Are there benefits to a bigger CPU?

[jimbojames]

Hi,

I am new to gaming rigs and OC'ing so sorry if what I am about to ask is wrong/simple/stupid

I have been reading this forum loads over the last week or so and I have seen a few posts that give me the impression that after a certain size or speed, the CPU has no contribution to improving FPS or quality ect.

I went to the supermarket and noticed that the custom pc magazine had a test of 2 identicle rigs but one on a 2500k and the other on a 3930k intel, everything else was the same. The results were about 20 FPS better on the 3930k.

So - my question is who's right and why?

 

[Darklight1990]

I'm quite new to this as well but as far as i can tell the real difference is Hyper threading. the I5 2500K has 4 physical cores but no hyper threading giving it four threads or 4 streams of data if you like, but an I7 3930K has 6 physical cores so it can handle 6 data streams but also has hyper threading so in effect it can handle 12 threads instead of six. hyper threading in effect adds virtual cores when the data stream being processed doesn't use the whole processing power on one core so it lets the processor handle 2 data "threads" at the same time. however most games don't make use of hyper threading so an i% can run them all fine, after all its mostly GPU intensive. but some games, i believe GTA IV is one, make use of hyper threading so you get a better FPS rate as the cpu picks up the extra processing. hope this helps. and hope i hit the salient points feel free to correct me!

 

[Lauralarry]

imagine a road with 1 lane going to 1 place, and tons of cars on it :/

Now imagine that road being turned into a 2 lane, more speed more cars, more more more...

 

[BKCXb]

a 2500K is a 4 core CPU and a 3930K is 6 of the very same cores with hyper threding, meaning 6 cores 12 threads.

core wise it is 50% more and thread wise it's 300% more.

keeping that in mind, if you take an older game designed to be run on an older dual core machine with hyperthreading, it will run great on the 2500K's 4 cores without hyper threading. It would run fine on any 4 core CPU or dual core with Hyperthreading. The only way to get better performance out of that game is a higher clock rate. This is why the 2500K is a gamers champion, it is the cheapest unlocked 4 core Sandy Bridge CPU, anything beyond that ,if all you do is game on it, is a waste of money, anything below you could lose out on performance , by not being able to overclock as high , with a non - unlocked CPU.

If your not gonna overclock and run the CPU at stock speeds most lower CPUs only trail by the percentage they run slower, compared to the 2500K. A 2400 is only 6% slower than a 2500K because 3.1GHz is 6% slower than 3.3GHz, the speeds that these processors run at. In real world gaming some cheap dual core CPUs with hyper threading are about this much slower than the full 4 cores of the 2500K. This is because a thread adds performance to a core, but can not make up for the absence of one. 4 full cores are better than 2 cores with 4 threads.

Now lets move on to a newer coded game, lets say one that can take advantage of 6 threads. A 2500K now has to do 50% more work, but this is not that much of a problem because most games dont fully load the CPU at all times, the 2500K remains the gamers champion, it just has to work alittle harder.

What about a future game that can take advantage of 12 threads, well thats 300% the work of 4 threads, the 2500K will be maxed out, trying to keep up. the game will have to run slower as the CPU catches up on the increased workload. In the same rig you could just throw in a 2700K which runs at 3.5Ghz(roughly 6% faster) but now has hyperthreading(8 threads). The 6% increase in clock spead will help it deal with some of the workload, but since extra threads are not as good as full core, it ends up being a wash, and the 2700K will do 200% of the work of the 2500K, which would be to overloaded some of the time. 8 threads is 66% of what the game is asking for, at 12 threads. In this example the much pricer CPU might just do just fine as the game doesn't load all threads at 100% all the time.

In gaming FPS isn't 100% linear, or just based on the CPU, same machine, same game ,same settings, even though the 2700K can process twice as much data in the same time you won't go from 30FPS to 60FPS. It might only go from 30 to 40 in that game.

Back to the 12 thread game, the 3930K is a 6 core ,12 thread CPU, It will run that game just fine just like the 4 thread game ran on the 2500K, as the game doesn't max out every thread all the time. The limit of this CPU won't come until games able to load it down with 36 threads become available(300%) if you go by my 2500K example and your FPS might jump up to 60.

This is the nature of things getting better, game desiners push the hardware, hardware makers make better hardware, than game desiners now have more to work with and make better games, so on and so on......

from our end it looks like this . My GTX 7800 runs unreal tournament 4 maxed out at 60FPS all the time, great! years later I'm tired of UT4, I'll try this new game I hear so much about Crysis2, WTF!! my GTX 7800 only runs it at 1FPS?????? OH well, back to the store for a 560Ti, WOW it runs UT4 at it's 300FPS cap all the time and Crysis2 at or above 30FPS, all the time.

Keep in mind this is grapfics cards that span several generations, because games are made more for graphics cards, than CPUs, if we needed a new CPU every time a new game came out, I don't think their would be as many computer gamers as there is, but gamers think nothing of buying a $500+ graphics card every year or so.

As it sits today most games out are made to run fine on 4 to less than 8 threads, because 99.8% of the market has this to game on. If a game developer made a game that needed 36 threads to run it would be sitting unsold on the store shelf for about 10 years until that types of CPU power becomes cheap enough for most of us to afford, why make it if no one will buy it. Plus you need the hardware to develope it on, with Sandy Bridge-E just recently being released, it will be a few years before we see games needing more than 8 threads. The games that use more threads run faster the more threads a CPU has, unless you overclock, If a 2500K would run at 6.6GHz nobody would ever buy anything more, and we wouldn't be talking about this.

hope that helps....

 

[Master&Puppet]

Total Capacity

What you have to remember is that the overal performance of the cpu is what matters, not the number of cores or GHz.

