How might memory effect R@h processing?

Message boards : Number crunching : How might memory effect R@h processing?

To post messages, you must log in.

Previous · 1 · 2

AuthorMessage
Keith Myers
Avatar

Send message
Joined: 29 Mar 20
Posts: 95
Credit: 289,903
RAC: 0
Message 93593 - Posted: 6 Apr 2020, 1:42:17 UTC - in response to Message 93582.  
Last modified: 6 Apr 2020, 1:47:36 UTC

This MoBo only has 2 memory slots so I cannot go past 16GB even if I decide to invest more dollars in this project.

Is the motherboard so old that it has a northbridge chipset?
The IMC in modern cpus is what controls the type and density of RAM DIMMS that are compatible.
From the Intel datasheet, the i5-6400 supports up to 64GB of memory.

Memory Specifications
Max Memory Size (dependent on memory type)
64 GB

Memory Types
DDR4-1866/2133, DDR3L-1333/1600 @ 1.35V

Max # of Memory Channels
2

Max Memory Bandwidth
34.1 GB/s

ECC Memory Supported ‡
No


pretty sure when I bought it the max was 16GB.

That statement was probably true at the time since two channels of the highest density RAM DIMMS at the time was probably 8GB. Now RAM density is much higher per stick with 8GB and 16GB dies and dual rank DIMMS. I don't see how the motherboard could have any limitation on how much memory can be installed other than still only two channels. The limitation would be in the cpu.
ID: 93593 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile robertmiles

Send message
Joined: 16 Jun 08
Posts: 1223
Credit: 13,824,497
RAC: 2,340
Message 93595 - Posted: 6 Apr 2020, 2:04:13 UTC - in response to Message 93593.  

[snip]


That statement was probably true at the time since two channels of the highest density RAM DIMMS at the time was probably 8GB. Now RAM density is much higher per stick with 8GB and 16GB dies and dual rank DIMMS. I don't see how the motherboard could have any limitation on how much memory can be installed other than still only two channels. The limitation would be in the cpu.

The limitation for total memory size used to be the number of bits the CPU could send to control the memory address. I haven't checked if this is still true.
ID: 93595 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Grant (SSSF)

Send message
Joined: 28 Mar 20
Posts: 1467
Credit: 14,320,025
RAC: 16,345
Message 93596 - Posted: 6 Apr 2020, 2:42:55 UTC - in response to Message 93595.  

The limitation for total memory size used to be the number of bits the CPU could send to control the memory address. I haven't checked if this is still true.
As Keith posted, these days the memory controller is built in to the CPU, not on the motherboard in the Northbridge.
While there might be some really cheap motherboards that restrict the maximum RAM a system can have (other than just a limited number of slots), it usually just comes down to what the CPU supports these days.
Grant
Darwin NT
ID: 93596 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Stephen "Heretic"

Send message
Joined: 2 Apr 20
Posts: 21
Credit: 11,028
RAC: 0
Message 93608 - Posted: 6 Apr 2020, 7:26:19 UTC - in response to Message 93593.  
Last modified: 6 Apr 2020, 8:01:01 UTC

Is the motherboard so old that it has a northbridge chipset?
That statement was probably true at the time since two channels of the highest density RAM DIMMS at the time was probably 8GB. Now RAM density is much higher per stick with 8GB and 16GB dies and dual rank DIMMS. I don't see how the motherboard could have any limitation on how much memory can be installed other than still only two channels. The limitation would be in the cpu.



. . It is a Medion computer, it has an MSI MoBo with no designation so I cannot get specs for it, it came with a booklet which basically said "congratulations on buying this computer" and little else (not sure where even that is now) but if I remember correctly the largest memory it supports is 8GB. I remember thinking 16GB should be enough if I have to expand it. But since it is shaping up that memory size is not the actual core problem then it is a moot point. Going on figures that others have posted I should not need more than 2GB each both for the GPU task and each of 2 CPU tasks so 8GB should be killing it, but that is not the case. After digging around I found Process Explorer just before getting Grant's reply and it showed the system Ram as 3.2GB physically in use and 5.4GB committed (not sure about that distinction but I presume it means that 5.4GB is reserved by current apps). E@H shows as 899MB private and 520MB as whatever. The Rosetta task was much lower at 289 Mb and 234 MB. But to me that says that the problem is NOT that 8GB is insufficient.

