Forged Alliance Forever

Re: Lets make simspeed benchmarks (revised!)

2014-09-22T04:58:19+02:00

Yeah I'm here. I emailed the OP a copy of the time demo I used for my original thread months ago based on my original idea. He didn't reply. At least I got credit as "some dude." I've since jumped to Sins of a Solar Empire from Supreme Commander and it really doesn't matter what CPU or RAM you use these days...it's still a slide show with thousands of units. I just basically wanted to check-in and say hi to to all you supcom winners. It doesn't matter the screen resolution or video card it's all CPU/RAM lag. oh and "strategic launch detected."

Statistics: Posted by spacing_guild — 22 Sep 2014, 04:58

Re: Lets make simspeed benchmarks (revised!)

2014-09-04T11:27:05+02:00

Alright. a good start would be to open up cpu-z and look at the spd tab for specifications of your memory. The xmp profiles should be visible right there. Otherwise google your specs. You'll see that the auto timings your BIOS sets are quite relaxed. Change the CAS, TRCD, TRP, TRAS and TRFC and command rate to match the clockspeed of your memory. Set the recommended Vdimm (volts for memory).

Even though motherboard manufacturers supply software to change BIOS settings, always set them manually in BIOS!

Run a few memory intensive benchmarks before and after changing the settings. Measure bandwith.

Getting tighter timings on the same clockspeed isn't everything. Depending on your cpu architecture / platform key components like the memory controller are unfluenced by the base clock and desired multipliers. Same goes for the L3 cache on Bloomfield i7 cpu's (the 9** series primarily). Later i7's are different. I've been not touched AMD in a long while but the same concepts apply. Obviously, memory controller and cache speeds influence RAM intensive benchmarks as well.

All in all, my prediction is that increasing the ultimate performance of the memory, influence on FA will be small. Doing this you can see for yourself. Keep cpu Mhz the same and mesure simspeed performance with your faster RAM. It won't give you much at all I'm sure. It's just a general rule of thumb for cpu intensive tasks.

Here you'll find excellent Intel overclocking guides for Intel, going into memory settings as well: http://www.xtremesystems.org/forums/sho ... ndy-Bridge)

Statistics: Posted by Col_Walter_Kurtz — 04 Sep 2014, 11:27

Re: Lets make simspeed benchmarks (revised!)

2014-09-04T17:20:31+02:00

Alright, so GPU influence is probably little, except if you're running on a unrecommended card for this game...

Thanks for the offer, I surely will do that. I don't know much about latencies, but I'll read into that first before I PM you. It seems though as if there is a correlation between simspeed and the memory read uncached score in PassMark, as you can see here in the 2nd row of graphs: https://docs.google.com/spreadsheet/ccc ... web#gid=14 You said cached scores are independend of latencies right? Then perhaps I already know enough by everyone's RAM MHz? Sorry but I'm quite new to overclocking

Statistics: Posted by Dr_Doolittle — 03 Sep 2014, 21:52

Re: Lets make simspeed benchmarks (revised!)

2014-09-03T16:53:51+02:00

I don;t know why there is a difference minimized. It's possible that the engine has a setting for max background fps, the maximum fps rendered when the game is minimized. I know this exists in other games but I've never looked for this in FA.

If you need help figuring out RAM settings I can be of service perhaps. Shoot me a PM with your situation and I'll gladly take a look. I was pretty interested in the overclocing scene for the last 10 years or so. I lost interest in recent years but I know the drill.

Statistics: Posted by Col_Walter_Kurtz — 03 Sep 2014, 16:53

Re: Lets make simspeed benchmarks (revised!)

2014-09-03T14:12:23+02:00

@_VODKA_:

Thanks. Yes, in total I've seen this replay over a 100 times now ^^

What 1600MHz bench is beaten by a 1333MHz? I don't see that one. Perhaps you missed out that the 1333MHz bench was OC'ed or one was benched windowed and the other full screen... As you can see in the 'FS/W comparison' tab RAM speed surely does matter, though you're right about latencies. I've made a chart on the bottom that displays the difference for different memory speeds. You can translate the benchmark scores to simspeed by this series:
>0.75 = 0 simspeed
>0.416 = -1 simspeed
>0.29 = -2 simspeed
>0.225 = -3 simspeed

I supposed GPU is not working minimized because I've benched minimized and full screen and full screen windowed, and minimized was by far the fastest. Don't remember whether it was windowed minimized or full screen minimized though (there's some difference in between those 2). So if GPU is still working benching minimized, how do you explain this?

About latencies... I suppose eventually I have to ask people to report their latencies/exact memory name as well, as I really need to know what the individual influence of RAM is on the bench. I hope I can test that on my own PC after I learned how to change latencies?

Statistics: Posted by Dr_Doolittle — 03 Sep 2014, 14:12

Re: Lets make simspeed benchmarks (revised!)

