CPU cores (multiple) and the debate

[Zak33]

Lo people

I'm reading the other thread about CPU cores and the need (or not need for them) but rather than hijack it I started this because I'm confused

I went, as you all did, from a single core, to a dual core and felt the entire Windows experience improve. Not just gaming. Then I went to a Quad.. the original Intel Core 2 Quad and I owned it for AGES!

I build a few dual cores for friends, but my quad was good.
Then I built a Hex AMD and I still have it.. I'm typing on it.
In gaming terms I will build a new rig in the next ... 9-12 months I guess. And I will build my Dad a movie editting rig to.

The question is simple: As I see most or all of my cores are being used on nearly every game I use, and on things like encoding, I see it too, surely multi cores is best and if not why not?

£ for £ an 8 core AMD and mobo looks like AWESOME value against something like an Intel Quad.

I see Xeon's turning up in video editing rig suggestions and one day I will have to advise my Dad.... 8 Core AMD with tons of ram for a set price.. or an lower spec intel with less cores.

With no fan boih ism, how will getting a lower spec Intel benefit every day life over an Octa with loads of ram for the same dough?

£106 buys an 8 core low energy 95 w AMD octa

A Haswell Quad will cost £147 and a Skylake for £152

A Xeon Haswell is £171

in real terms, the AMD does looks like astounding value.

And I know Ferrals new Octa 8350 goes like a train and it's £140

Surely Octa wins, no?

That £ difference buys a LOT of extra ram or another SSD in some cases!

[peterb]

Depends to some extent if the software you are planning to run can use them. For example, I am tipping and transcoding some material - the ripper uses one core, but the encoder uses two.

My guess is that things like games that are processor and GPU intensive will use multiple cores, similarly video processing software generally - I think Final Cut Pro does.

So a qualified yes for multiple cores, but make sure you use software that can use them.

[shaithis]

There is also the IPC to take into account. If your comparing CPUs of different architectures, you have to take that into account. There are almost certainly single core (with HT) CPUs now that will outperform the original dual cores.

Toms benchmarks from the 8350 launch has a 3770k beating it in all but one of the productivity tests (including media encoding and PS6).....
http://www.tomshardware.co.uk/fx-835...-32550-10.html

But it doesn't make the AMD a bad purchase for the workload, especially when you consider you would really want the xeon with HT (approx £210) and the mobo will certainly cost more.

[Zak33]

ah.. the Xeon with hyper threading WAS the one, you're right Shaithis

I have seen all my cores doing something nearly every time on nearly every game. So maybe an AMD load balances itself? Rather than the game?

[Zak33]

I've just played Dirt for a moment and it's using all 6 cores to the max

that fair enough...its quite power hungry

but I also loaded up Sol Survivor and it skips between 2 and 4 cores and sometimes drops core 3 and uses 1,2 and 4 for example

Does the mobo or cpu load balance ?

[DanceswithUnix]

ah.. the Xeon with hyper threading WAS the one, you're right Shaithis

I have seen all my cores doing something nearly every time on nearly every game. So maybe an AMD load balances itself? Rather than the game?

No, and that was one of the problems AMD had when PileDriver first came out. Windows was rubbish at scheduling tasks on it, throwing performance away. In some ways it is harder to schedule tasks with a lot of cores, because cores can not only help each other along they can tread on each others toes as well.

In short, all you can do is look at benchmarks for what you want to run and compare with the price you would be paying. I bought my 8350 ages ago, and at £125 it was a really good deal and I still am not that bothered about upgrading it. These days the FX has oddly gone up in price, and over several generations the i5 has both improved in performance and gone down in price, I find it hard to recommend an FX these days.

[Zak33]

so my Hex core might be doing more work than it needs and running over itself?

[DanceswithUnix]

so my Hex core might be doing more work than it needs and running over itself?

It shouldn't do. A program should divide its work up into well designed threads, the operating system should divide those threads nicely onto the available cores.

