Dual or quad core? Multicore rendering! Does it matter?

Stimmty · February 2013

I have a new 3570K, with a 480gtx running on my main comp but my old comp is a Q9400 yorkfield quad core running overclocked to 3.2... because of the board limits and the 8x multiplier limit. I wanted to know if I should sell my quad core processor and by a cheaper dual core and overclock it 3.6+ so me and my gf can play at the same time! Also a 9600gt. The resolution is poor but it is doable at this point! SO! CPU>GPU>MEMORY! Remember that im a broke ass college student and so is she! PLEASE LET ME KNOW IF ITS BETTER HIGHER CLOCK DUAL OR LOWER CLOCK QUAD! Multicore render?

Frothybeverage · February 2013

As far as I know, NS2 barely uses more than a single core, and doesn't use more than two cores at all.

|DFA| Havoc · February 2013

Yeah, most games are not heavily multi-threaded, and even those that benefit from multiple cores will still be limited by the primary single thread that essentially cannot be broken down past a certain point. For NS2 in particular, there was a guy who posted on this board at one point detailing a series of tests where he disabled more of his processor cores each time, and in the end he found that there was practically no difference past 2 cores, but that dropping down to 1 had a noticeable impact on performance.

Therefore, selling your old quad and moving up to a dual core i3 on a newer sandy or ivy bridge architecture could make a difference, but the catch is that those newer chips are going to be locked and you won't be able to OC in any significant way. To get an unlocked SKU, you have to move up to the more expensive i5 line. There maybe be some overclocking-friendly AMD dual core chips, but you'd be taking a step backwards in single-thread performance due to architecture there. In addition to getting an unlocked 'K' sku CPU, you also need a motherboard chipset which supports overclocking, and those tend to be more expensive as well.

Depending on exactly how tight your budget is, there are a few different options you can pursue. The first and most ideal scenario is that if you live near a microcenter, you can get a fantastic price on an unlocked 'K' sku i5 chip and typically a $40 motherboard discount on top in a combo deal. For reference, you'd want to pair it with a Z75 or Z77 board for overclocking.

If that's not an option, you can go with a non-overclocked i3 chip and still see some improvement just from the updated architecture. This article may be informative in that regard: http://www.anandtech.com/show/6670/dragging-core2duo-into-2013-time-for-an-upgrade

The good news if you're not overclocking is that you don't have to spend extra money on a 3rd party cooler, and you can spend significantly less on the motherboard too. If you don't mind buying used parts, you can sometimes snag a good deal from tech websites classified forums. For example, here's a guy selling a sandy bridge i3 with 4GB of RAM for $90: http://www.tomshardware.com/forum/105575-8-intel-2105 - You might even be able to talk him down a bit, and that wouldn't be a bad deal. You could pair that with a basic H61 board and be done: http://www.newegg.com/Product/Product.aspx?Item=N82E16813157315

The last option would be to take the long view approach, buy a more expensive overclocking friendly motherboard and pair it with a dirt cheap Pentium (such as http://www.amazon.com/dp/B005LLBBSS/?tag=pcpapi-20) for the time being, with the plan to swap out the CPU sometime down the road. To that end, newegg is currently running a promotion where you can get 8GB of RAM for free when purchasing certain Z77 motherboards, such as this one: http://www.newegg.com/Product/Product.aspx?Item=N82E16813157296 - The downside there is it's a single stick of memory, not a dual channel kit. They have run this promotion a number of times with various different kits, so if you are patient it might change. Then again, if you can stand to be patient then you might want to just wait until the new Haswell chips come out sometime in Q2 with an entirely new socket type.

I hope this helps. Feel free to ask if you have any more questions, I'll do my best to answer.

RainyCaturday · February 2013

I was actually looking at this earlier today, I found that NS2 actually uses all 4 of my cores (pushing them all to around 70-80 and spiking all over the place).

Not sure if this is entirely NS2 or some nvidia multi threading helping to spread the load along with the in game multi threading.

Ghosthree3 · February 2013

That's impossible since NS2 only runs on two cores at most.

|DFA| Havoc · February 2013

Nvidia also has no influence on how your CPU distributes threads. You probably just had some other application running alongside NS2 or in the background that were also hitting your CPU.

ScardyBob · February 2013

AFAIK, windows scheduling moves which threads NS2 runs on constantly, so it appears as if its using all of your cores. Also, in-game performance is going to be limited by only one core (because one of threads will be the bottleneck).

Ciro · February 2013

Stimmty wrote: »

I wanted to know if I should sell my quad core processor and by a cheaper dual core and overclock it 3.6+ so me and my gf can play at the same time! Also a 9600gt?