2014-09-02T17:23:24+02:00

@ Doolitlle. A couple of observations.

First, great work, it looks like you spent a lot of time making this.

400Mhz (800 DDR), although not below JEDEC standard, is terrible and not used by anyone in the real world. 533Mhz is borderline unrealistic. You could use this to underline that RAM speed doesn't do much at all. Then I don't see anything about latency in your doc. Without that Mhz is a useless number, unless you're looking at uncore or cache speeds. Fact is that a 1600Mhz RAM is beaten by a 133Mhz RAM system with the same cpu in your results. Also, the 1066Mhz RAM system is alsmost equal to the 1866RAM system with the same cpu. EVerything points to a very loose relation between RAM speed and simspeed.

How is GPU a factor? I'm sot sure I'm following this or why you come to that conclusion. Of course the GPU takes some cycles off your cpu time, but the same is true for all processors regardless of the type of GPU. It's just that the effects are more noticeable on slower cpu's because their resources are starved faster. In other words running FA on a PIII will impact simspeed more than running it on a 4770k with the same GPU.

Lastly, where I said windowed I meant minimized in the last paragraph. You're assuming that minimizing the game somehow the GPU doesn't matter anymore? As I explained the GPU is still working when you minimize stuff. Therefore your cpu is still being used to feed the GPU.

Statistics: Posted by Col_Walter_Kurtz — 02 Sep 2014, 17:23

Re: Lets make simspeed benchmarks (revised!)

2014-09-02T12:33:45+02:00

Benchmarks at windowed mode are now welcome as well, though I need to know whether you used windowed or full screen as scores are different for some configurations.

EDIT:

I'd like to ask you to bench with your memory at 1333MHz, so RAM influence on the correlation calculations are minimized.

Now I'm using the last lap as the FA benchmark to compare with other benches, but I'm gonna change that to the 2nd and 3rd lap, because GPU influence is present in the last lap because of the last ASF spawn fighting the buggs. I exclude the 1st one because some people include the pauze and some exclude it (that's 4s bias).

Statistics: Posted by Dr_Doolittle — 02 Sep 2014, 12:33

Re: Lets make simspeed benchmarks (revised!)

2014-09-02T11:50:07+02:00

_VODKA_ wrote:
Wait have you found an indication that RAM matters at all for simspeed? I'd be surprised and tempted to look at other factors such as cache and uncore speeds before deciding fast RAM does any good.

Yes, RAM speed does matter. I tested different RAM speeds with the same CPU clock and got better results for more MHz RAM, as you can see here: https://docs.google.com/spreadsheet/ccc ... web#gid=14

_VODKA_ wrote:
And before looking at RAM I would also worry first about background stuff clogging up the machine, unoptimized OS and so on.

That's a factor, but all I can do is to ask people to close as much background programs as possible. It's undoable to let everyone list all the programs they've running and calculate how that would affect your score. Also, for most processors in Windows 7 those programs get allocated to another core than FAF is using.

_VODKA_ wrote:
What I'm trying to say is, cpu architecture X at speed Y should give predictable results. I'm using a Xeon W3570 @ 4.0Ghz and it's just proportionally faster than an i7 920 @ 2.66Ghz for example.

As said above, RAM and GPU also matter.

_VODKA_ wrote:
Dr_Doolittle wrote:That's why an ideal bench would be minimized, as in that cace the GPU wont affect your scores.

I'm quite sure the GPU is still taxed in windowed mode. following your logic you could run one benchmark like Furmark minimized while running 3dMark on screen. That wont work.

Offcourse it is, but you're not reading well what I said. I said benching the game MINIMIZED (not windowed) would exclude GPU influence on the bench. Benching this way only CPU and RAM matters. To bench GPU influence on simspeed, one would need a ton of GPU's and I only have 3.

Statistics: Posted by Dr_Doolittle — 02 Sep 2014, 11:50

Re: Lets make simspeed benchmarks (revised!)

2014-09-01T18:56:36+02:00

1- ASUS ROG MAXIMUS VII RANGER Z97 SKT 1150
2- CORSAIR HYDRO SERIES H110
3- SAMSUNG SERIES 840 EVO ULTRASLIM 120GB 2.5" SATA III
4- PROCESSADOR INTEL CORE I7 4790K 4GHZ 8 MB SKT 1150 BOX.( 4.6 OC )
5- WESTERN DIGITAL CAVIAR BLACK 500GB 7200RPM 64MB SATA III 3.5
6- ARCTIC COOLING MX-4 4G
7- KINGSTON PC3-19200 2400MHZ 16GB XMP BEAST CL11 (2X8GB)
8- Corsair Full tower
9- GTX 780 TI
10-XFX gold serie 1000w Modular

I done the benchmark with stock 4.0 but Dual intelligence Maximus dont let and always put up the Stk clock to 4.4 without OC

But lets see iff i done well this

1-0:22 sec in game 0:30
2-1:01 sec in game 1:00
3-1:39 sec in game 1:30
4-2:12 sec in game 2:00

Other thing i notice on save inf ,cpus 4/5/6/7/8 werent use , i think FAF engine is use only 3/4 cores?