I was just saying, there is nothing automatic about it. The program has to be well written to benefit from a lot of cores.

[Rob_B]

Might be an interesting read

http://www.eurogamer.net/articles/di...est-gaming-cpu

[EndlessWaves]

surely multi cores is best and if not why not?

It depends entirely on the task at hand.

Imagine following a recipe. Some of the steps naturally work best when done side by side. Having the oven heat up while mixing the ingredients are things that you're going to want to do simultaneously.

Other tasks can be split up but may cause problems and need additional co-ordination. If there are lots of ingredients to prepare then to cut down the time you can get multiple people in to help and divide up the work, but you'll likely have to supervise them and make sure people aren't waiting for others to finish with measuring scales and other shared resources.

Some tasks can't be split up at all. There's no way you can get multiple people or multiple ovens involved to make the cooking time any shorter.

It's the same for computing tasks. That's why core counts have generally been kept low and the focus has been on improving performance in other ways.

Additionally, dividing up a task requires knowledge about it so there's no way to do it automatically. It has to be done by the programmer writing the application.

A task gets divided up between workers called threads and these threads then get assigned processor time by the operating system - although programs can request that certain threads go to certain cores to help balance the load on more demanding threads.

Writing threaded software is hard. In a single thread or core everything behaves sequentially. If you issue instruction B after instruction A then B happens after A every time. But if you issue instruction A to thread 1 and then issue instruction B to thread 2 they could happen in either order as thread A may be held up waiting for something else to finish on the core it's assigned to.

As such, you have to allow for a lot of contingencies and require one action to wait on another. If you issue a save and close command then you want the work to be saved before the program closes.

That's why multiple cores have been so to be supported for many tasks and when they are you often see some cores with much higher usage than others. They're difficult to work with and not well suited for everything

More cores have little or no intrinsic value. An 8 core processor offers nothing over a 4 core processor whose cores are twice as fast. Given the issues outlined above the double speed four core would offer significant performance advantages in many tasks. If you're dealing with a purely single threaded task where all steps would run sequentially then the four core would run it at 1/4 top speed while the eight core would run it at 1/8 top speed and take twice as long to finish.

AMD's options are more than half the speed of Intel's per core. However, AMD's design doesn't have eight identical cores but instead has four sets of two with each pair sharing certain resources. It only have four copies of these shared components - notably for gaming the floating point units which are used heavily by most games.

Hence the typically abysmal gaming performance, for one of the biggest tasks in a game they've got the same number of 'cores' to work on it but they're a lot slower.

In terms of hyper-threading (the generic name for which is Simultaneous Multi-Threading, SMT) that's a way to organise work more efficiently. Cores are not simple units so a thread assigned to a core doesn't always use all of the available resources. SMT allows multiple threads to be assigned to a core in order to use some of the unused capacity. It has all the drawbacks of multi-core, and a couple of it's own (scheduling issues) with only minor benefits. On the other hand, it presumably takes a tiny amount of extra silicon compared with adding more cores so the pay-off is likely to be in performance gain vs. extra manufacturing cost.


Oh, and those tasks that naturally like to split? Perhaps the most familiar one for you is 3D rendering and that was so much better done massively parallel that games started requiring dedicated massively parallel processor. That's why games require a GPU to run, rather than just using the general purpose CPU like every other program.

[CAT-THE-FIFTH]

When it comes to gaming performance,you tend to be mostly limited by graphics performance. Among a few of use we have the following gear:
1.)Myself - Xeon E3 1230 V2/Core i7 3770 and a GTX960
2.)Mate 1 - Xeon E3 1230 V3/Core i7 4770 and an R9 280
3.)Mate 2 - FX6350 at 4.5GHZ and a R9 390
4.)Mate 3 - FX6300 at 4.4GHZ and a HD7870XT
5.)Mate 4 - Phenom II X6 1045T and an HD4670(their HD7850 has problems)

When it comes to something like ARK - my mate with an R9 390 destroys all of us in performance.

