Dual v. Single Channel Benchmarks....

[Naja002]

Ok, in a post a while back, I offered to try to compare running single v. dual channel ram. In another post I asked How to Go About it...... The response was to run multiple benchmarks with dual and single channel. So, here are the numbers for a Q9450 @ 3.6 on a Asus P5E-VM HDMI and an E8400 @ 4.0 on a Gigabyte GA-G31M-S2L:









Feel free to check my math....


Not sure what the value of this is overall. Anyone care to comment?

[cadaveca]

Doing a render it has a greater impact. Very interesting though...means negligible performance loss for crunchers with just one DIMM installed...which might means more FSB too...

[STEvil]

I dont think measuring drystone/whetstone in a short period is a valid benchmark of how your system will perform for WCG.

I dont think this will be fixed any time soon so the best you can do to test performance would be to run for a week or two in dual then switch to single and repeat.. and keep the rig on a certain work unit type only possibly.

[Jaco]

I agree with STEvil.

To be sure you'll have to run single channel for a week and then compare with your dual channel output. The boinc benchmark doesn't show any difference between single/dual channel , but the dual channel config will be slightly faster , hence have a higher ppd.

[[XC] riptide]

For what its worth users have traditionally reported little difference (if any) in WCG output vs RAM speed/timings. Although it would be interesting to see what effect would something as drastic as single channel have.

[Archetype450]

From what I read, best crunchers are :

- Highest CPU frequency
- Highest CPU cache

My Prescott 3.0 GHz / 1 Mb L2 Cache / 512 Mb RAM has an output 30% better than my Northwood 2.87 GHz / 512k L2 Cache / 1.3 Gb RAM.

I know Prescott is not Northwood, but 30% is quite a difference!

[Naja002]

Ok, well it seems as though I went on a wild goose chase then.......

Anyway, it also seems as though I may have located the problem on my gigabyte MB--1 ballistix stick. So, I am going to have to rma both. That will leave me with 2, so I'll be running single channel for a while (1month?).

So, what numbers do I need to collect? The wcg results are affected by many different things, so how do I get this to provide some useable numbers?

I've had plenty of downtime, so my wcg numbers are only probably half or 2/3s what they should be right now. I can start single channel and then dual channel once the rma gets back--what numbers am I looking for?

[STEvil]

Put both sticks in one channel and run for a pre-determined amount of time using only a single work unit type (cancer or whatever)

Repeat using dual channel with both sticks.

Memory amount must remain the same in both tests.


Some work unit types may benefit more than others depending on importance of memory access....

[Naja002]

Memory amount must remain the same in both tests.

Confused on that part. Memory amount meaning running 2x 1GB dual channel v. 1x 2GB single channel.....?

[loonym]

Confused on that part. Memory amount meaning running 2x 1GB dual channel v. 1x 2GB single channel.....?

Yeah that's it, except I think he really meant 2x1gb single vs 2x1gb dual. That would be an interesting test since if the hit isn't big I might split up a couple of my 2x2gb kits.

[Naja002]

Ok, got it. For some reason the 2x 1GB single channel never crossed my mind.

So, what numbers am I looking for? My "averages" are wacked from the various downtime......

[loonym]

Well, I think kind of like steve said, it will work best if you do a test like this with all 1 type wu, and even better if you did all cancer and then another with all hpf and so on. The down time really wouldn't be a factor I don't think. All you really need are completion times.

edit: I don't really have the patience to do this kind of ambitious testing but I would certainly love to see the results. I'm sure a lot of other people would find it useful too.

[[XC] Hicks121]

Put both sticks in one channel and run for a pre-determined amount of time using only a single work unit type (cancer or whatever)

Repeat using dual channel with both sticks.

Memory amount must remain the same in both tests.


Some work unit types may benefit more than others depending on importance of memory access....

Yes, but doing that, I (you) get different length wu's for let's say HCC. Unless you would get all the wu's with the same time limit to complete the test wouldnt work IMO. Am I thinking right about this. I know I get several different HCC wu's. Some take longer than the others.

[Naja002]

Ok, well, completion times is one thing that I was thinking of because I already watch those with the OCing, etc.

Yesterday I swapped out my cpus--put the Q9450 on the Asus and the E8400 on the gigabyte. I've had issues with the gigabyte still, but as I mentioned earlier--I think its a bad ram stick. It runs well single-channel or with a different stick dual-channel. So, I am running it now dual with a different stick to see if the other stick is bad or if maybe its something with the MB. My money is on the stick. It doesn't run anything excpet WCG-HCC.

