Testing HOI late game slowdowns

[bholstege]

To summarize what we know so far

What we know:
-Clock speed on the CPU matters, but only up to a certain point.
-If you have less than 1gb of RAM, the game will page the hard disk, reducing performance
-CPU cache probably has an effect
-just using counters generally has no positive effect, you have to actually delete the files
-more than 2 cores probably is not beneficial (not sure on this?)
-higher difficulty slows down the game

What we need to find out:
- at what point does CPU speed stop mattering?
- what effect does speed and latency of RAM have?
- does AI settings at normal difficulty effect the game?
- what effect does the number of countries have? (IE, is 100 countries with 10 units slower than 1 country with a 1000)
- does music slow down the game? ( also do sound cards with dedicate hardware like x-fi's change this)

 

[unmerged(76495)]

To summarize what we know so far

What we know:
-higher difficulty slows down the game

Actually, from what I´ve seen what you tested was aggressiveness. Not difficulty level. And it seems very logical that higher aggressiveness equals more computations. On higher levels, the AI is far more willing to do attacks, often not being able to wait for dawn to press the assault. I don´t think high AI aggressiveness in on itself cause it to make more calculations, it is in the indirect effect. Higher aggressiveness cause more and longer battles which use up computering power and thus slow the game down.

 

[sbr]

I ran the same test with Tibet and it took 2:26.

Couldn't tell you exactly what CPU I have though. I think it is pretty good.

Right click on "My Computer", then select properties. That window will show you your CPU. Let us know what it is, and how much RAM you have.

Intel Core 2 CPU
6600@2.40 GHZ
1.98 GB of RAM

 

[bholstege]

Actually, from what I´ve seen what you tested was aggressiveness. Not difficulty level. And it seems very logical that higher aggressiveness equals more computations. On higher levels, the AI is far more willing to do attacks, often not being able to wait for dawn to press the assault. I don´t think high AI aggressiveness in on itself cause it to make more calculations, it is in the indirect effect. Higher aggressiveness cause more and longer battles which use up computering power and thus slow the game down.

I tested both at the same time, Normal/Normal vs VH/Furious

 

[bholstege]

I'm redoing the tests as tibet with difficulty and AI separate.

Results:
Normal/Furious: 2 minutes 26 seconds
VH/Normal: 2 minutes 44 seconds
Normal/Normal: 2 minutes 13 seconds

So they both have an effect, although the difficulty settings effect is far greater.

 

[Lennartos]

Current benchmark list:
4 cores @ 3.7Ghz 12MB cache= 1min 25 sec
4 cores @ 2.4Ghz 8MB cache = 2min 26 sec
2 cores @ 3.0Ghz 1MB cache = 2min 30 sec
1 core @ 2.2Ghz 0.5MB cache=3min 28 sec

Lets do some preliminary math here:
- i will spare you the raw numbers-

it seems the game scales with the cpu speed nicely.
there is big speedup from 2cores to 4cores but the net gain is covered by the much bigger cache.

conclusion of this preliminary research:
To fully maximize your speed you should get:
A triple core as the game seems to have 3 main threads.
More Mhz = better speed, more cache = better speed.

What remains to be tested is the speed relation to RAM I/O throughput and latency.

 

[unmerged(66346)]

Another thing that came to my mind : Is it realy the number of units on the map that slow down the game or the number of nations controling them ?

 

[Lennartos]

Another thing that came to my mind : Is it realy the number of units on the map that slow down the game or the number of nations controling them ?

In all propability its the AI that is secound guessing itself every hour.
So for every hour it is calculating wherther or not it should send unit x to province y, or Z... should it make a strategic redeployment, split the current group, and so on.

reducing the units in game will lower the load, but the AI depends on a lot of factors, so whether or not we reduce or even increase the load when merging countries is hard to answer.
But reducing Units and provinces should reduce the AI load, therby giving speed.

 

[bbasgen]

Great summaries bholstege and Lennartos. I would like to see a few more results, especially from some older CPUs, to get a high level of confidence in our findings.

Our results so far help to get an idea for things. I think for years folks have been frustrated that the latest processors don't *seem* to dent game performance, perhaps slimming off only a few seconds per month. Naturally, however, when the Core 2 Duo's came out, they started blowing those conceptions out of the water -- and this is especially clear with the Quad core results. That is a *substantial* improvement.

Regarding further testing:
Lennartos: It seems the most practical way we can test RAM is for people with similar processors but different RAM to step forward.

Bholstege: I'm not sure this point is worth pursuing. It is a point of interest, but I don't see a practical utility to knowing the answer:
- what effect does the number of countries have? (IE, is 100 countries with 10 units slower than 1 country with a 1000)

 

[bholstege]

Another thing we need to keep in mind is the processor architecture. A Pentium 4 at 3ghz us low end, a Core 2 at 3ghz is state of the art. I have a Pentium D at 2.8 on my backup machine, so if someone would like to test a more modern CPU at that speed, we can see the effect of the architecture.

Also, my motherboard can under clock the ram, so I can try running it at a lower speed. Will do that tonight, or maybe on the weekend.

Also, I have a celeron 440 coming in, which is based on the core architecture, but is a single core. We can compare this to the dual cores to see what benefits more cores give us on the low end.

 

[bholstege]

and this is especially clear with the Quad core results. That is a *substantial* improvement.

We also need to be aware that that test was run on a Raptor, which is noticeably faster than standard 7200rpm drives. We need to take into account the fact that hard drive speed may influence performance, since the game loads lots of data, and may not put it all into RAM.