Stephen

:(
ID: 93608 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Stephen "Heretic"

Send message
Joined: 2 Apr 20
Posts: 21
Credit: 11,028
RAC: 0
Message 93610 - Posted: 6 Apr 2020, 7:30:52 UTC - in response to Message 93585.  
Last modified: 6 Apr 2020, 8:03:15 UTC

Task Manager, Processes tab, Memory column.
It can vary by several hundred MB depending on where in the Task it is at.
Although Process Explorer is probably better as it shows how much is also reserved, but not using at the time- it's values are higher than those shown in Task Manager.
A Task using 730MB in Task Manager shows as 760MB Private/ 783MB Working Set in Process Explorer.

It doesn't support 16GB DIMMs?
. . It is an old cheapy MSI MoBo. I have very little info but I am pretty sure when I bought it the max was 16GB.
Check the motherboard manual, it should list what capacity & speed DIMMs are supported.


, , With process manager is it the greater of the 2 values or the sum of the 2 values that is the memory rquirement?

Stephen

??
ID: 93610 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Stephen "Heretic"

Send message
Joined: 2 Apr 20
Posts: 21
Credit: 11,028
RAC: 0
Message 93612 - Posted: 6 Apr 2020, 7:39:10 UTC - in response to Message 93587.  

Some of the ram merchants have a ram tester that will recommend the highest level of ram that a MB will support.
If you don't have a lot of cpu cores/threads running Rosetti@Home you should be able to run multiples of the task even if it is running up near 2GB per task on a 16GB ram MB. Lately mine have all been a lot smaller.
Tom


. . I tried CPUZ but it only shows what is installed not what is supported.

Stephen

:(
ID: 93612 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Stephen "Heretic"

Send message
Joined: 2 Apr 20
Posts: 21
Credit: 11,028
RAC: 0
Message 93613 - Posted: 6 Apr 2020, 7:43:18 UTC - in response to Message 93591.  

So in the end, with 32Gb Ram were you able to support all 22 tasks it was trying to run? Or did have to restrict it to any particular number?

I knocked it down to 16 x86_64 cpu tasks so I could run the three gpu tasks. If I tried to run more, I knocked gpu tasks offline with out of memory-postponed messages. That was with use 90% of memory while idle, 85% while active preferences. Was consuming a little shy of 30GB. The host is my daily driver.


. . So on a pro rata basis 8GB should suffice to run just 2 x x86_64 tasks and one GPU task with 90% available even when active.

Stephen

. .
ID: 93613 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Grant (SSSF)

Send message
Joined: 28 Mar 20
Posts: 1467
Credit: 14,320,025
RAC: 16,345
Message 93614 - Posted: 6 Apr 2020, 7:52:47 UTC - in response to Message 93610.  

, , With process manager is the greater of the 2 values or the sum of the 2 values that is the memory rquirement?
Working set is memory that application is using, plus the memory that another shared object it is using requires.
So the value to go by would be the Private Bytes value, which from a quick search will still be higher than that actual amount of RAM in use by that application (shared process memory etc again), but it does give you a good buffer for how many tasks you can run.
So If you allow 1.5GB per task (the largest i've seen to date in Task Manger was 1.3GB), and bump up the in use/not in use memory limits in your Computing Preferences to 90% (or higher), that should let you run 3 Rosetta Tasks, along with other projects at the same time.

I'd still replace the single memory module with 2, even if you only end up with the same amount of RAM (although more would be better).
Grant
Darwin NT
ID: 93614 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Stephen "Heretic"

Send message
Joined: 2 Apr 20
Posts: 21
Credit: 11,028
RAC: 0
Message 93616 - Posted: 6 Apr 2020, 8:33:07 UTC - in response to Message 93592.  
Last modified: 6 Apr 2020, 8:54:46 UTC

The Rosetta tasks seem to run best with 2 GB per task, plus more for the CPU portion of any GPU task, plus more for the operating system. For 8 GB, two Rosetta tasks running at once is about the limit. If you have enough virtual cores, 16 GB should raise the limit to about 6 tasks.
This assumes that you don't let any BOINC tasks stay in memory while they are not running. Expect lower numbers if you do.
Filling both slots with the same model of memory board gives a little more speed than mismatched memory boards. You might also check if you can get faster memory boards that are compatible with your computer.
Some models of computers are compatible with memory boards that contain more memory.
Back when I first started with BOINC, I used this site to find compatible types of memory:

https://www.crucial.com/store/systemscanner

I now order any new computers with the largest and fastest memories that are compatible with that model.