Last night I put dimes on the Q9450/Asus and today I put my 8800GT back on and added F@H GPU2. So, I have the Q9450 @ 3.6 (all 4 cores) running WCG-HCC and the F@H GPU2 is just swiping whatever cpu power it needs--which isn't much at all--and I'm running 32-bit XP. My point is that my WCG-HCC completion times are pretty much the same. I can't quite tell yet, but they are no more than a few minutes off at most. This is all on single-channel 1x 1GB (This is the board that I swiped the stick from to test the Gigabyte MB).

Here a 1 min video of my Task Manager-30sec/30sec:


Click here to see Video


As far as the testing goes, I can use the E8400/Gigabyte that runs nothing except WCG-HCC. I plan on running it dual-channel until probably Monday--if there are no issues, I'll rma the bad stick and run it single-channel for a while. As long as the Asus MB holds out with only 1 stick--it shouldn't be a problem.

Any other numbers? Anywhere that I can get daily WCG totals--not averages, but daily totals....?

[loonym]

I don't think the different length work units will be a huge factor if you avg a large enough sample. You can get daily totals at WCG my grid page. Shows number of work units returned, points earned, and cpu time.

[STEvil]

Yeah, the sample (a week) should help things average out.

Since points are determined by completion time as a function of dry/whet benchmarks you could in theory run 500 short WU's or 1 huge WU and receive nearly the same points for the given time worked.

This of course depends on the point value assigned to each WU as well.


If WCG would just include a benchmark WU with BOINC (a single WU that could be run as many times as you wished like a 3D bench) this would be a lot easier .... SETI@Home did that before they went BOINC and it was very easy to find the best performing combinations using it.

[Naja002]

Ok, NP. I'll start monitoring the E8400/Gigabyte WUs completion times. My biggest concern is going to be WUs uploaded before I get a chance to see them........I guess I'll reset the network useage time......

[D_A]

If you go to the My Grid page at WCG you can check on all your details, including completion times of past results.

[Naja002]

Yes, but doing that, I (you) get different length wu's for let's say HCC. Unless you would get all the wu's with the same time limit to complete the test wouldnt work IMO. Am I thinking right about this. I know I get several different HCC wu's. Some take longer than the others.

Hey Old Man, didn't see your ninja post earlier. My WCG-HCC WUs don't vary by much-- a few mins. And I agree that if enough WUs are averaged--it should give a decent idea.

Please folks: Don't consider this "Scientific" , but I will do the best that I can!

[D_A]

Please folks: Don't consider this "Scientific"

You want scientific? Got look at the thermal paste thread up in the stickies! Ya damn near need a slide rule to read the thread!

[Naja002]

If you go to the My Grid page at WCG you can check on all your details, including completion times of past results.

I've been going there to check my stats, but what I'm wanting is something that will tell me how many points/WUs/etc I did today, yesterday, etc. Without having to subtract my points from yesterday from my points from today, etc.

In other words, I can take my total points yesterday and subtract them from my total points from today to get my points for the last 24hrs. The graphs give me a rough idea, but is there anywhere that offers them in straight-forward numbers.....?

I guess it really doesn't matter though. The way I understand it is that the points are based (at least in part) on validated WUs and WUs completed/day don't always equal WUs validated/day. So, it would be kind of a mixed up way to try to do things. 123Bob offered a 58 day (I think) comparison a while back. I think in order to use those numbers it would need to be done over a long haul like that.

[D_A]

On that page, scroll down to the area where it says "Member Statistics History" and click on the link just under and to the left of it titled "14records". The "30 records", "60 records" and "365 records" all do the same thing but for different time frames. ie you can check you daily stats for points, results and run time in table for out to the last 365 days.

[[XC] Hicks121]

Hey Old Man, didn't see your ninja post earlier. My WCG-HCC WUs don't vary by much-- a few mins. And I agree that if enough WUs are averaged--it should give a decent idea.

Please folks: Don't consider this "Scientific" , but I will do the best that I can!

ok, sorry, i was just thinking again, you know how dangerous that is for me.

[Naja002]

On that page, scroll down to the area where it says "Member Statistics History" and click on the link just under and to the left of it titled "14records". The "30 records", "60 records" and "365 records" all do the same thing but for different time frames. ie you can check you daily stats for points, results and run time in table for out to the last 365 days.

Very Cool. Thanx! That's what I was looking for......I was kinda surprised that they didn't offer anything like that, but,.....uh......well,....uh.....I'ma go sit over here with Hicks for a while.....

[D_A]

No problem We all have those days sometimes.