Sounds like a losing battle. Going off benchmarks for the E8600 vs your current CPU, I'd say no, unless you get it for under 40 bucks or trade.

RainyCaturday · February 2013

Ghosthree3 wrote: »

That's impossible since NS2 only runs on two cores at most.

Nothing is impossible.. if you believe!

But really, there it is.

|DFA| Havoc wrote: »

... You probably just had some other application running alongside NS2 or in the background that were also hitting your CPU.

I make sure to not have this happen whenever I play games, NOD32 is also in "Game Mode" so no file scanning is being done.

BigTracer · February 2013

6 cores, 8 cores, 8 Gb Ram, 2 gb videocards. Good marketing, I say, making people buy what they don't even use.

Good, 2-cored CPU with high frequency and 4 gigs of ram is everything you need to play smoothly.

Sops · February 2013

@RainyCaturday refer to ScardyBob's post

RainyCaturday · February 2013

.

Sops wrote: »

@RainyCaturday refer to ScardyBob's post

Yes that may be true but it is still spreading the load more than it would on a dual core so there has to be some sort of performance increase there, even if minimal.

CommunistWithAGun · February 2013

Stimmty wrote: »

I have a new 3570K, with a 480gtx running on my main comp but my old comp is a Q9400 yorkfield quad core running overclocked to 3.2... because of the board limits and the 8x multiplier limit. I wanted to know if I should sell my quad core processor and by a cheaper dual core and overclock it 3.6+ so me and my gf can play at the same time! Also a 9600gt. The resolution is poor but it is doable at this point! SO! CPU>GPU>MEMORY! Remember that im a broke ass college student and so is she! PLEASE LET ME KNOW IF ITS BETTER HIGHER CLOCK DUAL OR LOWER CLOCK QUAD! Multicore render?

If you ever plan to use anything besides NS2, you're a fool to get rid of that CPU for the sake of MAYBE a frame rate increase. If anything you're side-grading at the best.

ScardyBob · February 2013

RainyCaturday wrote: »

Yes that may be true but it is still spreading the load more than it would on a dual core so there has to be some sort of performance increase there, even if minimal.

There isn't because it isn't actually spreading the load. It's simply just shuffling the bottleneck from thread to thread so fast that Task Manager's plot isn't precise enough to show the problem. It's like if you tried to distribute the weight of carrying a heavy suitcase by shuffling it back and forth between your hands very quickly. At any given time, one of your arms would be bearing the full weight of the suitcase.

An easy way to test this is to lock NS2 to specific threads via setting its affinity. You'll find no difference for in-game performance past using more than two threads (locking it to one thread will impact performance, because NS2 does multithread some rendering work).

RainyCaturday · February 2013

ScardyBob wrote: »

An easy way to test this is to lock NS2 to specific threads via setting its affinity. You'll find no difference for in-game performance past using more than two threads (locking it to one thread will impact performance, because NS2 does multithread some rendering work).

Coming right up!

I performed this benchmark using the same settings, one after another, no other programs running, and used Fraps to do a 2 minute benchmark of gameplay played back from a demo.

As you can see here, and this is just my system and one benchmark, but I gain about 10fps using all 4 cores instead of locking NS2 to 2 cores.

NS2 Set to 2 Cores:  Frames, Time (ms), Min, Max, Avg
                       6866,  120000,   34,  172, 57.217

NS2 Set to 4 Cores: Frames, Time (ms), Min, Max, Avg
                      8216,  120000,   43,  181, 68.467

*The addition of frames on the second bench is confusing me but it might be due to starting the bench at slightly different times, I tried my best to start it as soon as the loading screen finished.
Also, on the second bench, it processed more frames and still gave me a higher fps so I don't know exactly what that means in terms of performance. (Figure the addition of frames is from the higher fps)

ScardyBob · February 2013

RainyCaturday wrote: »
ScardyBob wrote: »

An easy way to test this is to lock NS2 to specific threads via setting its affinity. You'll find no difference for in-game performance past using more than two threads (locking it to one thread will impact performance, because NS2 does multithread some rendering work).

Coming right up!

I performed this benchmark using the same settings, one after another, no other programs running, and used Fraps to do a 2 minute benchmark of gameplay played back from a demo.

As you can see here, and this is just my system and one benchmark, but I gain about 10fps using all 4 cores instead of locking NS2 to 2 cores.
NS2 Set to 2 Cores:  Frames, Time (ms), Min, Max, Avg
                       6866,  120000,   34,  172, 57.217

NS2 Set to 4 Cores: Frames, Time (ms), Min, Max, Avg
                      8216,  120000,   43,  181, 68.467
*The addition of frames on the second bench is confusing me but it might be due to starting the bench at slightly different times, I tried my best to start it as soon as the loading screen finished.
Also, on the second bench, it processed more frames and still gave me a higher fps so I don't know exactly what that means in terms of performance. (Figure the addition of frames is from the higher fps)

Well, that's interesting. The previous times I've tried benchmarking by changing the number of threads available to NS2, there wasn't any difference past 2. Did you lock the cores using Task Manager or via a batch file command?

