Best Cpu For Arma 3?

[IGhzI]

Im thinking a 3770k, what does everyone else think?

 

[BallaTheFeared]

i5, i7 is a waste of money for gaming.

Unless you do other things that benefit from HT, gaming currently isn't one of them though.

 

[Prey2big]

Easy answer for all games: Highest clocked haswell quad core.

 

[ShintaiDK]

Easy answer for all games: Highest clocked haswell quad core.

^^ This.

 

[Maximilian]

Easy answer for all games: Highest clocked haswell quad core.

^^ That.

 

[BrightCandle]

According to gamegpu.ru and my groups own tests a 4930k.

Arma 3 does benefit moderately from additional CPU cores, although it also loves clock speed. Ivy bridge shows some quite noticeable gains and is worth it over and above the sb-e. Haswell also shows gains but not enough to beat the 6 cores. Of course if you aren't pairing it with dual GPUs you likely won't notice except in certain missions, but arma 3 in real multiplayer missions is largely CPU limited for most people.

 

[ozzy702]

As many threads as possible with as high a clock as possible.

 

[Prey2big]

From http://gamegpu.ru/action-/-fps-/-tps/arma-iii-alpha-test-gpu/testovaya-chast.html
Its an alpha test though and it doesnt say what kind of benchmark sequence they are running.

 

[ph2000]

the cpu load graph is a bit meaningless without FPS

 

[frozentundra123456]

the cpu load graph is a bit meaningless without FPS

Scroll down to see the FPS data. Arma 3 beta cpu benchmarks.

 

[frozentundra123456]

According to gamegpu.ru and my groups own tests a 4930k.

Arma 3 does benefit moderately from additional CPU cores, although it also loves clock speed. Ivy bridge shows some quite noticeable gains and is worth it over and above the sb-e. Haswell also shows gains but not enough to beat the 6 cores. Of course if you aren't pairing it with dual GPUs you likely won't notice except in certain missions, but arma 3 in real multiplayer missions is largely CPU limited for most people.

I will accept your data that in your gameplay a hex core is better, but the game.gpu benchmarks I linked in the above post dont really show that decisively.

At ultra quality, 3970x is only 5 to 10% faster (depending on the quality setting) than the 2500k, but it also has about a 6% clockspeed advantage and more cache. So really in that test, very little improvement from more cores. Results in real games could be different for sure though, so I am not trying to discount your experience.

I really like that site, but I wish they would upgrade the intel cpus to Haswell. Not going to ivy made very little difference, but it would be nice to see results for Haswell, since there should be a noticeable improvement over 2 full generations.

 

[BrightCandle]

The test we did was in a single player showcase. In a multiplayer real game the 6 core is only about 15% faster. We confirmed that using -cpucount setting for the game to force it to use less. Alas that doesn't account for cache different and such.

I too wish they would add a 4770 to that list. They are the only guys I know that are testing a wide enough range of games to say anything useful for comparison and yet their stutter analysis and missing lastest CPU is a real shame.

 

[sm625]

From http://gamegpu.ru/action-/-fps-/-tps/arma-iii-alpha-test-gpu/testovaya-chast.html

What does this chart mean? It looks like gobbledegook to me. Is this chart saying that a 4 core cpu will have less maximum single core loading than either a 6 core or an 8 core? Because that's nearly impossible to believe. That's saying adding 4 extra cores can increase the load on a single core.

 

[Jacky60]

The test we did was in a single player showcase. In a multiplayer real game the 6 core is only about 15% faster. We confirmed that using -cpucount setting for the game to force it to use less. Alas that doesn't account for cache different and such.

I too wish they would add a 4770 to that list. They are the only guys I know that are testing a wide enough range of games to say anything useful for comparison and yet their stutter analysis and missing lastest CPU is a real shame.

I agree, I'm torn between pulling the trigger on a 4930K and a 4770K. The former costs 50% more but my trusty old 920 is pulling a pitiful 22-28fps in single player. I have 8 weeks to make a decision, here's hoping I have better data to go on come November.

 

[frozentundra123456]

What does this chart mean? It looks like gobbledegook to me. Is this chart saying that a 4 core cpu will have less maximum single core loading than either a 6 core or an 8 core? Because that's nearly impossible to believe. That's saying adding 4 extra cores can increase the load on a single core.

Its pretty straightforward. What it shows is that as you get fewer cores, the load on each core increases. In this particular game, and a lot of others, even with higher core count, you are also dependent on single core performance. You also however need to look at the framerate graphs to see what the framerate is.

This is also why FX is at a disadvantage in a lot of games that claim to be multi threaded. Even though a game may use a lot of threads or cores, the workload has to be perfectly balanced for the FX to shine. Otherwise the faster per core performance of Intel will be faster, when the game is limited by one task.

 

[PPB]

Yeah, in a few words: Crappy thread load balance in a multi-threaded aware engine. Exactly what PS2's dev was referring to. Only that, as usual with Bohemia, it's taken to next level of game engine crapiness.

 

[sm625]

So why would it have better load balancing on a 4 core configuration. You got 45,78,36,46 on 4 cores but then you enable the other 4 cores and all the sudden one core is at 90%? It kinda looks like the primary game engine thread is also the same thread that handles the load balancing, so when you add more cores it makes more work for that thread pushing it up near 100% where it is probably limiting framerates. That's probably why there's almost no improvement going from 4 cores to 8 cores on AMD.

 

[PPB]

So why would it have better load balancing on a 4 core configuration. You got 45,78,36,46 on 4 cores but then you enable the other 4 cores and all the sudden one core is at 90%? It kinda looks like the primary game engine thread is also the same thread that handles the load balancing, so when you add more cores it makes more work for that thread pushing it up near 100% where it is probably limiting framerates. That's probably why there's almost no improvement going from 4 cores to 8 cores on AMD.