. . This is an off the rack bargain package. At the moment I have upped the memory reserves to 85% when in use and 95% when idle, and I have disabled 'leave non-GPU tasks in memory when suspended" though I am not sure it would be causing problems. I will trial these settings and see if it will play nice.

. . I ran that link and it identifies the MoBo as an Akoya P5110D (H110) chipset. It says that it will support up to 16GB memory at up to DDR4-4000. I find it hard to believe this old MoBo could support memory that fast, but it is recommending a confusing array of memory stick options that has my mind reeling. The sad thing is they must all be USA prices, when I look up similar RAM locally it is double the prices quoted. But I would probably settle for a 16GB kit at DDR4-3000, most of that is backwards compatible with DDR4-2666 so I should be covered, I am not so sure it would support DDR4-3200.

. . Thanks for that link. it seems to have answered some questions.

Stephen

{edit} OK it is running AOK with those settings but I had to set BOINC to use 3 cores to get the second 64 bit task to run. This is not a problem to me as long as it does not crash again :) I ran Process Explorer again and system memory commit has increased to 6GB (from 5.5) but memory physically used is still at 3.4/3.5GB. The second Rosetta task is only using half the memory of the first but there is still plenty of head room within the reserved memory for it to increase. Fingers crossed ...
:)
{/edit}
. .
ID: 93616 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Bryn Mawr

Send message
Joined: 26 Dec 18
Posts: 373
Credit: 10,595,209
RAC: 8,355
Message 93619 - Posted: 6 Apr 2020, 9:16:33 UTC - in response to Message 93593.  

[

That statement was probably true at the time since two channels of the highest density RAM DIMMS at the time was probably 8GB. Now RAM density is much higher per stick with 8GB and 16GB dies and dual rank DIMMS. I don't see how the motherboard could have any limitation on how much memory can be installed other than still only two channels. The limitation would be in the cpu.


The number of lines wired in the address bus?

In a cheap mb they could easily “simplify” the mask and restrict it to the max in use at the time.
ID: 93619 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Grant (SSSF)

Send message
Joined: 28 Mar 20
Posts: 1467
Credit: 14,320,025
RAC: 16,345
Message 93621 - Posted: 6 Apr 2020, 9:37:23 UTC - in response to Message 93619.  
Last modified: 6 Apr 2020, 9:44:49 UTC

That statement was probably true at the time since two channels of the highest density RAM DIMMS at the time was probably 8GB. Now RAM density is much higher per stick with 8GB and 16GB dies and dual rank DIMMS. I don't see how the motherboard could have any limitation on how much memory can be installed other than still only two channels. The limitation would be in the cpu.

The number of lines wired in the address bus?
You are thinking back to the days of the Pentium Pro and earlier (or just earlier- it was a long time ago).
Then they had the Northbridge to connect the CPU to the RAM- the amount, type & speed of RAM the system could use depended on the the chipset for the motherboard.
Now the RAM connects directly to the CPU, the memory controller is part of the CPU.
Grant
Darwin NT
ID: 93621 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile robertmiles

Send message
Joined: 16 Jun 08
Posts: 1223
Credit: 13,824,497
RAC: 2,340
Message 93640 - Posted: 6 Apr 2020, 14:24:53 UTC - in response to Message 93616.  

[snip]

. . This is an off the rack bargain package. At the moment I have upped the memory reserves to 85% when in use and 95% when idle, and I have disabled 'leave non-GPU tasks in memory when suspended" though I am not sure it would be causing problems. I will trial these settings and see if it will play nice.

. . I ran that link and it identifies the MoBo as an Akoya P5110D (H110) chipset. It says that it will support up to 16GB memory at up to DDR4-4000. I find it hard to believe this old MoBo could support memory that fast, but it is recommending a confusing array of memory stick options that has my mind reeling. The sad thing is they must all be USA prices, when I look up similar RAM locally it is double the prices quoted. But I would probably settle for a 16GB kit at DDR4-3000, most of that is backwards compatible with DDR4-2666 so I should be covered, I am not so sure it would support DDR4-3200.

. . Thanks for that link. it seems to have answered some questions.

Stephen

{edit} OK it is running AOK with those settings but I had to set BOINC to use 3 cores to get the second 64 bit task to run. This is not a problem to me as long as it does not crash again :) I ran Process Explorer again and system memory commit has increased to 6GB (from 5.5) but memory physically used is still at 3.4/3.5GB. The second Rosetta task is only using half the memory of the first but there is still plenty of head room within the reserved memory for it to increase. Fingers crossed ...
:)
{/edit}
. .