Also, mind uploading and posting the link to your demo file? I'd like to run it on my system to see if I can the same results.

YMICrazy · February 2013

He used the demo from this thread:

http://forums.unknownworlds.com/discussion/127867/benchmark-demo-thread/p2

But only the first 2 minutes and then cut it off. Either way in 19 minutes this demo will not work since it will be 240.

RainyCaturday · February 2013

Glad to spark some interest!

I used the demo in this thread http://forums.unknownworlds.com/discussion/127867/benchmark-demo-thread/p1

and I locked it using task manager

My specs are

i7 920@4GHz
GTX 580 Latest beta driver, No OC
6GB Ram 1600
Windows 8

*Gorgeous Update Edit*

I made a demo with really similar gameplay to the one I was using earlier but for some reason it shows the network disconnect symbol at the start and wont play. I made another one but having similar issues. Also when you try to disconnect the game will hang and you will have to force close it.

Until I can get a demo working I'll just post some bench from direct gameplay.

Here are the benches, 6 minutes each for consistency. Same map, same settings all new features set to low.

Frames, Time (ms), Min, Max, Avg
 56848,    600000,  58, 158, 94.747

Frames, Time (ms), Min, Max, Avg
 48615,    600000,  48, 162, 81.025

Frames, Time (ms), Min, Max, Avg
 42842,    600000,  44, 160, 71.403

Since these are longer benches, the Average FPS will appear higher. What I see here is that my min has gone up a little and my max has gone down.

Again, this isn't the same demo I used before so the data is going to be a bit different +/- but I tried to get a larger sample to make up for that.

Stimmty · March 2013

Thanks guys for your input, unless i find myself an awesome deal on a Core 2 Duo that I can clock close to 4ghz for around 50 dollars, imma keep the quad in. I also played the new gorgeous update w/ my old 3.2ghz quad w/ 9600gt and ran a decent 45-50fps w/ everything off, it seems they made it run better cause i was running 40fps on my quad before. My main is running 80fps everything turned on x1050p... on both. We really need a developers input, I asked Hugh, NS2HD on youtube and never received a response. Honestly, that quad benchmark is pretty convincing.

rockypockyteriyaki · March 2013

e5200 --> overclocks like hell (4.2 ghz easy) and if you can find one, price will be like 20$.

HeatSurge · March 2013

Careful with older processors. It'd be interesting to test with NS2, but clock-per-clock they have much worse performance than the latest intel ones.

You could probably overclock an ancient P4 on water or something to 6ghz+, but it might end up slower than an i7 2600mmmmmmkay at 3ghz (again, it'd be interesting to test if someone has lots of CPUs, mobos, patience, and time xD)...

matso · March 2013

There are three threads of interest in the NS2 client.

- logic thread; runs slower the more entities there are. Usually the bottleneck late game. No configuration options affects this.

- renderer thread; runs slower with more objects/effects to render. Can be configured to run faster using the graphics options. Depending on your CPU/graphics cards mix, it may or may not be the bottleneck early game.

- prediction thread; runs 20 times per second to handle new server data, updating the server data to the client time (prediction). Takes more time the higher the latency and the more entities there are around, but is usually never the bottleneck.

The prediction thread can consume something like 10-25% of a cpu, depending on your latency and entity count, so late game, quad cores will run faster than a dual core (early game, there is usually enough time left over in one of the other cores to sneak in the prediction thread).

Oh, do note that playing around with graphics option on a empty game will probably have no effect on your lategame fps (on a quad-core, at least), as the bottleneck will be the logic thread.

DC_Darkling · March 2013

Noone really?

Ok.. I will again then.
Intel cpus (I do not know about amd) have on there more expensive ones a turbo option.
i5 and i7 CPUs, almost all have a turbo option.

Turbo pauses some to almost all cores, puts them in a sleep mode.
less active cores = less heat.
So this means that the only core or cores left can be automaticly overclocked. The system does this completely automatic aslong as you configured the minimum settings.

So this means a i7 with 4 true cores or 8 'cores' can at times choose to run with only 1. If mine does this it goes up from 2.8 to 3ghz+ completely automatic.

So if you consider a i5 or i7, multithread or not should not be a to big of problem to think about. It will turbo on its own.
Also note that turbo, unlike manual overclock, is fully covered by warranty.

Dual or quad core? Multicore rendering! Does it matter?

Comments