8 Physical cores vs 4 Physical 8 Threads?

[HAL9000]

Hey guys,
Just curious to know what would perform better 8 Physical Cores no HT, or 4 Physical cores with HT i.e. 8 threads both. (assuming same clock speed and other considerations)

 

[ModestGamer]

if the workload is large in size 8 cores will beat 4v all day, in fact alot depends on the threading technuque used as well.

 

[MrWizzard]

Assuming you could actually take full advantage of all the cores the 8 would.

 

[HAL9000]

Splendid thanks guys, I'm talking mainly about video encoding so I imagine I would see use on all cores, what about 6 cores with 12 threads vs 8 physical? thanks in advance for the help

 

[ilkhan]

8c will always beat 4c+HT. For 6c/12t vs 8c, thats a harder question, and would depend on the exact workload.

 

[HAL9000]

8c will always beat 4c+HT. For 6c/12t vs 8c, thats a harder question, and would depend on the exact workload.

Fair enough I wasz just wondering what the theory was like. I.e does an additional thread add 20% 50% 100% per core?

 

[Makaveli]

Like others have said it depends on the workload. If you have an application that can take advantage of HT it will give you a nice boost however its not 100% efficient. So eight real cores will always do better than 4c+HT.

If i remember correctly HT gives you something like a 20-30% on applications that can use it.

 

[ModestGamer]

Like others have said it depends on the workload. If you have an application that can take advantage of HT it will give you a nice boost however its not 100% efficient. So eight real cores will always do better than 4c+HT.

If i remember correctly HT gives you something like a 20-30% on applications that can use it.

only if the workload is light and the pipeline is underutilized. when the workload is heavy and the pipeline is fully utilized, its not better then having 4 real core and in some cases can actually cuase a drop in performance as resource and memory contentions can cuase performance fallout, even though it is better then the p4 days it still happens.

 

[HAL9000]

only if the workload is light and the pipeline is underutilized. when the workload is heavy and the pipeline is fully utilized, its not better then having 4 real core and in some cases can actually cuase a drop in performance as resource and memory contentions can cuase performance fallout, even though it is better then the p4 days it still happens.

So then theoretically with a heavy workload 8 physical cores without HT would perform better than 6 physical cores with?

 

[ModestGamer]

So then theoretically with a heavy workload 8 physical cores without HT would perform better than 6 physical cores with?

in a high throughput situation where you needed to fully utilized the pipeline, absolutely.Now if you running MS word and you have alot of threading to keep the app from blocking, the ht might outperform "still running word" in a single application, if you encoding a dvd in the background and its using a high throughput highly thread execution engine. Most likely ht might get a advantage "for the secondary application" but it depends entirely on the pipeline.

AMD is correct on one front. HT is not really anything but a trick that could be accomplished with other methods. real cores matter at some point.

 

[HAL9000]

in a high throughput situation where you needed to fully utilized the pipeline, absolutely.Now if you running MS word and you have alot of threading to keep the app from blocking, the ht might outperform "still running word" in a single application, if you encoding a dvd in the background and its using a high throughput highly thread execution engine. Most likely ht might get a advantage "for the secondary application" but it depends entirely on the pipeline.

AMD is correct on one front. HT is not really anything but a trick that could be accomplished with other methods. real cores matter at some point.

Cool, well i'm thinking in the situation that your using the computer for a single high priority process with little running in the background to encode a high definition video or DVD

 

[ModestGamer]

Cool, well i'm thinking in the situation that your using the computer for a single high priority process with little running in the background to encode a high definition video or DVD

with a application like handbrake that can really push the cpu to its limits, 8 cores will beat 6 with ht all day provided the cores are of equal performance clock for clock.

 

[HAL9000]

with a application like handbrake that can really push the cpu to its limits, 8 cores will beat 6 with ht all day provided the cores are of equal performance clock for clock.

Splendid I made the right choice, thanks

 

[JFAMD]

I can't speak to client loads but it is ~14% on integer and ~20% on server if I remember correctly.

HT really only works well for lightly threaded workloads with a lot of gaps in the pipeline. Heavy threads and highly efficient software don't work well.

 

[HAL9000]

I can't speak to client loads but it is ~14% on integer and ~20% on server if I remember correctly.

HT really only works well for lightly threaded workloads with a lot of gaps in the pipeline. Heavy threads and highly efficient software don't work well.

