Who said Supereme Commander runs faster with a quad-core?

[Acanthus]

Originally posted by: lopri
I didn't get to the test yet - my apologies first.

At this point what I am curious about is whether this game really takes advantage of more than 2 cores, and if so how. So let's try to avoid GPU bottleneck issue. Of course we know after a certain point the game will be limited by GPU. (which game isn't?) That is rather a moot issue for modern games. One can make any game either GPU bound or CPU bound by manipulating various in-game/driver settings.

One thing that I find interesting from the linked reviews is that they used a quad-core for both scenarios (dual vs quad), via a method of disabling cores in the BIOS. One would think a site the size of HardOCP can afford E6600/E6700, instead of going the route of disabling cores via BIOS? It might or might not matter, but I found that to be a little odd.

Supcom has 4 active threads at all times, it just distributes the load of the threads on the available cores.

 

[btdvox]

Originally posted by: lopri
I didn't get to the test yet - my apologies first.

At this point what I am curious about is whether this game really takes advantage of more than 2 cores, and if so how. So let's try to avoid GPU bottleneck issue. Of course we know after a certain point the game will be limited by GPU. (which game isn't?) That is rather a moot issue for modern games. One can make any game either GPU bound or CPU bound by manipulating various in-game/driver settings.

One thing that I find interesting from the linked reviews is that they used a quad-core for both scenarios (dual vs quad), via a method of disabling cores in the BIOS. One would think a site the size of HardOCP can afford E6600/E6700, instead of going the route of disabling cores via BIOS? It might or might not matter, but I found that to be a little odd.

Well comparing to all other benches (many have been posted) HardOCP is the only site that has huge differences and it makes me think that there doing something wrong by changing cores on the bio. Its not a coincidence that there the only site thats using the same Quad CPU and are getting higher results whereas the other sites got Dual core and Quad core cpu's and had much better results on the Dual cores.

 

[lopri]

I'm curently trying to get some meainingful consistency out of these data.. So far I can't find any. It's very weird, to say the least. Will post the resluts real soon. (sorry.. bed time)

 

[Acanthus]

Originally posted by: btdvox

Originally posted by: lopri
I didn't get to the test yet - my apologies first.

At this point what I am curious about is whether this game really takes advantage of more than 2 cores, and if so how. So let's try to avoid GPU bottleneck issue. Of course we know after a certain point the game will be limited by GPU. (which game isn't?) That is rather a moot issue for modern games. One can make any game either GPU bound or CPU bound by manipulating various in-game/driver settings.

One thing that I find interesting from the linked reviews is that they used a quad-core for both scenarios (dual vs quad), via a method of disabling cores in the BIOS. One would think a site the size of HardOCP can afford E6600/E6700, instead of going the route of disabling cores via BIOS? It might or might not matter, but I found that to be a little odd.

Well comparing to all other benches (many have been posted) HardOCP is the only site that has huge differences and it makes me think that there doing something wrong by changing cores on the bio. Its not a coincidence that there the only site thats using the same Quad CPU and are getting higher results whereas the other sites got Dual core and Quad core cpu's and had much better results on the Dual cores.

Was HOCP using replays to test?

If so, then no AI is being used, and entire thread of load missing.

 

[Acanthus]

@ the OP, youll see much better gains from a more powerful vid card, the 8600GT isnt exactly a powerhouse.

If youre severely GPU limited, it doesnt really matter how many cores you have.

Quad core certainly does see large gains on systems that arent bottlenecked by the GPU.

 

[Nathelion]

The size of the map and the number of units on the map is primarily what determines CPU stress in SupCom. As mentioned, the AI is also an EXTREME resource hog. To test CPU limitation one should simply run with all the eye candy turned of at a low resolution...

 

[btdvox]

^^ But why? For the sake of seeing if a Quad is working better than a Dualcore? If you dont play with Eye Candy Turned down and at a Low res whats the point?
I dont think anyone should pay attention to benches that mean nothing to them.

 

[lopri]

Originally posted by: Acanthus
@ the OP, youll see much better gains from a more powerful vid card, the 8600GT isnt exactly a powerhouse.

If youre severely GPU limited, it doesnt really matter how many cores you have.

Quad core certainly does see large gains on systems that arent bottlenecked by the GPU.

Originally posted by: lopri
Found the DVD and managed to install the game on the other rig. The two rigs compared for this game is:

1. Q6600 / NF680i / 8GB RAM / 8800 GTX / Single Raptor (148GB) / Windows XP
2. E6600 / 975X / 2GB RAM / 8800 GTX / Quad-Raptors (4 x 74GB) in RAID0 / Windows XP

Please understand that the rig #2 isn't mine so I can't change the configuration on it. I dual-boot XP and Vista 64 on rig #1 so if anyone's interested I could run the test on Vista, too. It looks like the lowest resolution the game allows is 1024x768 so that's where I am going to test both rigs. What I can't figure out is how to test it correctly. Is there a timedemo funtion built in this game? If not, I will probably download a multiplayer (2v2 or 3v3) replay somewhere and run with FRAPS.

