Problem with CPU benchmark

Message boards : Questions and problems : Problem with CPU benchmark
Message board moderation

To post messages, you must log in.

AuthorMessage
Steveplanetary

Send message
Joined: 14 Feb 19
Posts: 5
United States
Message 90132 - Posted: 14 Feb 2019, 20:44:35 UTC

I have a quad core processor. Previously, when I ran Tools > CPU benchmark the first part of the test would use two of the CPUs running at 100% each, thus a total of 50% processor capability. The second part of the test would use one of the CPUs running at 100%, thus 25% of total processor capability. Now only one CPU is used (CPU 0), for both the first and second parts of the test, thus 25% of processor capability.

Also, the test reports one CPU in Event Log. But I read a post somewhere that showed four CPUs.

I'm wondering if there's something wrong with my processor, and the change in CPU benchmark activity is just one of the reasons. I run LHC@Home (which runs in a Virtual Machine in Oracle VirtualBox) and a GPU project at Einstein@Home. I used to be able to run both simultaneously with various configurations. Each E@H task uses one CPU to feed the GPU. For example, I could run two E@H tasks, with each using half of the GPU, and one LHC task using two CPUs. For a couple of days I didn't run E@H at all, and I used all four CPUs for one LHC task at a time.

Now I'm limited to just one CPU for a LHC task, whether I configure the project preferences to use two, three, or no limit to the number of CPUs. So I'm wondering if the BOINC CPU benchmark result is providing an explanation for this.

I might add that Task Manager still shows veriable activity in all four CPUs. Perhaps not as much as before, but I'm not as accustomed to running just one task with 25% of the processor capability. And Speccy shows variable CPU temperatures, as before, but it's pretty hard to correlate that with what I see in Task Manager.

I would really appreciate any help with this.
ID: 90132 · Report as offensive
Steveplanetary

Send message
Joined: 14 Feb 19
Posts: 5
United States
Message 90147 - Posted: 15 Feb 2019, 16:21:15 UTC - in response to Message 90132.  

I had no way to edit my post, so I had to reply to it to add some information.

I just ran CPU benchmark and it used all four CPUs, for 100% of processor capability. That never happened before. It will be interesting to see what happens when I download my next LHC task. I'll keep everybody posted.

The only thing I did differently last night was to run Memtest64 for ~3.5 hours. I have been wondering if the processor memory manager wan't working correctly. Maybe Memtes64, which fully used all four CPUs retrained it.

Don't ask me how. It's just a wild guess anyway. I'll keep everybody posted.
ID: 90147 · Report as offensive
Profile Ageless-Away
Volunteer moderator
Project administrator
Avatar

Send message
Joined: 29 Aug 05
Posts: 13300
Netherlands
Message 90148 - Posted: 15 Feb 2019, 16:36:18 UTC - in response to Message 90132.  
Last modified: 15 Feb 2019, 16:37:22 UTC

A quad core CPU is one CPU with four physical cores built into it. It counts as one CPU, hence why BOINC shows it as one CPU.
There are systems out there that use two or more CPUs, all with anywhere from 1 to 32 cores, some of these hyper threaded as well. For these the BOINC benchmarks will only show the physical CPU count, not the core count, or the logical count.

When you start BOINC, it will show the amount of cores your CPU has, both physical and logical. For example, my i5 without hyper threading shows as:
15/02/2019 17:26:58 | | OpenCL CPU: Intel(R) Core(TM) i5-3470 CPU @ 3.20GHz (OpenCL driver vendor: Intel(R) Corporation, driver version 3.0.1.10891, device version OpenCL 1.2 (Build 76427))
15/02/2019 17:26:58 | | Processor: 4 GenuineIntel Intel(R) Core(TM) i5-3470 CPU @ 3.20GHz [Family 6 Model 58 Stepping 9]

1 CPU, with 4 cores: a quad core CPU.

I'll leave the answer to the question as to why you run what you see to others who run the same combination. I don't, so don't have experience with it.

