| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Community
Forum: Hardware Chat Room
Thread: Single channel memory. Does it restrict crunching? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 44
|
|
| Author |
|
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7846 Status: Offline Project Badges: 
|
Hmmm:2.3ghz vs. 3.1 ghz. That might have something to do with the discrepancy. You did not specify if the performance per thread was the difference or if total throughput was the difference.
----------------------------------------Cheers
Sgt. Joe
*Minnesota Crunchers* |
||
|
|
thunder7
Senior Cruncher Netherlands Joined: Mar 6, 2013 Post Count: 238 Status: Offline Project Badges: 
|
Total throughput. From 'Device statistics history' :
----------------------------------------V2: Statistics Date Total Run Time Points Results 09/12/2020 0:058:08:58:13 227,085 440 09/11/2020 0:035:01:00:45 123,314 256 09/10/2020 0:046:23:46:26 174,401 363 V4: Statistics Date Total Run Time Points Results 09/12/2020 0:097:12:19:08 188,591 401 09/11/2020 0:057:19:21:23 86,324 179 09/10/2020 0:075:15:56:33 85,049 194 both systems run all projects, by the way. [Edit 1 times, last edit by thunder7 at Sep 13, 2020 6:30:23 AM] |
||
|
|
KerSamson
Master Cruncher Switzerland Joined: Jan 29, 2007 Post Count: 1684 Status: Offline Project Badges:
|
Based on CPU and mainboard design, from a performance point of view, it is not recommended to operate a system with a single channel memory, if you look for achieving the best performance as possible.
----------------------------------------In particular with memory stressing sciences such as MIP1 (and ARP1), the wider is the memory bandwidth, the better the resulting performance. Honestly I do not see the overall objective of investing money in a high performing CPU and to castrate its memory performance with a single channel configuration. Cheers, Yves --- PS: Don't forget that the memory bandwidth becomes significant as soon as "cache miss" occurs; i.e. data is missing in Level 3 cache. Likewise I would prefer a Ryzen 9 3900X(T) with only 24 threads but 64 MB cache than a 3950X with 32 threads with the same cache size (64 MB). |
||
|
|
thunder7
Senior Cruncher Netherlands Joined: Mar 6, 2013 Post Count: 238 Status: Offline Project Badges:
|
I'll update when I receive my 8 sticks of RAM.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
A lenovo I7 octa core running 6 of the 8 cores on WCG, Ubuntu and the system monitor ( nothing else running ) only utilizes 2.7GB of the available 16gb. So having a lot of ram definitely isn't that critical.
|
||
|
|
DukeBox
Cruncher Joined: Nov 18, 2004 Post Count: 30 Status: Offline Project Badges: 
|
@gta198 It's not about the amount of used RAM but if the CPU can address it in single or dual channel mode. The memory usages will also vary per project, for example ARP uses ~1GB of RAM per WU so for every core running that project you need at least 1 GB of free memory.
----------------------------------------Regarding single/dual setup there is no 'best' answer. Overall it will depend mostly how often the CPU has to skip a cycle for waiting on memory. If calculations can be done within CPU L1/L2/L3 cache that will offload memory IO. If not then it will help to have the lowest latency access to RAM however if it needs a lot of data, then the fastest (bandwidth) way of accessing would be better then latency in general. So in short.. Yes, a CPU with small L1/L2/L3 cache will in most cases benefit from low latency memory preferable in dual (triple/quad/+) channel. Is it noticeable.. ? maybe. An additional RAM DIMM will also use a bit more power (Watt). [Edit 3 times, last edit by DukeBox at Sep 21, 2020 7:18:10 AM] |
||
|
|
thunder7
Senior Cruncher Netherlands Joined: Mar 6, 2013 Post Count: 238 Status: Offline Project Badges:
|
Well, first results seem to indicate a clear change.
----------------------------------------Below is 'Device Contribution History' since 1st of september: 09/20/2020 0:080:07:21:23 243,349 476 09/19/2020 0:081:11:35:04 249,877 525 09/18/2020 0:086:06:04:56 216,735 493 09/17/2020 0:080:00:13:19 135,804 317 09/16/2020 0:071:21:00:16 106,405 215 09/15/2020 0:078:18:17:15 202,499 374 09/14/2020 0:073:23:27:10 95,037 181 09/13/2020 0:087:15:27:19 164,923 338 09/12/2020 0:097:12:19:08 188,591 401 09/11/2020 0:057:19:21:23 86,324 179 09/10/2020 0:075:15:56:33 85,049 194 09/09/2020 0:070:20:34:27 112,617 226 09/08/2020 0:082:14:04:04 110,160 225 09/07/2020 0:078:16:40:42 101,077 197 09/06/2020 0:087:16:00:29 127,059 256 09/05/2020 0:074:15:37:27 108,606 232 09/04/2020 0:073:14:42:28 114,728 239 09/03/2020 0:092:07:03:39 118,876 245 09/02/2020 0:058:18:33:50 89,875 160 09/01/2020 0:089:06:15:58 159,490 323 I exchanged 2x 16 for 8x 4 dimms on the 17th early in the day. I'll report back after a week or two. [Edit 2 times, last edit by thunder7 at Sep 21, 2020 4:33:40 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
So the question is no longer, to multichannel or not to multichannel ;o
|
||
|
|
thunder7
Senior Cruncher Netherlands Joined: Mar 6, 2013 Post Count: 238 Status: Offline Project Badges:
|
No, multichannel is the way to go:
09/28/2020 0:078:21:53:10 296,760 510 09/27/2020 0:079:17:17:56 290,292 507 09/26/2020 0:079:08:31:57 282,647 512 09/25/2020 0:077:11:57:56 275,673 491 09/24/2020 0:080:03:37:31 283,859 502 09/23/2020 0:078:02:30:40 242,028 502 09/22/2020 0:080:21:48:40 257,256 511 09/21/2020 0:080:04:42:39 245,634 516 09/20/2020 0:080:07:21:23 243,349 476 09/19/2020 0:081:11:35:04 249,877 525 09/18/2020 0:086:06:04:56 216,735 493 09/17/2020 0:080:00:13:19 135,804 317 |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Interesting data thunder7!
To better interpret the data a made a chart (my specialty  ) So starting on 17th when you switched to dual channel there was a clear rise in points. Now it would be very interesting, which mix of projects you run? As the conclusion could be valid for some projects and not for others. Could you give us a qualitative idea about that? (i.e. something like bulk of the WUs are MCM, only 3 MIP at a time, very few ARP or something like that.) Noteworthy may also be, that Childhood Cancer restarted on 23rd, which gives very high points on Linux. That may explain the further slight rise in points around that time. |
||
|
|
|