Are quads worth getting?

[V|per]

As many people have said, it all depends what you want to use it for.

If you are an encoder, gfx designer, renderer, or something along those lines, then they will be very beneficial.

Clock for clock the quads are identical to the duals (depending on cache and nm process, ie, an e6600 and a q6600 are spot on clock for clock, a\s would a wolfdale be with a yorkfield) Quad cores are most certainly not useless for games either (depending on the genre you play)

If you are primarily an FPS gamer, then a faster dual core would probably suit you better, likewise if you have crossfire or sli, as you will need sheer clock speed to unleash the full potential of your graphics setup.

If on the other hand you like your RTS games, then a quad can be beneficial, especially in supreme commander (and more so with core maximiser 1.03 [free download]) Infact, in Forged Alliance, i have still yet to see a big map with lots of units play smoothly and a 28k 3dmark06 machine

Conclusion:

rendering/video and RTS game = Quad
FPS gaming and general tasks = Dual

[iranu]

I am of the opinion that if you have to ask whether you need a quad then you don't. Those that think, blimey I could really use those extra cores for all my video rendering etc will simply buy one.

V|per's conclusion is the answer.

[Mblaster]

>> How long till multi threaded apps become common? and Quad become mainstream?

Quads ARE mainstream in video/audio/graphics circles, and have been for a year or two.

For multimedia usage the quads own everything out there.

So Quad core is mainstream in a small section of the market?

As previously stated it's all to do with what you are using your PC for, and how often you are planing to upgrade. At the moment only cirtain types of programs see significant benefits from 4+ cores, and I think it'll be a couple of years before we see a large ammount of programs able to make use of the 4/8 core CPUs.

If you don't want to upgrade for another 4-5 years, then I'd recommend the quad core, but if you'll be replacing your machine in 2 years time you'll probably get more benefit out of the dual core setup, assuming you're not going to be doing a lot of encoding/rendering work, which scales great with as many cores as you can get your hands on.

[Rosaline]

I'd like to the original poster's question, "will quad cores be a good investment for a good long term investment in computing power?". This is different to if it is worth getting one now, and to me seemed more of a general enquiry. I'm going to assume for the nature of answering this that we are talking with respect to applications which do not already at this time make a quad core a required purchase, since the answer for those is quite obvious

To answer this, we need to look at the changing state of the computing market, and the reasons for these changes.

It is commonly believed that Moore's law refers to processor speed, but it in fact is a speculation upon transistor count. It is relatively easy to add more transistors to an integrated circuit, you simply need to use more space on the silicon wafer, or to shrink the size of your transistors.

Processor speed is ultimately governed by how fast the fields within the transistors can form and decay, how one can chain multiple transistors to get valid results considering the time lag (It's been a while since I did my electrical physics, so this is probably not perfectly accurate), and hence how fast the slowest components within can run. It is comparatively run.

On the other hand, adding cores is actually fairly simple to get right. All you need to do is to allocate them some space, and hope no defects occur in manufacturing that break that additional core. There are some design issues to adding cores, but these are simple in comparison to speeding up existing functionality.

When a defect does occur within an additional core, careful design can allow you to disable the broken core and still leave you with a marketable product.

As such, adding cores is massively preferable than attempting to increase processor speed.

Within a processing core, there are often parts with little to do. Some instructions might be delayed as they need a result from an instruction that is still being worked out, others might need information fetching from memory, and others might only use a small part of the core's processing powers. Ideally we want to push more instructions in that will use different details to the slow set, so as to maximise the activity of all parts of the core. We have seen before in Hyperthreading, and shall be seeing again in the Nehalem architecture design, and it is called Simultaneous Multi-Threading (SMT). When it is done right, it is very similar to adding another core, but when it is done wrong it is, well... hyperthreading

Looking at performance graphs for quad cores running highly multi-threaded programs, they appear to be stalling, not getting the maximum possible gain from the additional cores. As such, Nehalem's SMT ability should be quite effective indeed.

Effective SMT is quite hard to build, requiring monitoring of two (or more) threads at once, and making sure that the two threads do not interfere with each other. However, given how much of a processor can lie fallow, the investment in building in this feature is a far more easy route to increase performance than by attempting radical redesigns or by attempting to speed up components.

Given all of the above, it should be no surprise that the processor market is going heavily multi-threaded, both by increasing the core count, and by adding SMT functionality. Intel is due to release Nehalem at the end of the year, a new processor design that will put four, maybe even eight, cores onto one integrated circuit, and each core will have the ability to run two simultaneous threads. I would expect AMD to be working on similar means to increase their threads per processor count.

However, the hardware market is only one side of the equation, as a computer is nothing without software. It is here, not with the hardware, that the importance of multi threading processors will be defined.

Until very recently, software developers did not have to actively consider the possibility of more than one thread running at a time. Despite this, however, they were encouraged to take up good practices such as using an interface thread and a back-end thread. By doing this they could keep a program appearing responsive whilst processing information. This is possible because even with a single core, no thread can have total dominance (this used to be the case, but modern operating systems prevent this). Threads are swapped in and out, depending upon a number of conditions. As such, it made sense that whilst a networking thread was waiting for the next packet, to have the display thread still update for the user and react to input.

