Forged Alliance Forever

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-26T18:42:05+02:00

We just want better performance... it may not be that we're confusing sim-speed with FPS, but rather since both contributed to a better play experience, we're looking at the net gain. Back when I had a lower-end machine, a little less sim-speed was worth not choking out when the ASFs started flying.

There are many reasons you may not want to bring that to a network game today where everyone else has fast computers, but back when I used it, my friends and I all used it on our slow crappy machines and it helped make the game smoother for us.

Statistics: Posted by Doompants — 26 Aug 2012, 18:42

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-26T14:50:00+02:00

rabidradish wrote:
I don't believe this works at all, and why should it? The simulation thread is what determines simspeed and it can only run on 1 core. Maximizing core usage by some kind of swapping threads around shouldn't help at all. All I hear is people saying "it seems to be working for me" but never any concrete evidence. Or they confuse better fps with better simspeed.

Yep, it's probably nothing but voodoo. I hear drinking a glass of cat blood or shaking a dead chicken over your PC has similar effects on performance!

If you run perftest a five times for supcom, you will get five different results. It's a complex game, with lots of race conditions.

Statistics: Posted by thygrrr — 26 Aug 2012, 14:50

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-25T16:34:14+02:00

I don't believe this works at all, and why should it? The simulation thread is what determines simspeed and it can only run on 1 core. Maximizing core usage by some kind of swapping threads around shouldn't help at all. All I hear is people saying "it seems to be working for me" but never any concrete evidence. Or they confuse better fps with better simspeed.

Statistics: Posted by rabidradish — 25 Aug 2012, 16:34

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-25T12:20:53+02:00

I have used Coremax for many years, 1st with a Q6600[WinXP] and currently with a 2700K[Win7].
Testing late game performance of AIx on a big map was near impossible without this tool, as gamespeed would choke to -4. Measuring the effect of coremax was complicated by the fact that Sorian and Duncan made much improved AIs and the community did bugfixs. Also AIx ran better under FA than original Supcom. Now FAF has improved AI performance even further allowing fast games against 2 teamed AIx on 20x20 maps. The tool defiantly helped my Q6600 with OS WinXP by spreading load over more cores ---> core 0 [60%] - core1 [85%] - core 2 [30%] - core 3 [10% or less] rather than seeing core 0 @ 100% without coremax. Back then AIx games would see a slowdown to barely playable at 45mins without coremax and at 1hr with it on 20x20 size maps.

I thought the 2700K with Win7 OS would be bulletproof and capable of even larger AIx games using 40x40 maps... but alas those games seem to take a speed hit at about 1hr and choke to between 0 and -2, even though you may have killed one of the AIx. On a relatively small 20x10 map playing two humans verse two teamed AIx with 1000units, I see a slowdown when the players and AIx are at or near the unit count limit. I noticed that my core 0 was 100% and yet other cores were budging at low load. So I figured Win7 load sharing wasn't working to well for me with FAF when lots of AI is present.

Ok - coremax installed - it now does not auto start with FAF like it did on GpGnet.
Manual startup each time
Open 'Options', set FA and set to use only real cores not virtual cores.
Ran a replay of big AIx game without a slowdown.
Core load spread looked better on core load graphs.
Tried a TeamPlayAI 20x10 map and the game seemed smoother, faster and the AIx smarter.
On 40x40 maps it doesn't help that much as you will still see a slowdown at 1hr - 1hr15min. Don't really know why ... seems to be much the same in a Skirmish.

Sometimes in multiplayer games I forget to start coremax and I feel the game is slowing so I [Ctrl+Esc], start coremax, check ren_shownetstats and I see a +2 speed increase for about 10 mins until it still slows with say core 2 at 100%. I think the main thing it does is move the networking thread to a core of its own, but once the sim thread hits 100% on any players core 0 ... you see lagg.

Now if I read this subjective dribble I would not believe any of it... the only way to determine the truth is try it yourself... maybe I have been 'just lucky' hehe.

