Of game engines, CPU cores and threading...

[Sylvanas]

Bit-Tech and how many 'cores' do you need

To note are slight differences with how ATI and Nvidia handle threading, but the basic idea is that most engines don't really make use of more than 2 cores if that, sometimes 3. The general trend seems to follow the CPU's single threaded performance, hence i7 being ahead of Quad core Phenom2's. So it remains that we still haven't got to the stage yet where games are capable of harnessing all the resources at their disposal.

 

[ModestGamer]

Poorly written software is still garbage. this should be handled at the OS level though.

 

[cbn]

That is interesting how BCBF2 is optimized for dual core when using an ATI card.

 

[BFG10K]

I was just about to post this.

It’s true, 2 cores are still perfectly fine for gaming at any reasonably detail level in most cases.

Even the much touted BC2 is showing marginal improvements when going from 2 cores to 4 cores (ATi), and that’s at 1080p with no AA. Add just 2xAA and the difference will likely be zero.

nVidia’s drivers are clearly sub-optimal in that game because ATi is able to saturate their faster 5870 with just two cores, while it takes nVidia four cores to do the same on their slower GTX470.

 

[cbn]

It’s true, 2 cores are still perfectly fine for gaming at any reasonably detail level in most cases.

At the moment Core i5 750 is just so much better at running two threads compared to Clarkdale.

I just wonder what happens when Sandy Bridge is released?

 

[Sylvanas]

Poorly written software is still garbage. this should be handled at the OS level though.

Unfortunately in most cases it comes down to cost vs benefit for most developers. Everyone would like their games to run at 60fps maxed and make use of 6+ cores but the added development time and resources become a hindrance to most studios tight production schedules. It may take months to 'properly' optimise your game engine to use 4 cores which comes at the expense of development time you could put into say the game world- something which 100% of people who play the game will see and interact with, rather than less game, but higher performing for the <5% with 6 core CPUs.

This is probably why we see Dual cores being the common denominator through all engines still, most of the player base is still using them. I have no doubt games with longer development cycles like id's tech 5 engine for Rage will make use of multiple cores, can only hope engines like this get adopted more often.

 

[cbn]

This is probably why we see Dual cores being the common denominator through all engines still, most of the player base is still using them.

http://store.steampowered.com/hwsurvey/

Yep, according to this "dual core" installation base is holding steady. "quad core" is increasing at the expense of the "single core" installation base.

Maybe future console ports will use more x86 cores?

 

[ModestGamer]

It shouldn't be that much work in reality. the Os is supposed to handle threading task. Look at apple for instance. the Os handles the threading not the software. So it has to run through 2 layers of BS to get to where it needs to go. Total BS and a complete fialure on microsofts ends.

Look at the BEOS or HIAKU those OS's handling the threading and do incrediable things with it.

the problem is we don't take microsoft to task on this deficancy in there code.

Unfortunately in most cases it comes down to cost vs benefit for most developers. Everyone would like their games to run at 60fps maxed and make use of 6+ cores but the added development time and resources become a hindrance to most studios tight production schedules. It may take months to 'properly' optimise your game engine to use 4 cores which comes at the expense of development time you could put into say the game world- something which 100% of people who play the game will see and interact with, rather than less game, but higher performing for the <5% with 6 core CPUs.

This is probably why we see Dual cores being the common denominator through all engines still, most of the player base is still using them. I have no doubt games with longer development cycles like id's tech 5 engine for Rage will make use of multiple cores, can only hope engines like this get adopted more often.

 

[Lonbjerg]

Try playing ARMA2...it will gobble up CPU cores like no other game.

 

[thedosbox]

Try playing ARMA2...it will gobble up CPU cores like no other game.

Not according to these guys.

 

[Lonbjerg]

Not according to these guys.

So forumsposter are cluesless...is that it?

Take a look at the real world:

http://www.pcgameshardware.com/aid,687620/ArmA-2-tested-Benchmarks-with-18-CPUs/Practice/


(and they even forgot to use the "tweak": -cpuCount=x
Where x = the number of real CPU cores you have, as this game dosn't play nive with HT
This has been implentend into the lastest patch BTW)

You don't suffer for the dellusion that multicore = 100&#37; usage per core used...do you?

That the second post you have made to me with nothing relevant in it...care to make it a hat-trick?

 

[cbn]

So forumsposter are cluesless...is that it?