Ok great thanks for the info, as I do a lot of video encoding I made the right choice!

 

[IntelUser2000]

Can I ask what CPUs are you comparing? 8C/8T being better than 4C/8T is applicable if the architecture is close enough. If you are using say a dual Harpertown(Xeon based on Core 2 Penryn) and comparing that to the Core i7 quad core, then its hard to say. In some rendering applications the Core i7 4 core outperforms the dual 4 core Xeon.

 

[HAL9000]

Can I ask what CPUs are you comparing? 8C/8T being better than 4C/8T is applicable if the architecture is close enough. If you are using say a dual Harpertown(Xeon based on Core 2 Penryn) and comparing that to the Core i7 quad core, then its hard to say. In some rendering applications the Core i7 4 core outperforms the dual 4 core Xeon.

As it happens those are exactly the processors I was comparing, dual Xeon quads vs a single i7 strictly for batch video encoding in things like handbrake

 

[biostud]

The more physical cores, the less the advantage HT provides.

 

[Massive Overkill]

Can I ask what CPUs are you comparing? 8C/8T being better than 4C/8T is applicable if the architecture is close enough. If you are using say a dual Harpertown(Xeon based on Core 2 Penryn) and comparing that to the Core i7 quad core, then its hard to say. In some rendering applications the Core i7 4 core outperforms the dual 4 core Xeon.

As it happens those are exactly the processors I was comparing, dual Xeon quads vs a single i7 strictly for batch video encoding in things like handbrake

Most peeps in this thread are assuming you were asking about same-architecture. Latest gen processors do a hell of alot more work per CPU/cycle than older generation CPUs. Apples-to-apples. Some of the newer CPU's also have instruction sets that may or may not speed up your app that the older stuff doesn't have.

You also have to consider that dual-socket server boards aren't exactly cheap (x58 boards can get up their in price too, but are generally cheaper). Also you can easily OC an i7 to 3.8~4.0 Ghz to squeeze free extra-power............can't do that with a server board.

6 cores vs 4:

http://www.anandtech.com/show/3674/amds-sixcore-phenom-ii-x6-1090t-1055t-reviewed/1

 

[Cogman]

The more physical cores, the less the advantage HT provides.

Not necessarily true. It really depends on the application being threaded. x264, for example, actually sees some real improvements from hyper threading.

 

[HAL9000]

Most peeps in this thread are assuming you were asking about same-architecture. Latest gen processors do a hell of alot more work per CPU/cycle than older generation CPUs. Apples-to-apples. Some of the newer CPU's also have instruction sets that may or may not speed up your app that the older stuff doesn't have.

You also have to consider that dual-socket server boards aren't exactly cheap (x58 boards can get up their in price too, but are generally cheaper). Also you can easily OC an i7 to 3.8~4.0 Ghz to squeeze free extra-power............can't do that with a server board.

6 cores vs 4:

http://www.anandtech.com/show/3674/amds-sixcore-phenom-ii-x6-1090t-1055t-reviewed/1

Fair enough, but I'm talking about running a computer without a GUI, just using something like HandbrakeCLI to do large quantities of batch encoding, I had the option of buying a dual xeon board and a pair of X5355's @ 2.66 ghz, which I've overclocked to 3.2ghz, or an i7 and i7 board, the xeon set up cost me a little less, as I bought it from ebay, and got a good deal. So I went with it. from what I'm reading I made a good choice, I do tons of video encoding so it was important to me to be able to get the most of my processors, without spending too much.

 

[IntelUser2000]

Oh, then its not even Harpertown but Clovertown.

The i7 965 can beat Core 2 QX9775 x2 in about half of the rendering and image processing benchmarks. And Core 2 QX9775 is 3.2Ghz, 1600MHz FSB part.

Looks like even on Handbrake the i7 might be faster: http://gizmodo.com/5407204/benchmarked-the-quad+core-i7-imac-is-super-fast

 

[HAL9000]

I've clocked mine to 3.2ghz and 1600fsb, should it make that much difference at the same clock speed?

 

[soccerballtux]

if the coding is done right, then 8 will always beat 6+HT, assuming same clock speed and processor design.

 

[HAL9000]

if the coding is done right, then 8 will always beat 6+HT, assuming same clock speed and processor design.

Great, that's what I expected, which is why I went for what I did, ty.