If you compare a Phenom II 1100T (6 cores running at 3.3 ghz) and a i5 2500k (4 cores running at 3.3ghz) you would assume that the 1100T is 50% faster.

Now take a look at these pages covering cpu performance start with General performance and flick through until the Visio Studio page. What you will find is that the 2500k actually kicks 10 bells out of the phenom in most tests. The capacity of 2500k's individual cores is actually much greater than the 1100Ts.

To exemplify this, take a look at the Cinebench R10 single threaded test. This is a test which only allows the cpu to run 1 thread, the work load of a thread cannot be divided up between cores - so only 1 core can be used. It's like testing a 2500k running on 1 core against the 1100T running on 1 core. What you can see is that in this example the 2500k's cores are about 50% more powerful than the 1100Ts.

Now moving on to the 2600k and hyperthreading. HT allows each core to work on two threads at the same time so it can work on 8. However a hyperthreading core isn't like like doubling the cores. Here's a desciption from wikipedia but this example should be good enough to make the point:

A 2500k core will be queued threads to start on by the operating system. The core will work on one and once finished will move on the next.

But what if the thread only takes up 70% of the cores capacity?

A 2600k core will work on the threads in order, just the same as the 2500k. However, rather than sitting with 30% of it's capacity idling it will simultaneously start working on the next thread as well.

So you can see that the 2600k basically doesn't waste it's capacity.

If you look at that single threaded image again you can see that single threading performance is virtually identical between the 2500k and 2600k (remember that the 2600k works at 3.4ghz not 3.3ghz). This is because they are essentially identical cpus, it's just that the 2600k is running a bit faster @ 3.4ghz and is allowed to HT, where the 2500k is not. So if you now take a look at the multi threaded test you will see that the 2600k is able to work quicker because it can make use of its spare capacity.

The actual performance gain is very dependant on the difficulty of the threads being handled. The more difficult the thread, the more capacity of the core is filled, so the less spare capacity can be used for HT and therefore the less gain you will see. Typically, however, you will see a 10% to 30% improvement in total processing power of a 2600k over a 2500k.

Now to the main point - total capacity and gaming.

There are a lot of conflicting reports on the internet of cpu performance on games. Unfortunately these are often misleading because they test in different ways.

In games the cpu basically does the maths - it calculates what you are shooting at and how much damage you take when you get hit and how the ai will respond and behave. Regardless of the resolution this will place a fixed amount of demand on the CPU.

The GPU's job is to make the game look pretty, so it takes what CPU says is happening and literally turns it into a slideshow (defined as the fps).

Now at a low resolution with low game settings the gpu has a very easy job. The higher the resolution and the higher the settings, the harder it has to work.

As a proportion of overall work done, as the resolution and settings increase, the gpu will do an increasing amount of the combined work.

So where do you draw the line? A game which has a lot of mechanics (I believe that Shogun 2 is like this) will put more demands on the cpu. A game which looks prettier will put more demands on the gpu.

If you take a look an experiment I did on the forums earlier in the year you can see an example of cpu usage on fps in Battlefield 3:

Picture 1 - This was all done with the same gpu clock and game settings. The only thing I changed was the cpu. You can see that performance improved as I increased the clock and opened up the full 6 cores. However you can see that there is no difference at all between the fps on the 6 core tests. This suggests the entire overclock was not necessary and the cpu was comfortably dealing with the game. At this point the fps scores were limited by the gpus.

Taking a look at the Just Cause 2 and Shogun 2 results you can see that the performance scales as I increase the power of the cpu. This suggests that these games put more emphasis on the cpu than bf3, which could make sense given the large sandbox nature of just cause and the model complexity of Shogun 2. In both cases my cpu was probably bottlenecking the gpus.

Does it matter that my cpu is bottleneck the gpus?

Not really, depending on the game. Taking a look at the cpu bottlenecked scores:

Battlefield 3 looks quite light on the cpu demands, even when I turned off 2 cores its bottleneck still produced a min frame rate of 45 and in shogun 2, which was the most cpu demanding it still made 32fps.

You could run the same tests on a 3930k overclocked at 4.8ghz and the fps would be much higher. But who cares? You can't see the difference.

Conclusion

Yes a poorer cpu bottlenecks cards. Something is always bottlenecked but there is no point going overboard. Even a phenom II locked down on 4 cores at stock produced a 32 minimum fps score in it's hardest test. That is actually a mental score. The gameplay would be seamless. Check out this page to see the fps of different cpus not bottlenecked by cards.

Going back to bf3, which is known to be hard on the framerates, you can see that the quality of the graphics cards are much more important than the cpu. In order to produce the 45 minimum fps cpu bottleneck you just need a phenom II x4 at stock ghz. You can get a 965 for £90 off Aria (4 core @ 3.4 ghz). In order to produce 45 fps at 1080p ultra settings to match the CPU you would need a top card, 7950/7970/680 which would set you back £300 or more.

Now you could replace that cpu with the 3930k and overclock it to 5ghz. The game would still only run at 45 minimum fps because the gpus can't handle the graphics any faster, so why would you spend nearly £300 pounds more on the cpu?

Having said that I'm not suggesting that everyone should buy a 965, the 32fps on shogun 2 doesn't give you much future proofing and for general performance it sucks compared to a i5 2500k. The i5 2500k is really a sweet spot, it is really at the cheap end of the cpu market (£145 off Aria) and will handle game mechanics with ease for a long time to come.

Hope this helps!

M&P

 

[jimbojames]

Wow - thanks to BKCXb and Master&Puppet for the detailed explainations. I have only been a member of this forum recently and I must say it is the best one I've been a member of (pc related or otherwise).

Cheers guys - very much appreciated!