On those tests, there is no info regarding if they were taken at the very same time during the game test, in the same area, etc.

In general, thread load tests are, at best, dubious. Most of the time they might help you get a general pan of the situation regarding the game's engine and CPU scaling, but hiccups like this one aren't strange

 

[Jacky60]

What does this chart mean? It looks like gobbledegook to me. Is this chart saying that a 4 core cpu will have less maximum single core loading than either a 6 core or an 8 core? Because that's nearly impossible to believe. That's saying adding 4 extra cores can increase the load on a single core.

It's saying AMD CPU's are turds for Arma 3 (and incidentally Arma 2). Single threaded performance and IPC are so piss poor they probably should not be in the house.

 

[tweakboy]

Pentium D @ 5Ghz .......... just a little humor for this thread. Thanks.

Best CPU is Sandy Bridge 2600k or higher and it will run just fine not using all the threads. gl

 

[BrightCandle]

The more useful chart is this comparing the CPUs:

Here we can see a very slight advantage to the 6 cores but its really not worth talking about. However this is in a single player environment in the middle of beta when everyone in tier1 agreed they seemed to have broken the engine from the alpha to the beta. They have now fixed it again for release and its back to scaling a bit better and running a lot better than it did during the beta.

In multiplayer, in real games using all 6 cores is better than limiting the game to 4 cores. The average FPS is most of the time improved by around 15% and the minimum also. However that is SB-E compared to SB-E. That 15% will almost certainly be eroded by the difference in architecture that Haswell has compared to IB-E, certainly much of the advantage will be gone. Practically I don't think there is much point getting a 6 core for the game, the price/performance isn't worth it, but it is the fastest solution we know of (we run a heavily scripted highly modded version of the game, as I think do most Arma communities).

BIS have actually published quite a lot of details so to why they have this poor multithread usage and what they intend to do about it. But this is a problem they have been working on for years and its not easy to fix. Arma has one of the most advanced simulation engines ever made and its extremely complicated. They haven't yet found a nice way to utilise those additional threads for the AI (which is our biggest concern) or players for the server. They are working on a course grained threading approach but it likely wont work past a relatively small number of cores (4-6 I guess). Arma is a prime example of why so many games aren't multithreaded, its intrinsically difficult and it might be impossible (mathematically provable) to use multithreading fully in many games.

I really like the way Arma looks, its got a realistic lighting system that isn't over done or highly stylised. But it really does need a fast CPU and preferably an overclocked CPU to run well.

 

[Jacky60]

BrightCandle do you know how well Arma3 is using multithreading for network play? You and your community have clearly experimented. I know Arma 3 usually prefers HT off but I read something from BIS about trying to offload stuff to other threads like AI for Arma 3. For example it's really difficult to test HT on vs off on someone elses server with constantly changing other variables I'm really curious to know what you've discovered,I want to know if on a network with all the chat and sounds and peoples pings changing HT is better? Everything's looking a lot better/faster here after some tweaking, I still can't really believe how good it looks.

 

[SPBHM]

probably a 2500-3570-4670K with a lot of overclocking is the best way to go...

http://pclab.pl/art54599-4.html

 

[BrightCandle]

Arma 3 chooses the number of CPUs it is using to mostly avoid HT. Back in the A2 days that was a big performance boost, the game would loose a lot of performance with HT on. Nowadays its less exciting but still a net detriment and is usually best left off. Thus CPUs that are 4C/4T with the highest clock speed possible (overclock it is possible, it helps quite a bit) are probably the best bang for the buck. The game automatically selects to turn off HT when it starts but you can force it on again but its a bad idea.

On the server side we tested with a variety of core numbers and found the game basically only uses 2 cores, on the clients it does use more than that. A few different configs with a PvP map and determined you could fit a certain numbers of players on before server FPS dropped and the game started to crash:

Xeon 2.0 Ghz (Virtualised) - 28
Opteron (recent) 3.0 Ghz (Virtualised) - 12
3570 (bare) - 60
2500k (bare) - 55

AI counts along with that on the 3570 are in and around 150 AI before we start having issues. Bare in mind if you test on other peoples servers that they often have many game instances running on the same box and often try to get near 100% CPU usage. The issue there is of course any single game could become CPU constrained and we have found that once a server drops below 35 FPS that the client game FPS starts dropping as well. By the time server FPS is 20 the game FPS is normally halved and around 10 they are basically both 10. So server FPS really matters in this performance testing, it needs to be a constant 50.

So basically AMD is kind of bad on the client but terrible on the servers. More modern and faster CPUs result in almost linear scaling. We don't have any Haswell's in the group yet so we haven't seen what benefit it will bring but we already know the game likes new architecture (our 2Ghz Opteron from 1900BC didn't even manage 5 players concurrently whereas in Arma 2 we had 30 no problem).

I did some tests with retail today on a mix of Ultra and Very High graphics on Altis and saw no more than about 5% boost in single player.

Last nights games was very smooth but I think that is just retail code verses beta code, its noticeably less spiky in its frame rates. We'll do another 2 organised missions this Sunday and I'll take some traces and see what release looks like 6 c verses 4c but there is nothing in the release notes about them changing anything multithreaded wise and my discussions with their developer tells me not before the end of the year. So I doubt I'll see any useful difference to the tests I did back in Beta and also in Alpha where it was about 15%.

 

[Jacky60]

Thanks for the heads up. I'm going 4770K at 4.6 with 9970 xfire or 7990 Quadfire or 780 tri SLI or something else bigger and better, hopefully that will keep the wolf from the door.