 

[bholstege]

Intel Core 2 CPU
6600@2.40 GHZ
1.98 GB of RAM

do you know if it a Q6600, or a e6600? Open task manager, and see how many CPU usage boxes it shows,

 

[unmerged(44926)]

I am using a P4 2.6ghz with 1.5gb of RAM. It's due for an upgrade (building a new PC is this summer's project) but until then, I can use it to run a couple tests... What tests are you guys looking for?

On another note, my modding skills are rather limited, but I've been toying with the idea of creating a simple mod that removes all of the neutral minors and replaces them with one country simply marked "Neutral." The effect would be similar to the blank countries on the board for the board game Axis & Allies.

I would remove all IC, VP and resources from this "country" and set all remaining nations to ignore it diplomatically. I imagine that most Carribean, South and Central American nations would be removed, with the exceptions of Brazil, which did historically join the allies and make a notable contribution to the war effort, and Venezuela, which is often a useful partner for trading oil. From Europe I'm tempted to neutral out Switzerland, and in Asia Afghanistan, Saudi Arabia and Turkey can probably all go without interfering in a historical game.

Of course this simple mod would remove ahistorical options, possibly screw up international trade, and probably look very unpleasant... But for those interested in speeding up their game, it might improve the late game up some by removing many (usually meaningless anyway) calculations.

Any interest in such a project?

 

[sbr]

do you know if it a Q6600, or a e6600? Open task manager, and see how many CPU usage boxes it shows,

Just one CPU usage box.

 

[bbasgen]

I am using a P4 2.6ghz with 1.5gb of RAM. It's due for an upgrade (building a new PC is this summer's project) but until then, I can use it to run a couple tests... What tests are you guys looking for?

This is our baseline:
1. Restart your computer
2. Open HOI2, ARMA, version 1.2
3. Select the Doomsday scenario, set for Normal/Normal
4. Select Tibet, and turn counters on
5. Run on "Extremely Fast" until November 1, 0:00
6. Time it with a stop watch utility
7. Repeat process to account for variances.

This will help us figure out the role of hardware -- forgetting for the moment how different game settings might effect results. After all, hardware being equal, the changed rates based on settings should also be equal.

On another note, my modding skills are rather limited, but I've been toying with the idea of creating a simple mod that removes all of the neutral minors and replaces them with one country simply marked "Neutral." The effect would be similar to the blank countries on the board for the board game Axis & Allies.

Interesting idea! The Axis & Allies mod. Wonder if anyone has done/tried it before? The main issue is how it would effect trading.

 

[unmerged(44926)]

This is our baseline:
1. Restart your computer
2. Open HOI2, ARMA, version 1.2
3. Select the Doomsday scenario, set for Normal/Normal
4. Select Tibet, and turn counters on
5. Run on "Extremely Fast" until November 1, 0:00
6. Time it with a stop watch utility
7. Repeat process to account for variances.

This will help us figure out the role of hardware -- forgetting for the moment how different game settings might effect results. After all, hardware being equal, the changed rates based on settings should also be equal.

On Normal/Normal, with counters on, but sprites not deleted... After 3 tests... *drumroll please* ... an average of 5:50 for my old P4 2.6ghz to run one month. Ack.

Mentioned above, one poster with a 2.2 ghz single core processor did a run in 3:28... I've been (or at least I think I've been) really good about keeping my machine free of crap, defragmented etc. etc... Any ideas why my run might be so bloody slow?

EDIT: For anybody interested in the "Axis and Allies" neutrals mod idea...
Click Here.

 

[bholstege]

Megahertz are not everything. The pentium 4's need to be about 40% faster than the Athlons, and almost 90% faster than the core 2's, to perform at the same level.

 

[bbasgen]

On Normal/Normal, with counters on, but sprites not deleted... After 3 tests... *drumroll please* ... an average of 5:50 for my old P4 2.6ghz to run one month. Ack.

Ouch! That is PAINFUL. That is 12 seconds per day! That is four times slower than our fastest time. Wow, that hurts. The funny thing is... your processor was top of the line when HOI2 was released!

Anyway, I was the one with the 3:30. Remember that Intel was far behind AMD for several years, so my Athlon 3500 is considerably faster than your PIV, even though they are both from similar generations. Nowadays, of course, Intel has blown AMD out of the water with the Core 2 Duos. All that said, you are still a bit higher than I would expect -- I would have though 4 - 4:30 would make sense. When was the last time you reinstalled Windows?

 

[bbasgen]

I have added our findings to date to the Wiki:
http://www.paradoxian.org/hoi2wiki/index.php/Technical_Issues#Game_Slow_downs

 

[MacLeod]

Oh, is my P4 2.4GHz computer no longer state-of-the-art? I even upgraded the memory to 1GB at one point.

I've run some tests, loading up as Cuba, and letting it run for one month with sprites:

Starting in the 1936 scenario, 2:04.
Starting in the Doomsday scenario, 8:18.

What I have done for a couple of years now is to modify the manpower in the misc.txt to one third.

# National Province Manpower Multiplier (income per day)
	0.0033
# Non-National Province Manpower Multiplier (income per day.) Note: Some types of ministers can affect this value.
	0.00033

This reduces the ridiculous amount of divisions and makes unit choices more important. It also delays end game slowdowns.

From various save games I have loaded and ran for a month as Cuba:

1943 - 4:50
1945 - 6:00
1947 - 6:35

The last one froze for 5 seconds as the Soviets decided to release puppets.

Probably all these saves have less units and less countries than the Doomsday scenario.