 

[lopri]

I have run the built-in benchmark test with E6600 rig first. The numbers didn't make sense so I had to run several times per setting. (and this thing takes almost 10 mins per run!) Someone tells me what these mean.

E6600 @3.60GHz (1333 strap): http://img502.imageshack.us/my...e=e660036001333ro1.png

E6600 @3.20GHz (1333 strap):
http://img358.imageshack.us/my...e=e660032001333gq9.png

E6600 @3.20GHz (1066 strap):
http://img115.imageshack.us/my...e=e660032001066wd5.png

 

[lopri]

Easier way to make sense from above (as far as the effect of CPU goes) was to read the count of 'logged frames'. (whether it's a correct way or not, I don't know) Too many jargons in the analysis text for me to understand and you can't go by the 'scores' or even FPS that's for sure.

 

[lopri]

Originally posted by: Acanthus
Supcom has 4 active threads at all times, it just distributes the load of the threads on the available cores.

It doesn't look that way even from a site that shows a huge advantage for the quads.

http://enthusiast.hardocp.com/...w4LCxoZW50aHVzaWFzdA==

I know the Windows Task Manager isn't the best way to read loads on each core but at least it could give a ball-park idea? Do you know any better way to measure the loads on each core? If so please let me know.

 

[firewolfsm]

Looking at that, I guess you could say a quad uses about uses 200-250% of one core and a dual uses 140-160%.

Eh...

 

[stevty2889]

Originally posted by: AWhackWhiteBoy
Please use the SupCom Core Maximizer, its a supplement to create perfect load balancing across all cores. It's valuable even for a duel core system.

I actualy found that on the GPG forums, and started using it. While my FPS didn't really increase, on large maps, and in multiplayer games with a large amount of units on the screen, the lag and slow downs I was getting completely disappeared. I was normaly getting full usage of core 1, and partial usage of core 2. After starting the application, core 1 was barely being used, but core 2 was being used more, as well as core 3 and 4 being balanced. The lag and slowdowns are gone. Total CPU usage got over 50% on occasion, but most of the time, it is only effectivly using 2 cores. My roomate ran the program on his E6600. It was orginaly using all of core 1, and part of core 2, while now the 2 cores are more balaced, and the lag and slowdowns are gone for him as well.

My Specs:
Q6600 @3.1ghz
6gb ram
Vista x64
8000GTX
2x160gb WD sata II 16mb raid-0

His specs
E6600 @3.4ghz
2gb ram
XP pro
2x8800GTX SLI
2x400gb SATA II raid 0

Both running at 1900x1080 at max settings.

 

[btdvox]

Article on Anandtech shows the E6850 beating Supreme commander when pitted against the Q6600?!?!?!

http://www.anandtech.com/cpuch...owdoc.aspx?i=3038&p=15

 

[theprodigalrebel]

We ran Supreme Commander at 1024 x 768 with medium quality presets. We've changed our Supreme Commander benchmark a bit at the advice of Gas Powered Games' senior engineers. We created a skirmish with 7 AI players and let them duke it out for almost 30 minutes, we saved the replay and then measured the amount of time it took to playback the recording at maximum gamespeed (+10). The figures below are expressed in minutes, lower playback time being better:

That's a strange way to bench a game.

 

[RussianSensation]

Toms Hardware = 1024x768 60 sec gameplay

Q6600 2.4ghz = 49.45fps
E6600 2.4ghz = 40.17fps
E6850 3.0ghz = 50.27fps

 

[superbooga]

Originally posted by: theprodigalrebel

We ran Supreme Commander at 1024 x 768 with medium quality presets. We've changed our Supreme Commander benchmark a bit at the advice of Gas Powered Games' senior engineers. We created a skirmish with 7 AI players and let them duke it out for almost 30 minutes, we saved the replay and then measured the amount of time it took to playback the recording at maximum gamespeed (+10). The figures below are expressed in minutes, lower playback time being better:

That's a strange way to bench a game.

That's how I've always benched the game. It really shows the differences between cpu speeds. In the beginning, when there are very few units, all the CPUs will playback the demo extremely fast (about x10 speed). But once more and more units appear, the speed drops. On a slower CPU, the drop from maximum speed occurs much earlier than it does on a faster CPU.

 

[Laminator]

Originally posted by: btdvox
Article on Anandtech shows the E6850 beating Supreme commander when pitted against the Q6600?!?!?!

http://www.anandtech.com/cpuch...owdoc.aspx?i=3038&p=15

The clock speed on the E6850 makes up for the cores. At the same clock speeds, the cores a slightly faster (look at the bar graph).

