World Community Grid - View Thread - Single channel memory. Does it restrict crunching?

World Community Grid Forums

Category: Community

Forum: Hardware Chat Room

Thread: Single channel memory. Does it restrict crunching?

Quick Go »

No member browsing this thread

Thread Status: Active
Total posts in this thread: 44

[ ]

Author

This topic has been viewed 9198 times and has 43 replies

Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline


Re: Single channel memory. Does it restrict crunching?

Ok, I did a test myself now. I upgraded a Ryzen 1700 running at 3 GHz to 2x 4 GB dual channel @2400 MHz. Before it was running with only 1x 4 GB.
I did the switch on October 2nd in the morning.

The project mix was 14x Smash Childhood Cancer and 2x Microbiome Immunity Project during the whole period.

Here the results:


Date      run time          points       results
10/08/2020  0:016:06:27:38 	142,145 	309
10/07/2020  0:015:23:57:12 	140,409 	304
10/06/2020  0:015:11:24:17 	132,125 	288
10/05/2020  0:016:10:04:20 	144,700 	311
10/04/2020  0:016:01:26:12 	137,140 	301
10/03/2020  0:015:17:07:28 	131,415 	294
10/02/2020  0:015:13:46:55 	121,344 	268
10/01/2020  0:015:11:39:29 	122,149 	291
09/30/2020  0:016:06:50:12 	119,206 	279
09/29/2020  0:016:09:08:57 	124,156 	304
09/28/2020  0:016:22:56:17 	119,911 	287
09/27/2020  0:017:15:21:36 	131,166 	317
09/26/2020  0:016:14:43:22 	116,059 	280
09/25/2020  0:016:11:22:56 	100,761 	263

Here the calculated points per hour of run time:

There was a clear rise in points, but not as clear as the one thunder7 experienced.
The period with dual channel RAM created about 20% more points per hour as the period with single channel did.

I wished I had tested this before...

Nonetheless, I will now switch back to single channel, as I borrowed the RAM from another cruncher, which was switched off during the test. I will then continue to minotor the points, to make sure my results were not only a coincidence.

----------------------------------------
[Edit 1 times, last edit by Former Member at Oct 19, 2020 12:27:49 PM]

[Oct 9, 2020 7:25:13 AM]

Former Member
Cruncher
Joined: May 22, 2018
Post Count: 0
Status: Offline


Re: Single channel memory. Does it restrict crunching?

Short update after switching back to single channel.

As expected, points decreased again, but not to levels as low as before.

360 points per hour of run time for the 6 days with dual channel memory
15.5% (308 points) less for the 6 days before with single channel memory
6.2% (337 points) less for the 6 days thereafter with single channel memory

(I excluded the days when I made the switch from the calculation)

As mentioned before, 14 units SCC and 2 units MIP were running during the whole period. Results for other projects mix can be completely different.

So it seems not to be clear how large the benefit of dual channel memory with my SCC/MIP project mix is, but it seems to be likely, that there is one.

Even if the benefit is only 6%, it seems to be worth it. 2x4 GB instead of 1x4GB increases the overall cost of a cruncher by a similar amount.
Electricity consumtion, which is a major cost factor, increases by approx. 1.5 W with 2x4 GB instead of 1x4 GB, which is only 2% of the current consumption of 75 W that my crunchers have.

So in the future I will build new crunchers with dual channel.

[Oct 16, 2020 8:20:19 AM]