Now it seems that dual core systems are becoming dominant, but software really has not changed much since the single core days. There are a number of reasons for this.

Firstly, there is the fact that it is hard to program highly multi-threaded systems to work at peak efficiency. Even most computer science courses now still only teach basic multi-threading, rather than the techniques you need to get the most from running many threads,

Secondly, most computers really have too few cores - and that includes quad core. Games parallelise very well in theory - the activities of all the NPCs and physics calculations can be done simultaneously. The problem is the amount we can currently run in parallel.

Given N possible simultaneous threads, and M objects (NPCs, physics interactions, etc), you need to divide the objects evenly between the threads in terms of time-to-completion, having on average M/N objects per thread. Ideally this should not be done randomly, but done to minimise the amount each processing thread depends upon the results of any other thread. Both the time to sort, and the time to process are dependant upon the number of objects being looked at. The speed of performing the calculations in parallel, given the need to minimise the data hazards between threads, is effected strongly by the difficulty of sorting the objects into separate sets and by the number of sets that then can be done at once. With only a few threads able to run at once, the speed bonus from running in parallel is lessened yet time is still needed to spread the workload between threads. It is possible to get around the data hazard issue by using more memory, but then you hit issues with memory access times becoming a major factor.

::Edit: I released that I introduced the term 'data hazard' here without explaining what they are and why you want to avoid them. I'm using the term here to mean all the various problems you can get when running multiple threads upon the same overall data set at once. In reality there are a number of different specific terms for each of the various problems that can occur. The simplest problem is that of thread A writing to a memory location, and another thread B reading from it. If thread B is expecting the result of thread A to be there, it must read after thread A writes to the location. If thread B is expecting the original data, and thread A writes the new value in first, then everything gets messed up. As such, you have to be careful about the order in which things are done. There are even more complex issues such as deadlocks, where all the threads are patiently waiting for each other to finish - as they're all being kind and waiting, none can ever complete and allow the rest to finish! I hope that helps you understand better why you can't just let all the threads go at once without being careful beforehand And as mentioned, this is only an issue when you are writing to the same memory locations, so you could say "that was the last frame, and we will make a whole new area in memory for the next frame", which means you don't have to worry about any details changing whilst you work, but it needs at least twice as much memory to work! Edit over::

Furthermore, it must be remembered that a game now cannot guarantee any more than a single core really being available to it and it alone. It is reasonable to expect that other processes will always eat up the availability of at least one thread (how many of you game with firefox, IMs, e-mail and explorer windows open in the background?), and if games themselves have seperate input, display and back-end threads, that leaves even fewer open thread slots for the processing to be spread over.

The good news at least is that hopefully that will begin to change. As has already been mentioned, Intel is expected to launch Nehalem by the end of this year, which should offer the ability to run eight threads. Intel is also aggressively encouraging companies to write more and more multi-threaded software, reminding them that the existing scheduling by the operating systems means that they loose little now, but stand to have their programs really benefit as processors can run more and more threads. Games companies are also getting used to writing multi-threaded code, both because of the drive by processor manufacturers, and because to some degree that is how shaders on graphics cards are run.

Much like we have found with the graphics card market, once the developers learn how to multi-thread to full effect, they soon should become able to write code that will need future, even more heavily multi-threading, hardware to run effectively. This point is certainly coming, but I suspect we will have to wait until the average installed hardware base has notably more available threads (at least eight, I suspect).

So, finally we can look back upon the question asked, and see about formulating a proper answer. The hardware market certainly is going more and more towards many simultaneous threads, and it is pressuring the software market to follow. However, the software market is for various reasons lagging behind, and appears to be waiting for the installed hardware base to reach a critical point with available threads for advanced techniques to be applied.

Whilst I believe that the current quad cores actually do not offer enough threads, they are certainly far more future proofed than dual cores. I suspect that it will not be until the end of 2009 that four threads will be considered the normal parallel processing ability of a new computer, and it will probably take until the end of 2010 for software manufacturers to be able to consider their target audience as all being able to run four simultaneous threads. After this, it all depends on the changing nature of the computing market, and so cannot really be speculated upon, but I suspect the average number of simultaneous threads could well start to rise rapidly as fabrication costs lower further.

In conclusion, if you are looking to build a system now, a quad core processor is, price-for-price, going to give you a system that lasts longer and is more proofed against the future changes to the software market.

[Agent]

Rosaline - raising the HEXUS IQ level since July 2007

[SiM]

Wow... Saracen now has competition...

[-iceblade^]

****.

what a great, IQ heavy post, definitely thankworthy... Put simply, yes, a Quad is worth getting, with the only possible exception being if you are gaming, and even then i'd still argue that it's better than a Wolfie. As to which one? Well, Phenom 9850 BE if you're an AMD fan, and a Q6600 if you prefer Intel

[hse180]

One hell of a post Rosaline........will also answer future questions on this topic too

[GSte]

Good work Rosaline .