 

[Acanthus]

Originally posted by: btdvox
Article on Anandtech shows the E6850 beating Supreme commander when pitted against the Q6600?!?!?!

http://www.anandtech.com/cpuch...owdoc.aspx?i=3038&p=15

3.0ghz vs 2.4ghz.

 

[Acanthus]

Originally posted by: lopri

Originally posted by: Acanthus
Supcom has 4 active threads at all times, it just distributes the load of the threads on the available cores.

It doesn't look that way even from a site that shows a huge advantage for the quads.

http://enthusiast.hardocp.com/...w4LCxoZW50aHVzaWFzdA==

I know the Windows Task Manager isn't the best way to read loads on each core but at least it could give a ball-park idea? Do you know any better way to measure the loads on each core? If so please let me know.

It doesnt maximise load on each core, it just distributes the load across them haphazardly.

2 of the threads are much greater load.

For the best (but not so accurate) way to bench performance, id put 5 or 6 AIs against each other on a skirmish map and save the game late game (when the load is the greatest) and see where the quad core lands.

Obviously the AI isnt going to the do the same thing every time, so there will be a margain of error, it will however give you an idea of how each should perform under real conditions.

Im geting a Q6600 when the price drop hits, and 4GB of DDR2 800.

Ill be glad to do E6600 vs Q6600 stock and OCed in supcom on vista ultimate x64.

Rig:
Either CPU mentioned above
Gigabyte 965P-DS3 version 2.0
4GB DDR800 (2x2GB)
Azalia Integrated Audio
Integrated LAN
Geforce 8800 GTS 320MB

 

[btdvox]

Overclocking both CPU's is mainly 2.8-3.0 GHZ (Q6600) vs 3.8-4.0 GHZ (E6850)!!

 

[Acanthus]

Originally posted by: btdvox
Overclocking both CPU's is mainly 2.8-3.0 GHZ (Q6600) vs 3.8-4.0 GHZ (E6850)!!

3.8+ghz is rare on a 6850.

3.6ghz ish is more common.

 

[imported_kiwik]

Originally posted by: theprodigalrebel

We ran Supreme Commander at 1024 x 768 with medium quality presets. We've changed our Supreme Commander benchmark a bit at the advice of Gas Powered Games' senior engineers. We created a skirmish with 7 AI players and let them duke it out for almost 30 minutes, we saved the replay and then measured the amount of time it took to playback the recording at maximum gamespeed (+10). The figures below are expressed in minutes, lower playback time being better:

That's a strange way to bench a game.

That's the best way to bench this game. The simulation speed is independant from the FPS. Simulation is the work that is concentrated on the CPU. Setting it at X10 forces the simulation to go up to 10X as fast. If the CPU becomes incappable of simulating at X10, it will drag down to X9, then X8, then it goes to 0 (which is normal speed, and normal time). If the CPU become really overloaded, then the sim will go from -1 to -10. At -2 it becomes a slideshow so imagine how it is at -10.


You need to be really familiar with the game to understand what I've explained tho...

 

[bryanW1995]

Originally posted by: btdvox
Overclocking both CPU's is mainly 2.8-3.0 GHZ (Q6600) vs 3.8-4.0 GHZ (E6850)!!

if you compare a b3 stepping q6600 to a G0 stepping e6850 then it's a pretty tough decision b/c the G0 looks like it's going to overclock so much better, but try comparing a G0 stepping Q6600 to the e6850 and you're not likely to end up much more than the original 600mhz apart.

 

[Ryan Smith]

Originally posted by: theprodigalrebel

We ran Supreme Commander at 1024 x 768 with medium quality presets. We've changed our Supreme Commander benchmark a bit at the advice of Gas Powered Games' senior engineers. We created a skirmish with 7 AI players and let them duke it out for almost 30 minutes, we saved the replay and then measured the amount of time it took to playback the recording at maximum gamespeed (+10). The figures below are expressed in minutes, lower playback time being better:

That's a strange way to bench a game.

To be fair, that's what I thought at first too when GPG told us about that method. However given some thought it makes sense because SupCom's framerate isn't directly tied to the speed of the simulation. The exact note we got from GPG was the following:

1. Create a game with a large number of Ais and let them duke it out.
2. At some point exit and save the replay to 'test.scrply'
3. Run the game with "/replay test.scrply /exitongameover" on the commandline.
4. When the game is up, type 'wld_RunWithTheWind' on the console and increase the gamespeed to +10.
5. Time how long it takes to finish the replay and exit.

Generally we sync the simulation thread with the rendering thread based on game speed. With wld_RunWithTheWind enabled, the sync throttle is turned off and the sync happens as soon as the Sim finishes its tick.
You still need to jack the game speed up to force the game to simulate faster, hence the instruction to raise it to +10 (the fastest we allow).