The only assassination game I have ever won was with Coremax running ...lol...that should make some people at least try it.

roj

Statistics: Posted by AwarE — 25 Aug 2012, 12:20

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-24T01:07:15+02:00

LOL WUT? I missed something. Where is it you think I'm contradicting the author? And why the snark?

First off, I was just trying to relate my anecdotal experience for people who were curious about Maximizer, and had never tried this tool back in the day. But second, I experienced, more or less, precisely what the tool was made to do.If anything, maybe I explained something poorly that you've misunderstood, but if you read this blurb from the author himself, it explains clearly how the maximizer tool made possible an overall increase in performance even though you may also sometimes see a minor drop in simspeed. No contradiction.

MadBoris wrote:
Tainek wrote:So if i comprehend correctly, this increases your FPS, but lowers your sim speed?

It's a complicated issue...

You have to remember all that the tool does is spread threads out across cores, so that cores are not saturated where threads are usually starving and colliding. There is no real world hardware penalty being incurred there as was once considered, infact, just the opposite. Now why would that be a bad thing? It shouldn't be! But the engine 'shows' something strange at times in a potentially lower sim number on the same point in a game. There's several possible reasons for this but none of them fixable by me AFAIK yet.

Some would quickly answer yes to your question. But I actually think it is more accurate to say it "may" hit negatives sooner. It's only until you hit negative -1 before there is any real world impact in gameplay with or without the tool. Without the tool CPU 0 is crushed at this point and a player is really feeling the impact on multiple levels. Also if he doesn't have a fast CPU, networking code maybe hampered with CPU 0 in feeding clients, I dunno either way on that. The point is by the time you are hitting negative sim, which is all that really matters in sim speed, then the rest of the engine is really hurting big time on CPU 0, so this tool takes the stress off which could be beneficial to all involved, but definitely the individual player notices benefits most.

For instance, a replay can be played back at +10 that screams far faster than regular supcom at +10 without the tool. Now that is not an indicator of lower or faster sim speed but it shows the whole game is much more fluid and responsive. Also, where a game is chugging at 8 - 10 fps which is really unplayable, the tool shoots it up to 30 fps on fast hardware with no perceivable slow down but a hugely better feel all around and smoother game. Also the game runs all tests faster with the tool in actual time, unless certain GPU settings are enabled like AA has a wierd effect at times on real world time it takes a simulation to run it's course.

I would really recommend this 100% and that isn't because I made it and have some ulterior motive in pushing my tools, but because I tested it.

But I am all for people making up their own mind. I would say test it by getting a huge skirmish game going until it starts slowing down your machine until you hit negative sim or the game really starts suffering, then save it. Start it with and without the tool and make up your own minds how it plays. Or see what it is like online for yourself.

Statistics: Posted by Doompants — 24 Aug 2012, 01:07

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-22T20:02:35+02:00

Doompants wrote:
There is no contradiction. While the overall sim speed was slighty smaller to start, the game wouldn't nosedive into unplayable simspeeds in the long term.

Well that would very much contradict the authors benchmarks. Too bad you didn't let him know, because he could of at least used your exceptional setup to help debug the problem.

Statistics: Posted by Ghoti — 22 Aug 2012, 20:02

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-22T18:23:25+02:00

Ghoti wrote:
Perhaps windows 7's thread management is better behaved?

Another possibilty. I was back on XP at the time obviously.

Statistics: Posted by Doompants — 22 Aug 2012, 18:23

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-22T18:22:37+02:00

Ghoti wrote:
Doompants wrote:While it might lower the overall sim speed slightly, it DID *seem* to improve how much large games tended to slow down over time.
Could you resolve that contradiction for me? I don't understand. Games slowing down over time is the sim speed getting worse.

There is no contradiction. While the overall sim speed was slighty smaller to start, the game wouldn't nosedive into unplayable simspeeds in the long term.