A lot of the time people look at review sites with very artificial benchmarks - either a single very high end card or two of them.

Most people will not have a GTX980TI or a Fury X. They are more likely to have a GTX970 or an R9 390 class graphics card and slower.

This is why I tend to now go on YT to see if people get playable framerates on more normal hardware.

In many cases,MOST modern games will run perfectly fine on slower CPUs and graphics cards.

Remember,with a fast graphics card lots of games will get you over 60FPS and most people own 60HZ monitors.

At this point all the E-PEEN,I haz more FPS than you really means nothing.

For instance,this a Core i3 6100 "owning" an FX8350 according to some on a large enthusiast forum:

http://pclab.pl/zdjecia/artykuly/rad...vision_cpu.png

This an FX8350 being owned:

http://gamegpu.com/images/stories/Te...rimal_proz.jpg

Yet,look at the Core i7 4770K - it is only 10% to 15% faster than an FX8350.

But look at the difference when you move to a faster card:

http://gamegpu.com/images/stories/Te...mal_1920_1.jpg



There is nearly a 60% improvement in going from a GTX970 to a GTX980TI.

Now,no doubt I do better in D3 than my mates with an FX6350,but even then when the massive dips to crap framerates happen,it happens to all of us. It takes a tad longer for it to happen to me but it still happens.

So,its not like an Intel Core i7 is better than a FX8350 or a slower CPU but the problem is that you tend to more limited by your GPU horsepower most of the times IMHO OFC.

I also find the who obsession with CPU benchmarks by enthusiasts mean that plenty of people instead of buying a £100 to £150 graphics card to upgrade their rig,just end up buying a console since they think the card will be "bottlenecked" and guess what 10 years ago when people were more into modding budget parts,OFC the high end stuff was faster,but it wasn't about that - it was about compromises but much lower cost meaning we could get decent gameplay.

In the realworld,just because the latest 1000BHP Veyron exists,it does not mean the market for town cars is pointless.

I just think that the whole all or nothing thing which has recently emerged with computer enthusiasts is just missing the point.

[Jonj1611]

I do 3D rendering so multi-core is a must have, started out with a single core and the speed just made you want to cry, now on a i7-4790K and rendering is so much faster from previews to final. Obviously the bonus is its great for games too.

[AETAaAS]

If you're waiting 9-12 months, Zen and Kaby Lake could well be out so I'd hold for them. IMO, your CPU is something which can stick around for 3-4 GPU generations if not more, so I personally splash out on the CPU and expect to keep it a while.

[Ttaskmaster]

AMD's options are more than half the speed of Intel's per core. However, AMD's design doesn't have eight identical cores but instead has four sets of two with each pair sharing certain resources. It only have four copies of these shared components - notably for gaming the floating point units which are used heavily by most games.

Hence the typically abysmal gaming performance, for one of the biggest tasks in a game they've got the same number of 'cores' to work on it but they're a lot slower.

Wasn't there some hoo-hah recently over AMD's Six-Core actually being only 3-core (ie 3 real, 3 virtual) and thus misadvertised as 6, with the same on the 8-core technically only being 4?

[scaryjim]

Wasn't there some hoo-hah recently over AMD's Six-Core actually being only 3-core ...

Someone in the USA tried bringing a class action law suit for false advertising. My understanding of processor design is relatively limited, but AFAIK x86 is the integer instruction set (x87/MMX/SSE etc. are FP extensions to it) and since it's the integer cores that *aren't* shared on bulldozer et al. there' not much of an issue with saying that the processors have 8 x86 cores (and 4 shared FP co-processors, effectively)

[Jonj1611]

Wasn't there some hoo-hah recently over AMD's Six-Core actually being only 3-core (ie 3 real, 3 virtual) and thus misadvertised as 6, with the same on the 8-core technically only being 4?

Think this is what you mean :-

http://www.pcworld.com/article/30031...rocessors.html