Is Dual-Core Better Than Hyperthreading?

[ptabor]

So what's the difference between these two technologies? seems to me they're almost the same, no?

 

[chroot]

They are not at all the same.

A hyperthreaded processor has the same number of function units as an older, non-hyperthreaded processor. It just has two execution contexts, so it can maybe achieve better function unit utilization by letting more than one program execute concurrently. On the other hand, if you're running two programs which compete for the same function units, there is no advantage at all to having both running "concurrently." When one is running, the other is necessarily waiting on the same function units.

A dual core processor literally has two times as many function units as a single-core processor, and can really run two programs concurrently, with no competition for function units.

- Warren

 

[GluonZ]

HyperThreading was an attempt to do what dual-core is doing. It was only 'faking' it though, whereas dual-core is the real thing. And, in all honesty, HT never did much. Hell, some Windows applications recommended turning it off!
Basically what Chroot said was correct, and Intel has actual dropped (for now) HT on the new Core models, as dual-cores have been achieved now.

 

[MeJennifer]

A dual core processor literally has two times as many function units as a single-core processor, and can really run two programs concurrently, with no competition for function units.

But there is some competition with regards to memory access.
For instance the AMD dual cores are better optimized for this than the current Intel processors.

 

[Anttech]

But there is some competition with regards to memory access.
For instance the AMD dual cores are better optimized for this than the current Intel processors.

Source please?

I have a dual core 2.0Ghz with 2 gig Ram runing osX 10.4. I have Umbuntu and XP virtually running on top, with only a small performance hit. I use a LOT of RAM doing this however.

I don't think I could have done that with hyperthreaded proc

 

[nmtim]

Source please?

I have a dual core 2.0Ghz with 2 gig Ram runing osX 10.4. I have Umbuntu and XP virtually running on top, with only a small performance hit. I use a LOT of RAM doing this however.

I don't think I could have done that with hyperthreaded proc

I thought MeJennifers comparison was between current AMD dual cores and Intel dual cores, not between a dual core and an HT single core.

I think AMD still holds the edge in memory bandwidth with its on-die controller, versus Intel's FSB. This is huge as you scale beyond 2-4 cores/die: it doesn't matter how many operations per clock you can do if you're waiting for operands. AMD's Hypertransport scales better than FSB.

There was a huge spate of head-to-head comparisons when Woodcrest was released. Poke around arstechnica or slashdot for some links.

Cheers,
Tim

 

[Anttech]

Hmmm maybe it came across as if I was doubting her comparison, I wasnt, I was just wanting the source for that information.

 

[AO eye 5]

I suppose MeJennifer is referring to Core microarchitecture, in relation to the AMD integrated memory controller.

It would help to identify the technologies also, the particular pentium 4 with hyperthreading are Netburst microarchitecture, AMD is K8 microarchitecture, and the latest intel microarchitecture is Core -- as in Core 2 (Conroe, Merom, Woodcrest, Allendale)-- Although it seems Core "mobile" yonah is not regarded to have Core microarchitecture.
Sorry about the general use of "Core," I have heard of intel to use single processors in Core 2--much in the way as mobile Core yonah (Core Solo, with Solo as single core and Duo for dual core), with one processor core disabled in the dual core die giving a Core Solo CPU.

Correct my errors as necessary, and it would be helpful to refer to hyperthreading as a type of SMT (simultaneaous multithreading).

EDIT: I am uncertain of which AMD CPU's are K8 microarchitecture. i.e. post Athlon 64?

Corrected SMT also...

 

[rcgldr]

It's my understanding that the Intel Dual Core CPU's share the cache, which eliminates the logic required to invalidate one CPU's cache when another CPU modifies memory with a non-shared cache.

 

[nmtim]

It's my understanding that the Intel Dual Core CPU's share the cache, which eliminates the logic required to invalidate one CPU's cache when another CPU modifies memory with a non-shared cache.

Core still needs some synchronization methods to maintain coherence and avoid race conditions.

A potential bonus of the shared cache is that one thread could take more than half the cache if needed. A potential pitfall is if two threads, for example two data-hungry, number-happy scientific threads, both want more than half of the cache and start evicting each other. Presumably Intel has some sort of damping term in its cache algorithms to minimize this.

Core has some radical new thinking in the cache. Time will tell how much that benefits scientific work loads.

 

[chroot]

The algorithms for cache-coherent non-uniform memory access have been in existence for many years. SGI has had such technology in their supercomputers for something like 15 years now.

It's nothing really new, but it's new to the desktop PC scene.

- Warren

 

[MeJennifer]

The algorithms for cache-coherent non-uniform memory access have been in existence for many years. SGI has had such technology in their supercomputers for something like 15 years now.

For 15 years? I would never have characterised them as supercomputers.
They just did the graphics fast.

 

[chroot]

For 15 years? I would never have characterised them as supercomputers.
They just did the graphics fast.

What are you, a joker? They make the finest supercomputers on the planet. The fourth largest supercomputer in existence right now is an SGI machine with 10,160 processors.

http://www.top500.org/lists/2006/06

It's too bad their marketing always stank, because they literally made the best computers on the market.

- Warren

 

[MeJennifer]

What are you, a joker? They make the finest supercomputers on the planet. The fourth largest supercomputer in existence right now is an SGI machine with 10,160 processors.

Yes right now, but certainly not 15 years ago!
No need to ask if I am a joker.
See for a history of SGI for instance http://en.wikipedia.org/wiki/Silicon_Graphics"

 

[chroot]

Yes right now, but certainly not 15 years ago! Some years ago they reinvented themselves, rather than calling people names it would be nicer to study the history of the company.
See for instance http://en.wikipedia.org/wiki/Silicon_Graphics"

Thirteen years ago, on the first Top500 list, Cray held the #9 and #10 positions. http://www.top500.org/lists/1993/06 As you know, Cray was later purchased by SGI.

These machines used essentially the same technology, called ccNUMA, by SGI in later years.

Like I said originally, cache-coherent non-uniform memory access architecture has been around ~15 years. Please keep in mind that I used to work for SGI, and don't really have much use for a page on Wikipedia.

- Warren

 

[advitiyas]

Thanks for such an extensive info...Dual-core is definitely better than HT