CPU experiment with ISBoxer - Help me collect stats!

[Rakurai]

I realize that there are plenty of discussions about AMD vs Intel, which one sucks, whatever. This thread is for a very specific experiment regarding the D3/DB instance/dollar ratio for as many CPUs as we can get stats on.

The Experiment:
How many D3/DB instances can you actually fit on a CPU, if you give it plenty of memory and video card? The goal is to get a bang for the buck ratio, both in just the CPU and the entire system. What I'm interested in here is instances per dollar! Not whether it will be a "good gaming rig" with bots in the background, or any other criteria. If enough people post their stats, I can try to make some kind of sense of this information and post a price/instance ratio for each CPU. Maybe this thread can help others make a decision when building a pure botting rig.

EDIT:
OK, so the below setup procedure and stuff is probably asking a bit much of your time for this. Here's a shortened version that everyone can do in 30 seconds. Just reply to this post with:

• Your CPU type
• Clock speed if it is different than stock
• How many bots you're running
• Are they in VMWare?
• A reasonable guess at your average CPU load

You can still do all the stuff below if you really want to, but its not like I'm paying you or something I'll go first! See my example reply below.

RESULTS:
Here is a temporary results table, I'll make something prettier if we get more results.

[U]Processor[/U]         [U]Freq[/U]      [U]Bots[/U]    [U]VMWare[/U]   [U]Load%[/U]    [U]Load%/Bot[/U]     [U]MaxBots[/U]
Celeron G530      2.4ghz     2        No      45%       22.5%          4
Core i5-3570K     3.4ghz     2        No      20%       10%           10

(Old) Requirements:

• ISBoxer - I guess you could use some other kind of boxing software, I just don't know what kind of overhead they have. If you use method 1 in collecting the data (below), it shouldn't matter.
• AMD CPU - AMD 4-, 6-, or 8-core desktop CPU. I don't think we'd have enough samples to really compare Opterons.
• Intel CPU - i5 or i7 4-core
• Motherboard - anything that fully utilizes your CPU
• Other - SSD, memory, temperature, astrological sign, etc - shouldn't matter, but see "Getting a good reading" below!

Normalizing the Data:
Running 10 bots at 30 FPS each is a far different CPU load than 10 bots at 10 FPS. Running them at 800x600 is different than 200x150. So, while any results that people submit would be helpful, what would really be great is if you could spend 10 minutes of your time and set your machine up with these settings:

• All bots at 400x300 resolution (enough to see what they're doing)
• All framerates capped at 12 FPS
• Ticks per second at the default of 10 (if you don't know what this is, then its at default for you)

Getting a Good Reading:
With the above settings, we need to see that the CPU is being fully utilized. If your hard drive, memory, or video card are throttling the whole thing, then your CPU usage will be lower. Take a look at the FPS of your instances. If they are lower than the 12 FPS cap, there is a bottleneck somewhere. Here is a set of things to check (AFTER you conform to the specs above!):

• Memory - make sure it is not maxed out (Task Manager/Performance). If you have no more memory then it is swapping to the hard drive, slowly.
• Video card - Get GPU-Z and see if it is maxed out, or if your VRAM is maxed out.
• Hard Drive - Harder to diagnose, but could be an issue if all your instances are sharing a slow hard drive.

The idea is that you run as many instances as you can while not hitting a bottleneck. If your FPS is low, decrease the number of instances until you're mostly at 12 on all of them.

Collecting the Data:
Get all your instances running and actually doing something (at least walking around). Open Task Manager and on the Performance tab. At this point, you can go two ways:

• The more helpful way: click Resource Monitor. You can sort them by Average by clicking it, that should put D3 and DB at the top of the list. Wait a few minutes for the averages to normalize. Either take a screenshot (stretch it out so we can see them all) or write down the approximate averages for each instance. You can average it yourself if you want, or post it here and I'll do it for you. The important thing is figuring out %load per instance.
• The less helpful, but still useful way: Just get an estimate of the CPU load and report that along with the number of instances. Please make sure that all of your cores are being utilized! If not, tell us that, or even better, fix it so they are.
• While you're at it, get some readings on your video card for the other experiment! See the thread here!

Submission:
Please make sure to read "Normalizing the Data" and "Getting a Good Reading" above! If you don't... well, we'll try to extrapolate some kind of information from your submission, but at least tell us where you deviated from the instructions. Please include this information:

• CPU brand and model
• CPU frequency (ghz)
• ISBoxer - cores per instance
• CPU Usage - per instance of D3 and DB, or overall. This can be a screenshot if you want.
• Number of instances - how many pairs of D3/DB.
• Any deviations from the above guidelines so we don't have bogus data!

Thanks for the help!

 

[Rakurai]

Reserved to post stats, eventually.

 

[eugenius]

how do you force isboxer to make the windos 400x300?

PLEASE TELL ME

 

[jimjohnson]

how do you force isboxer to make the windos 400x300?

PLEASE TELL ME

ISBoxer Suite go to Window Layout --> CharacterSetName --> Regions and set the size

For the OP:

Good idea but how are you going to combat people who are getting poor scores based on spyware/bloatware/JustNotKnowingHowToActuallyConfigureWindows? Seems like that would drastically increase bogus data or at least not accurate data.

 

[Rakurai]

ISBoxer Suite go to Window Layout --> CharacterSetName --> Regions and set the size

For the OP:

Good idea but how are you going to combat people who are getting poor scores based on spyware/bloatware/JustNotKnowingHowToActuallyConfigureWindows? Seems like that would drastically increase bogus data or at least not accurate data.

Yeah, not much I can do about that. I'm probably going to limit the results to people who have more than 1 bot. If you're running two bots, you're probably know what you're doing and can tell the difference between your bots eating cpu and your computer being part of a botnet or something. The bad kind of botnet.

That said, I've modified the rules for entries a LOT. Setting up your comp to get results for this post is probably asking too much, so now I'm going for just a good guess of CPU usage per bot, and not perfectly precise results. The point of this exercise is to give semi-concrete evidence that can help people make hardware decisions, and that's not going to happen if the conditions are so tough that no one posts results. The new 30-second submission requirements will (hopefully) get a few more responses, even if they're less precise.

 

[Rakurai]

PLEASE read the new EDIT in the original post!
Submissions now take 30 seconds of your time!

My (current) specs are: Celeron G530 at 2.4ghz, 2 bots, ISBoxer (no vmware), about 45% average load total.

In a few days I'll post results for an i5 3570K! Anxiously waiting.

 

[Rakurai]

New system! Core i5-3570K @ 3.4, Asus Z77 board, 16gb, Samsung 830 SSD. 2 bots, ISBoxer, less than 20% average load.

I'll be buying more keys before too long, and I'll update this. 2 bots is a pretty light load for this CPU. I'd also like to see how clock speed affects it, so I'll overclock at that point. Might just be something boring like linear improvement, though.