Take a look at the real world:

http://www.pcgameshardware.com/aid,687620/ArmA-2-tested-Benchmarks-with-18-CPUs/Practice/


(and they even forgot to use the "tweak": -cpuCount=x
Where x = the number of real CPU cores you have, as this game dosn't play nive with HT
This has been implentend into the lastest patch BTW)

You don't suffer for the dellusion that multicore = 100&#37; usage per core used...do you?

That the second post you have made to me with nothing relevant in it...care to make it a hat-trick?

That is a ~20% increase in FPS with quad core q6600 (2.4 Ghz/4MB L2 cache per core) over dual core E6600 (2.4 Ghz/4MB L2 cache per core).

A better example of quad core scaling would be something like Dragon Age Origins or Left for dead 2.

 

[Lonbjerg]

That is a ~20&#37; increase in FPS with quad core q6600 (2.4 Ghz/4MB L2 cache per core) over dual core E6600 (2.4 Ghz/4MB L2 cache per core).

A better example of quad core scaling would be something like Dragon Age Origins or Left for dead 2.

Where is the law staitng that the scaling must be linear?
Some threads need to wait for other threads to finish.
The fact that it gains performance tells you that it is using more cores.

 

[thedosbox]

So forumsposter are cluesless...is that it?

Take a look at the real world:

http://www.pcgameshardware.com/aid,687620/ArmA-2-tested-Benchmarks-with-18-CPUs/Practice/

Seeing as they don't specify whether they're using a scripted benchmark, or how many times they ran through their test to eliminate variances, the results would seem as valid as the random forum posters.

(and they even forgot to use the "tweak": -cpuCount=x
Where x = the number of real CPU cores you have, as this game dosn't play nive with HT
This has been implentend into the lastest patch BTW)

http://forums.bistudio.com/showpost.php?p=1546711&postcount=58

You don't suffer for the dellusion that multicore = 100&#37; usage per core used...do you?

Being obtuse again I see. The point is that performance improvement going from a dual-core to a quad-core is not even close to being linear. Even your link shows this:

C2Q 9650 24/26.7
C2D E8400 20/22.8

Double the cores at the same speed = ~17 - 20% improvement. Is that worth the additional cost? Your wallet may vary.

That the second post you have made to me with nothing relevant in it...care to make it a hat-trick?

Get over yourself. While I'll admit your recent post history was amusing, I had no idea who you were prior to 20 minutes ago.

 

[cbn]

Where is the law staing that the scaling must be linear?
Some threads need to wait for other threads to finish.
The fact that it gains performance tells you that it is using more cores.

That is true.

However, Dragon Age Origins scales 75&#37; higher FPS with a quad core over a dual core. Left for Dead 2 and Resident Evil 5 (among others) scale 50% better with quad core over dual core.

 

[Lonbjerg]

Seeing as they don't specify whether they're using a scripted benchmark, or how many times they ran through their test to eliminate variances, the results would seem as valid as the random forum posters.

Sure...you got the same validity as Anand eg.



http://forums.bistudio.com/showpost.php?p=1546711&postcount=58

Being obtuse again I see. The point is that performance improvement going from a dual-core to a quad-core is not even close to being linear. Even your link shows this:

C2Q 9650 24/26.7
C2D E8400 20/22.8

Using fallcies...you must be kidding...didn't you just accuse me of the same?!
+ performance = it's using the added core for better performance...not matter how you try and spin it.

Double the cores at the same speed = ~17 - 20&#37; improvement. Is that worth the additional cost? Your wallet may vary.

Did you not read the -cpuCount=x part? :thumbsdown:

Get over yourself. While I'll admit your recent post history was amusing, I had no idea who you were prior to 20 minutes ago.

Was that ontopic?
Pot, Kettle, black, eh?

 

[Lonbjerg]

That is true.

However, Dragon Age Origins scales 75% higher FPS with a quad core over a dual core. Left for Dead 2 and Resident Evil 5 (among others) scale 50% better with quad core over dual core.

That test was made without the cpuCount=x (x should be 4 on an bloomfield, 6 on a Gulftown) so they should run this test again, as HT and ARMA2 dosn't play nice.

 

[cbn]

That test was made without the cpuCount=x (x should be 4 on an bloomfield, 6 on a Gulftown) so they should run this test again, as HT and ARMA2 dosn't play nice.

How much better would Arma 2 do if cpuCount=x was changed?

Would it scale better than Dragon Age Origins or some of the other games out there?

 

[Lonbjerg]

How much better would Arma 2 do if cpuCount=x was changed?

Would it scale better than Dragon Age Origins or some of the other games out there?

I don't have Dragon origons so can't say anything about that game.

I could try it when I get home and have had some sleep (been at work all night).