Statistics: Posted by LittleInferno — 01 Sep 2014, 18:56

Re: Lets make simspeed benchmarks (revised!)

2014-09-01T14:50:55+02:00

Dr_Doolittle wrote:
Although as noted in my OP, I'd rather have stock speed benchmarks as they can be compared to other benches to achieve my ultimate goal: predict what a certain CPU and RAM configuration would perform for a certain match in FA

Wait have you found an indication that RAM matters at all for simspeed? I'd be surprised and tempted to look at other factors such as cache and uncore speeds before deciding fast RAM does any good. And before looking at RAM I would also worry first about background stuff clogging up the machine, unoptimized OS and so on.

What I'm trying to say is, cpu architecture X at speed Y should give predictable results. I'm using a Xeon W3570 @ 4.0Ghz and it's just proportionally faster than an i7 920 @ 2.66Ghz for example.

Dr_Doolittle wrote:
That's why an ideal bench would be minimized, as in that cace the GPU wont affect your scores.

I'm quite sure the GPU is still taxed in windowed mode. following your logic you could run one benchmark like Furmark minimized while running 3dMark on screen. That wont work.

Statistics: Posted by Col_Walter_Kurtz — 01 Sep 2014, 14:50

Re: Lets make simspeed benchmarks (revised!)

2014-08-30T16:14:00+02:00

Hi,

here are some benchmarks of my new configuration (graphics AT 4850, RAM G.Skill F3-2400C10D-16GTX CL10-12-12-31)
I tested OC to 4400 and stock @3500 Mhz as well as RAM @1333 and @2400 Mhz.

Mhz
4690k4400
16GB 2400
30
34,5 | 64,5
42,65 |107,15
53|160,15

Mhz
4690k3500
16GB 1333
30
38,5| 68,5
51,7| 120,2
68,9| 189,1

Mhz
4690k3500
16GB 2400
30
37,32 |67,32
48| 115,32
64| 179,32

Mhz
4690k4400
16GB 1333
30
37,2|67,2
48,8| 116
64,3 |180,3

Bye Jens

Statistics: Posted by JensJensen — 30 Aug 2014, 16:14

Re: Lets make simspeed benchmarks (revised!)

2014-08-29T14:50:25+02:00

Yes, but as you can run this game with a decent 7 year old graphics card I think we can take the risk...

Honestly I think it's way to difficult to also bench the influence of a GPU on this replay. That's why an ideal bench would be minimized, as in that cace the GPU wont affect your scores. This is one of the most important reasons I wanted to make a new benchmark thread, as you can see some posts back.

Statistics: Posted by Dr_Doolittle — 29 Aug 2014, 14:50

Re: Lets make simspeed benchmarks (revised!)

2014-08-29T14:41:26+02:00

Isn't there a (small) graphics component too?

Statistics: Posted by E8400-CV — 29 Aug 2014, 14:41

Re: Lets make simspeed benchmarks (revised!)

2014-09-02T12:08:37+02:00

Thanks Shamble for the benchmark. I've added it in the spreadsheet. Although as noted in my OP, I'd rather have stock speed benchmarks as they can be compared to other benches to achieve my ultimate goal: predict what a certain CPU and RAM configuration would perform for a certain match in FA.

I've been doing some research lately on why windowed has other scores than full screen. Also, I want to be able to convert windowed score to full screen score. It seemed unpredictable but I've found a correlation now with memory read uncashed scores in PassMark Performance Test. For my pc (FX 6300 and pc3 17000 Gskill RipjawsX) windowed became faster than full screen when read uncashed was above 6800Mb/s (around 1500-1800Mhz, depending on overclock). Under that score windowed became slower. Though speeds are different for Intel CPU configurations (twice as fast), I suspect that threshhold to be around 750-900Mhz.

My calculations can be found here, at another tab of the public spreadsheet: https://docs.google.com/spreadsheet/ccc ... web#gid=14

To be able to convert your FA benchmark score from windowed to full screen (if benched at windowed), I need your memory read uncashed score, besides the info stated in the OP. You can bench this with this 30 day trial of the PassMark Performance Test here: http://www.passmark.com/products/pt.htm

EDIT: it's impossible to convert windowed to full screen for a processor I don't own.

Statistics: Posted by Dr_Doolittle — 29 Aug 2014, 13:08

Re: Lets make simspeed benchmarks (revised!)

2014-08-29T10:36:32+02:00

AMD 965 @ 4.04GHz
XFX 7870
win 7 64 bit
8gb ram

37
1.27
2.38
4.15

Statistics: Posted by Shamble — 29 Aug 2014, 10:36