Keep in mind, I am talking about my experiences on the first generation Core Duo processor, several years ago. Think it might've been a 6600. But to use Setons as an example, in a game where I hosted 6 of us (and 2 AI players), we could all play a nice fast, crisp game to start, but by midgame, we'd be starting to crawl kinda bad, and we'd be slogging at -4 before the experimentals started rolling. I agree it's purely anecdotal evidence, but we talked about it and we all agreed that core maximizer had the effect of dropping sim speed a small amount to start, but we could play a lot farther into the game before we hit horrible -4 sims. short term loss for long term game.

But I have no idea if there is still any truth to this, or benefit with today's CPUs, or if it was only because I was hosting AIs, or if any of the performance improvements made to FAF make a difference.

Statistics: Posted by Doompants — 22 Aug 2012, 18:22

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-22T17:09:36+02:00

tried as well. I think it was causing stuttering on my comp... strange.

Perhaps windows 7's thread management is better behaved?

Statistics: Posted by Ghoti — 22 Aug 2012, 17:09

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-22T16:37:17+02:00

Tried with and without. No difference or whatsover

Statistics: Posted by -_V_- — 22 Aug 2012, 16:37

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-26T15:25:31+02:00

i use it and it forces FA to use more than one core. i also tested it in large games with 7 AIs and 1000 unit limit. after 20min in the game, i get about 1-2 fps (this is almost independent of the setting of the graphic options). WITH core maximizer i get about 10 fps. i have a laptop with an i7 @ 1.6ghz.

i have not tested the simspeed, but this is not as important to me. i want my games playable and core maximizer does this.

i would like to have an option in FAF "use core maximizer", but there should be a warning as well saying it can actually slow down the simspeed and should only get used after extensive testing.

Statistics: Posted by eXcalibur — 22 Aug 2012, 15:57

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-22T10:08:45+02:00

Ghoti wrote:
Could you resolve that contradiction for me? I don't understand. Games slowing down over time is the sim speed getting worse.

Also a read through the threads on the application were interesting. I'll give it a wing.

Read: "id DID *seem*"
Contradiction solved.

Statistics: Posted by ExituS — 22 Aug 2012, 10:08

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-22T04:34:58+02:00

Doompants wrote:
While it might lower the overall sim speed slightly, it DID *seem* to improve how much large games tended to slow down over time.

Could you resolve that contradiction for me? I don't understand. Games slowing down over time is the sim speed getting worse.

Also a read through the threads on the application were interesting. I'll give it a wing.

Statistics: Posted by Ghoti — 22 Aug 2012, 04:34

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-22T04:04:33+02:00

I used to run this years ago on my older core duo, as did a couple friends. While it might lower the overall sim speed slightly, it DID *seem* to improve how much large games tended to slow down over time. So it was a tradeoff, but it was no miracle maker. Big games still slowed down somewhat and huge games still choked outright.

Nowadays, I have no idea how much of an impact, if any, it still has with FAF and more modern machines.

Statistics: Posted by Doompants — 22 Aug 2012, 04:04

Re: Coremaximiser tool for multi core CPUs/FAF

2012-08-22T03:31:22+02:00

now, call me crazy, but I'm pretty sure the guy who released it, said it would actually SLOW sim speed down.

edit: I remembered correctly.

There is a slight side effect where the sim runs a little slower. This is due to the nature of how GPG's two major threads work together in concert, Sim and Render. Unfortunately the game was not tuned with this kind of thread freedom expected on core 0 and when the render thread is allowed more freedom to the CPU time having huge benefits, the sim code slows down for internal reasons unknown.

http://forums.gaspowered.com/viewtopic.php?t=15556

Funny disconnect between what we expect to happen, and what actually happens.

Actually I've wanted to try it, but I'm not 100% on what exactly it's doing.

Statistics: Posted by Ghoti — 22 Aug 2012, 03:31