[Bugbait]

Rosaline gets a thanks for the effort alone and bonus for the information contained .

[SiM]

No, you're both wrong. Install UT3, Crysis, or any recent game, and find out for yourself, and this isn't even getting into the need for concurrency. Just stop perpetuating this antiquated fallacy.

Do you have any evidence to back your claims...

AMD Phenom X3 processor family performance - TechSpot

I think someone is a quad fanb0y
Its seems you are wrong...

[aidanjt]

Do you have any evidence to back your claims...

AMD Phenom X3 processor family performance - TechSpot

I think someone is a quad fanb0y
Its seems you are wrong...

You're compairing a new core revision, 1333Mhz FSB CPU to an older core revision 1066Mhz FSB CPU, lets stick to apples-apples shall we? Crank a Q6600's FSB up to 1333Mhz and drop the multipler to 7, and then we can start to talk. Use your brain, benchmarks aren't a replacement for thinking before you speak. And I'll point out that the E8400 only *marginally* beat out the Q6600, on equal footing (a Q9450) it'll get wasted entirely.

[SiM]

You're compairing a new core revision,

I am not. The reviewers are. And rightfully so. They are in the same price region.

lets stick to apples-apples shall we?

Lets stick to comparing same priced apples...

Crank a Q6600's FSB up to 1333Mhz and drop the multipler to 7, and then we can start to talk.

We are comparing stock speeds... overclocking is never guaranteed. A Q6600 might bsod at 1333mhz FSB...

Use your brain, benchmarks aren't a replacement for thinking before you speak.

No need to be rude, especially because you are wrong. These are actually gaming frame rates, not synthetic benchmarks. Your rude comment is very ironic

And I'll point out that the E8400 only *marginally* beat out the Q6600, on equal footing (a Q9450) it'll get wasted entirely.

So you would compare the almost double priced Q9450 to the E8400

Again I ask, do you have any evidence to back your claims?
If you don't then do not bother replying... you will only save yourself from being embarrassed...

[staffsMike]

You're both making good points of course.

I'd be interested to see the two chips running the same FSB but it would make little difference to the debate I fear.

The fact of the matter is, however good the E8400 may seem or even be, now, it will fade quickly and quietly into nothingness far sooner than the older Q6600 will.

At what could only be considered the same price, the one which has the better long term benefit and the current multitasking benefits is the Q6600, I don't think anyone will argue that.

As demonstrated if you want the absolute best stock gaming performance and flat out refuse to overclock then the E8400 will be better (now and for a little while longer).

What I will say though, is remove the "which bar is bigger" mentality and look at the figures. Neither chip is struggling in any of those games, so you might as well get the quad and reap the benefits of 2 extra cores In my most humble of opinions of course.

[TheAnimus]

Ladies! I've long been an advocate of dual proccessor machines for gaming, back in the day my dual core PII showed this quite well, the reason was there where twelvety apps i left open in the background. If its pure FPS you want, and the game makes no use of the other cores, then there is little point buying cores that will remain dormant.

But On Topic.

As it stands, very few programmers know how to use multiple threads. The amount of people i've interviewed for technical programming rolls who's just don't understand the difference between a kernel object for locking, vrs volatile memory are staggering. Here is a very simple performance killer.

All CS courses, should have in the first year an assignment where students are required to build a lockless queue or stack or tree or something, just so they can understand whats going on.

Personally i think that a shift towards functional programming, mabye not as Hascal/CamL people think, more Aspect Orientated Programming (AOP) will allow easyer parralllisation of tasks to be done by the CPU/JIT-er. Abstraction ideas like the .Net events system for WinForms have been re-developed a bit in later versions one of the main changes is the movement away from all the GUI work happening syncronusly. This means that a more modern runtime environment could leverage extra cores with no demand for code changes!

Multi threading in itself is no pannecia, many anti-virus programs have great problems in allowing multiple threads to access multiple streams (ie FileSystem or Network) concurrently.

It will take a long time before multiple CPUs are fully leveraged, but given that its hard to buy a CPU that isn't dual core at the very least, its only a matter of time before people are forced (all be it kicking and screaming) into the future.

[SiM]

You're both making good points of course.

I'd be interested to see the two chips running the same FSB but it would make little difference to the debate I fear.

The fact of the matter is, however good the E8400 may seem or even be, now, it will fade quickly and quietly into nothingness far sooner than the older Q6600 will.

At what could only be considered the same price, the one which has the better long term benefit and the current multitasking benefits is the Q6600, I don't think anyone will argue that.

As demonstrated if you want the absolute best stock gaming performance and flat out refuse to overclock then the E8400 will be better (now and for a little while longer).

What I will say though, is remove the "which bar is bigger" mentality and look at the figures. Neither chip is struggling in any of those games, so you might as well get the quad and reap the benefits of 2 extra cores In my most humble of opinions of course.

I agree with you completely. I will be getting a quad soon for these reasons. But Jay's original quote is indeed correct, which is what this discussion is about

for gaming, not really at the moment, for anything else, yes.