I take you want a
cpuCount=2
&
cpuCount=4
&
cpuCount=8

Style bench?

I can't promise that I get the time though, am currently moving from Denmark to Sweden, so there is a lot of stuff I need to do, but will try and squeze in some time for this.

 

[toyota]

kind of silly to make the conclusion that games only need a dual core based on testing just a few games and only using an i7 with cores disabled. IMO if building a higher end gaming pc from scratch at this point it would be foolish to use anything other than an i5/i7. that will handle any game now and for the foreseeable future and will also allow you to fully take advantage of whatever gpu that you upgrade to during that time.

for somebody like me there is no real reason to run out in get a quad as my cpu does good enough 99% of the time. not to mention my gpu would also be the limiting factor in most games. only way stepping up to an i7/i5 quad would be of any real benefit for me as well as many others is if we wanted to get the most out of a very high end gpu. of course to some users a quad is beneficial in many other ways than just gaming too.

 

[Lonbjerg]

kind of silly to make the conclusion that games only need a dual core based on testing just a few games and only using an i7 with cores disabled. IMO if building a higher end gaming pc from scratch at this point it would be foolish to use anything other than an i5/i7. that will handle any game now and for the foreseeable future and will also allow you to fully take advantage of whatever gpu that you upgrade to during that time.

for somebody like me there is no real reason to run out in get a quad as my cpu does good enough 99% of the time. not to mention my gpu would also be the limiting factor in most games. only way stepping up to an i7/i5 quad would be of any real benefit for me as well as many others is if we wanted to get the most out of a very high end gpu. of course to some users a quad is beneficial in many other ways than just gaming too.

Add in that it looks like 50% af the games tested are console ports...

 

[pmurgs]

I think the article posted by the op suffers from a very major flaw. They use only one graphics card at a high resolution, hence they could easily end up gpu bound and not cpu bound in the test. Pretty much makes the article and the results somewhat pointless.

Test should of been down with tri-fire or tri-sli and at a lower resolution and lower gfx settings.

For instance, when playing BC2 online, I go from a minimium fps of 30 with my quad core (Q9400) at 2.66Ghz (stock) to minimium fps of 50 with my quad core at 3.9Ghz. The average also increases significantly. This is on a single overclocked 5850 with all gfx settings on max. The article the op posted says you only need a dual or tri-core before bc2 maxes out.

 

[toyota]

I think the article posted by the op suffers from a very major flaw. They use only one graphics card at a high resolution, hence they could easily end up gpu bound and not cpu bound in the test. Pretty much makes the article and the results somewhat pointless.

Test should of been down with tri-fire or tri-sli and at a lower resolution and lower gfx settings.

For instance, when playing BC2 online, I go from a minimium fps of 30 with my quad core (Q9400) at 2.66Ghz (stock) to minimium fps of 50 with my quad core at 3.9Ghz. The average also increases significantly. This is on a single overclocked 5850 with all gfx settings on max. The article the op posted says you only need a dual or tri-core before bc2 maxes out.

yeah you can see in this review that a Phenom 2 X4 demolishes an X2 at the same clockspeed in BF2 even at 1920 on high settings. thats a 50&#37; improvement just going to 4 cores from 2 for the Phenom 2 using a fairly mild gtx260. http://www.legionhardware.com/articles_pages/battlefield_bad_company_2_tuning_guide,3.html

 

[Scali]

An OS can't take care of multithreading.
It provides the means for creating and managing threads to the developer... but if the developer just writes a single thread, there's nothing the OS can do about it.
If splitting up workloads into parallel units was this trivial, we wouldn't even be having this discussion.

 

[Dribble]

I think the article posted by the op suffers from a very major flaw. They use only one graphics card at a high resolution, hence they could easily end up gpu bound and not cpu bound in the test. Pretty much makes the article and the results somewhat pointless.

Test should of been down with tri-fire or tri-sli and at a lower resolution and lower gfx settings.

For instance, when playing BC2 online, I go from a minimium fps of 30 with my quad core (Q9400) at 2.66Ghz (stock) to minimium fps of 50 with my quad core at 3.9Ghz. The average also increases significantly. This is on a single overclocked 5850 with all gfx settings on max. The article the op posted says you only need a dual or tri-core before bc2 maxes out.

+1

Same comments in bit-tech - they use such a fast cpu that even with 2 cores it's maxing out the gpu. They should have used a slower cpu so that you don't end up graphics bound (e.g. a Q6600). That would have been a much better test of how many cores the games were really trying to use.

That said I don't expect they would use much more then 3 because the game was probably developed to run on the triple core xbox 360.