CPU. Hyperthreading as I know it allows the CPU to use unutilized resources (i.e. pipes). If you're doing 2 stressful and dissimilar processes you'll see a boost but if the processes are similar (e.g. FPU intensive) HT won't help.

That's not true at all. It depends on the applications you're using, otherwise using two instances of SETI would not produce the ~30% increase in throughput that is seen on the P4 with Hyperthreading.

 

[Spack]

The big problem with Intels implementation of SMT is that the cache is not seperated between the threads. This means is one of the threads going through the core is monopolising the core, then the other thread stalls or if one threadline is using too much of the cache it can cause the other thread to starve due to how the schedular works. In many OS and office apps, a hyperthreaded P4 feels smoother doing multiple things than a non-HT cpu. For most intensive apps, there really isn't that big of a difference. In fact, apps that have specifically have been written for SMP can take a huge hit because they see the P4 as a true SMP system and not just a multithreading single CPU. Also, it is really difficult to program (at least for big gains) multithreaded apps. The biggest gains the P4 sees over the Athlon 64 right now is from sheer clock speed of its SSE2 (the Hammer core really needs a 128 bit wallace tree), but HT can be nice in day to day use.

 

[shaihulud]

it is very interesting to note how relative things are to people. in general rule of science the human race is general inobservant. when a person sees what should or should not be they are truly amazed and sometimes changes the paradigm of that individual.

anyway, when on my athlon 700mhz system back in the day i alt-tab from quake 3 to desktop to search for a map and download it at a faster speed than the server can upload. after getting the map name i switched to the desktop and notice a slow down in browsing. i was upset when i looked at my system monitor to see that the cpu is at 100%. i was like, this is a super scalar processor and it can not handle this at all, at 700mhz??!! this was exhibited on all the athlon i had up to the 2700+ until i purchased a p4-3c. it was nice to finally alt-tab to the desktop in multiple games and not have a slow down. it does and can make a difference. even as simple going to the desktop and doing a small task.

gaming companies are going to make use of ht. there has been multiple interviews on this subject. even if not certain os functions and do not forget drivers. intel's ethernet driver for example can make use of ht. ht is nice and i do believe will push us quicker and smoother into dual cores. personally, i think this was the intent. even if this was not so it inadvertantly did.

 

[USMC2Hard4U]

I dunno, I happen to beleive that Hyperthreading rocks.... when I built my current system (in sig) HT was turned off in the Bios by default and I didnt even know.. I was using my PC as usual and thoguht the peformance, resonsivmess was meh.. about average...

it was all good until I turned on HT (had to format again) but whatever, but woah, I instanly noticed a difference. in everything I do, it just seems snappier... thats all....

 

[SLee]

The big problem with Intels implementation of SMT is that the cache is not seperated between the threads. This means is one of the threads going through the core is monopolising the core, then the other thread stalls or if one threadline is using too much of the cache it can cause the other thread to starve due to how the schedular works.

No, the P4 in HT mode partitions the caches and instruction buffers so that this does not occur.

In many OS and office apps, a hyperthreaded P4 feels smoother doing multiple things than a non-HT cpu. For most intensive apps, there really isn't that big of a difference.

Running intensive multiple or multi-threaded apps are where the big differences are seen, since simple apps don't push the P4 much for any user to notice.