Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
World Community Grid Forums
Category: Completed Research Forum: Microbiome Immunity Project Thread: sigificant credit drop - only for me or did someone else see this? |
No member browsing this thread |
Thread Status: Active Total posts in this thread: 117
|
Author |
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I would assume it is handle the same as the CPU since there are only 4 CPUs with Hyperthreading
|
||
|
KerSamson
Master Cruncher Switzerland Joined: Jan 29, 2007 Post Count: 1664 Status: Offline Project Badges: |
I'm happy being able to notice that I am not alone with my observations and questions related to MIP1 (poor) performances.
----------------------------------------As I already mentioned, the problem does not seem to be related to hyperthreading (even if HT could amplify the effects). Even on a 4 or 6 core CPU (without HT) - e.g. Athlon II x4 or Phenom II x6 - MIP1 is impacting significantly the overall performance. On my side, I initially focused on possible Memory Management problem (RAM) , since I did not observe problem with Disk I/O and that the swap space remains empty (enough available RAM). Even if a concurrent demanding use of a shared FPU could be relevant in some cases, this explanation does not cover all relevant observations. The root cause is somewhere else. Cheers, Yves |
||
|
rod4x4
Cruncher Joined: Apr 29, 2014 Post Count: 12 Status: Offline Project Badges: |
Hi Yves
I share your assumption that it may be related to memory and or memory related operations. As my PC has 16Gb RAM, the swap runs empty - always. Even though my cpu was running cool, I did notice the motherboard temperatures running a touch higher than normal. Not sure where the MB sensor is exactly, but it could indicate something unusual occurring in RAM. My PC is Linux (kernel 4.10.0-37) AMD Ryzen 1700 and 16Gb RAM. magiceye04 also has a ryzen 1700. Perhap could be related to the new ryzen platform? |
||
|
KLiK
Master Cruncher Croatia Joined: Nov 13, 2006 Post Count: 3103 Status: Offline Project Badges: |
Hi Yves I share your assumption that it may be related to memory and or memory related operations. As my PC has 16Gb RAM, the swap runs empty - always. Even though my cpu was running cool, I did notice the motherboard temperatures running a touch higher than normal. Not sure where the MB sensor is exactly, but it could indicate something unusual occurring in RAM. My PC is Linux (kernel 4.10.0-37) AMD Ryzen 1700 and 16Gb RAM. magiceye04 also has a ryzen 1700. Perhap could be related to the new ryzen platform? Check flow through of the air the casing & check MBO data for Tjunc of Northbridge / Southbridge. |
||
|
rod4x4
Cruncher Joined: Apr 29, 2014 Post Count: 12 Status: Offline Project Badges: |
Hi KliK
Thanks for the suggestions. I have a case with 4 chassis fans plus a Noctua NH-u14s CPU cooler. The PC case has been modified with custom baffles to maximise the air flow over the MB. I have also install Linux lm-sensors package as maintained by groeck. Unfortunately Linux support for AMD Ryzen temperature monitoring is still being developed. (Not integrated into the kernel until v4.15, I am still on v4.10) I have to recompile the temperature package with every kernel increment. Hence finding the Tjunc temp will be an adventure for me. If you can add any light on identifying the Tjunc for Ryzen in Linux, I will certainly give it a go. thanks again. |
||
|
B2I
Senior Cruncher usa Joined: Jan 23, 2011 Post Count: 232 Status: Offline Project Badges: |
I also have noticed a significant drop in points earned in the last 3-4 days. While I have been a volunteer cruncher for years I am now a member of the gridcoin team, trying to mine enough gridcoins to pay for my electricity. The way they figure your earnings has to do with how many points you put one the board in relationship to others crunching the same project. They consider WCG a single project and do not see the various research projects independently.
----------------------------------------So, if something is going wrong in MBI, It is literally costing me money. If this doesn't pick up in the next few days, I must switch projects. I am not alone in this and as more crunchers sign up with gridcoin, the science projects that don't deliver WU and get the validated quickly will see fewer participants. Right now, you have to join the gridcoin to mine gridcoin as you work. The administrators are working on a program that would allow people to stay with their own and also mine at the same time. When that happens, I would expect a high percentage of WCG participants to sign on. I did notice that i had 187 pages of "in process" tasks. Could this indicate a backlog problem? B2I |
||
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7223 Status: Offline Project Badges: |
I did notice that i had 187 pages of "in process" tasks. Could this indicate a backlog problem? That may or may not indicate a backlog problem on your machine(s). How many machines do you have and how many cores ? Which projects are you crunching ? Are you running Windows, Linux or Android ? To how many days are your cache settings set ? On my dual hexcore machines running SCC1 I have cache set to 1/2 day. This gives me about 20 pages of "in progress." If I set the cache to 5 days I would probably have over 100 pages of "in progress" per machine. With two of these machines I could easily have over 200 pages "in progress." Unless you are missing deadlines or having your machines run in panic mode, you are probably OK. Without knowing more information that is just a guess. Cheers
Sgt. Joe
*Minnesota Crunchers* |
||
|
B2I
Senior Cruncher usa Joined: Jan 23, 2011 Post Count: 232 Status: Offline Project Badges: |
got 7 'puters. mix of windows 7, 10 and linux. one is set to max cache because it is in a remote site that on gets an internet connection once a week when I can get to it to connect via smart phone, the rest are set to 3-4 days because of spotty internet coverage. all but one are i7s, one i5. missed some deadlines last week because I couldn't get to the remote 'puter but that is cleared now.
----------------------------------------Whew!. Think I answered all your questions but it is Saturday night and I just finished my 4th drink :). Sgt Joe. I've been seeing your logo for years now. Are you or were you a Sgt I'm a retired USAF E8. B2I From the wilds of Colorado |
||
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I'm a retired USAF E8. - and sometimes you were a Base Camper B2I B2I From the wilds of Colorado Sent you an invite to travel to the wilds of Boise Idaho with some of all those computers a while ago in the revolving door of the thread of the Four Letter Game where you pass so quickly not noticin' nothin' like ships at sea at night we have drinks in our camp in Idaho, too [Edit 1 times, last edit by Former Member at Oct 22, 2017 7:18:45 AM] |
||
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
After monitoring both chips on a machine running 32 threads with the large memory WUs, both chips report L1_Cache_Miss=0, L2_Cache_Miss=0, and L3_Cache_Miss=0. Since there aren't any L3 Cache misses, it's unlikely there are any cross-chip "snoops" happening. The QPI links are very busy; in the range of 15.3GB/s across 2 links. 5 to 1 read to write ratio. Unfortunately, I don't have the QPI counters turned on in the BIOS so can't see QPI details but it looks like the QPI links are being saturated and I assume this would happen on the AMD HyperTransport Links also. Since control flow for the links happens on-chip, the OS doesn't see any wait state so reports the thread as dispatchable. Also explains why chips with fewer threads see less of a problem than chips with high thread counts. Without QPI counters, its only a guess. Could try and configure the machine for non-NUMA and see if that helps but I suspect it would only be a minor benefit.
----------------------------------------Sorry, posted in wrong thread..Can't delete since there is a reply. [Edit 2 times, last edit by Doneske at Oct 22, 2017 3:17:45 PM] |
||
|
|