Mine uses an Intel X99 chipset.

I ran that scan on my computer, and also found the results confusing. I asked them for help comparing the speeds to those of my current memory.

The type of memory used on memory boards does not allow a memory access to complete within one CPU clock cycle for today's CPUs. Instead, the memory controller must start the memory access, wait a certain number of CPU clock cycles, then tell the CPU or rest of the CPU that the access is complete. The memory controller cannot start another access on the same memory board until the last one was finished, but can start one on a different memory board. The type of memory inside the CPU completes memory accesses in one CPU clock cycle, but using that type on memory boards would make them far more expensive.

The site shows me USA prices. Today's restrictions on airline flights would probably make shipping to other countries very slow or very expensive.

Executables usually come with sections marked as something the program can change, and sections the program cannot change. Sections that cannot change, and also do not have to be in a certain position within the memory relative to any changeable section, can be shared between copies of the executable running at the same time.

Are you able to test if it runs better without having any program using the GPU for any purpose other than keeping your display working? You could then check if a program that would use the GPU for another purpose is still in memory.
ID: 93640 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Bryn Mawr

Send message
Joined: 26 Dec 18
Posts: 373
Credit: 10,595,209
RAC: 8,355
Message 93644 - Posted: 6 Apr 2020, 15:41:39 UTC - in response to Message 93621.  

That statement was probably true at the time since two channels of the highest density RAM DIMMS at the time was probably 8GB. Now RAM density is much higher per stick with 8GB and 16GB dies and dual rank DIMMS. I don't see how the motherboard could have any limitation on how much memory can be installed other than still only two channels. The limitation would be in the cpu.

The number of lines wired in the address bus?
You are thinking back to the days of the Pentium Pro and earlier (or just earlier- it was a long time ago).
Then they had the Northbridge to connect the CPU to the RAM- the amount, type & speed of RAM the system could use depended on the the chipset for the motherboard.
Now the RAM connects directly to the CPU, the memory controller is part of the CPU.


There must still be physical connections between the ram sockets and the cpu socket. To simplify the (mask used to create the) motherboard it is possible to narrow the address bus to only those lines that the ram of the day actually uses.
ID: 93644 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Keith Myers
Avatar

Send message
Joined: 29 Mar 20
Posts: 95
Credit: 289,903
RAC: 0
Message 93650 - Posted: 6 Apr 2020, 17:09:31 UTC - in response to Message 93644.  

There must still be physical connections between the ram sockets and the cpu socket. To simplify the (mask used to create the) motherboard it is possible to narrow the address bus to only those lines that the ram of the day actually uses.

Yes, certainly doable. But why would you want to. That would make for a very limited production run to use a special mask chopping off address lines.
ID: 93650 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Bryn Mawr

Send message
Joined: 26 Dec 18
Posts: 373
Credit: 10,595,209
RAC: 8,355
Message 93656 - Posted: 6 Apr 2020, 18:40:44 UTC - in response to Message 93650.  

There must still be physical connections between the ram sockets and the cpu socket. To simplify the (mask used to create the) motherboard it is possible to narrow the address bus to only those lines that the ram of the day actually uses.

Yes, certainly doable. But why would you want to. That would make for a very limited production run to use a special mask chopping off address lines.


The mask for each motherboard is a custom design so, by definition, a special mask.
ID: 93656 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Keith Myers
Avatar

Send message
Joined: 29 Mar 20
Posts: 95
Credit: 289,903
RAC: 0
Message 93658 - Posted: 6 Apr 2020, 18:44:33 UTC - in response to Message 93656.  

I could see this type of cost savings for private label low end consumer products or OEM motherboards. Not for main product lines sold to the general public.
ID: 93658 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Bryn Mawr

Send message
Joined: 26 Dec 18
Posts: 373
Credit: 10,595,209
RAC: 8,355
Message 93663 - Posted: 6 Apr 2020, 18:54:06 UTC - in response to Message 93658.  

I could see this type of cost savings for private label low end consumer products or OEM motherboards. Not for main product lines sold to the general public.


I was responding to Grant's comment :-

I don't see how the motherboard could have any limitation on how much memory can be installed other than still only two channels.


and coming up with a way that it could happen - not that I'm at all pedantic, honest :-)
ID: 93663 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Previous · 1 · 2

Message boards : Number crunching : How might memory effect R@h processing?



©2024 University of Washington
https://www.bakerlab.org