Forums:
Editing posts can only be done up to an hour after posting. Moderators have unlimited post editing times, but can only edit their own posts, not those of others.
ID: 90148 · Report as offensive
robsmith
Help desk expert

Send message
Joined: 25 May 09
Posts: 470
United Kingdom
Message 90149 - Posted: 15 Feb 2019, 17:17:51 UTC

It's quite simple really - Memtest is exactly what its name implies, a memory test tool.
Other benchmarking tools test various other aspects of a system. Some will be testing the performance of a single CPU core, some will test multiple cores, in the latter case most will do two cores and scale the result to however many cores are in the CPU chip. Some tests will take note of the CPU architecture and will test both "real" and "virtual" cores under hyper-threading.
Then of course there is what sort of benchmark being run - a simple set of maths (integer, floating point, fixed point, single or multi precision etc.), a set of logic, a real-world mix of both, and so it goes on. Last time I really looked there were over 500 different combinations, then probably ten times that number of actual implementations.

BOINC uses a fairly simple, single core, single-float benchmarking tool to give a performance guide. It can only be a guide as every project uses a very different balance of operand types. It then scales for the number of CPU cores found.
Of course projects may have their own benchmarking tools, and they will almost certainly be tailored around that project's applications.
ID: 90149 · Report as offensive
Steveplanetary

Send message
Joined: 14 Feb 19
Posts: 5
United States
Message 90150 - Posted: 15 Feb 2019, 18:02:18 UTC - in response to Message 90148.  
Last modified: 15 Feb 2019, 18:35:56 UTC

A quad core CPU is one CPU with four physical cores built into it. It counts as one CPU, hence why BOINC shows it as one CPU.
There are systems out there that use two or more CPUs, all with anywhere from 1 to 32 cores, some of these hyper threaded as well. For these the BOINC benchmarks will only show the physical CPU count, not the core count, or the logical count.

15/02/2019 17:26:58 | | Processor: ]B]4[/B] GenuineIntel Intel(R) Core(TM) i5-3470 CPU @ 3.20GHz [Family 6 Model 58 Stepping 9]

I know. I'm just using the same terminology that LHC@Home uses on their project preferences page, which is consistent with the Event Log. Thus I correctly refer to the number of CPUs in my processor.

2/15/2019 3:21:46 AM | | Processor: 4 GenuineIntel Intel(R) Core(TM) i5-2310 CPU @ 2.90GHz [Family 6 Model 42 Stepping 7]
2/15/2019 5:29:00 AM | | Running CPU benchmarks
2/15/2019 5:29:00 AM | | Suspending computation - CPU benchmarks in progress
2/15/2019 5:29:31 AM | | Benchmark results:
2/15/2019 5:29:31 AM | | Number of CPUs: 4
2/15/2019 5:29:31 AM | | 3577 floating point MIPS (Whetstone) per CPU
2/15/2019 5:29:31 AM | | 10379 integer MIPS (Dhrystone) per CPU
2/15/2019 5:29:32 AM | | Resuming computation

Forums:
Editing posts can only be done up to an hour after posting. Moderators have unlimited post editing times, but can only edit their own posts, not those of others.

Thank you for that clarification.
ID: 90150 · Report as offensive
Steveplanetary

Send message
Joined: 14 Feb 19
Posts: 5
United States
Message 90151 - Posted: 15 Feb 2019, 18:15:15 UTC - in response to Message 90147.  

I said I'd keep everybody posted. This morning when I resumed a 1 CPU Einstein@Home task the CPU usage reported by Task Manager jumped up to ~50%. The LHC task did not go into Waiting to run. For whatever reason it appears the problem has been resolved.

Thank you both for your replies.
ID: 90151 · Report as offensive
jglrogujgv

Send message
Joined: 6 Jul 18
Posts: 49
Barbados
Message 90153 - Posted: 15 Feb 2019, 20:32:47 UTC - in response to Message 90151.  

I suspect the problem might return and eventually go away again. I don't have any GPUs so the following info is just hearsay but there might be something to it. I recall seeing posts at LHC@home mentioning that on hosts running CPU only projects alongside GPU projects the scheduler doesn't do a very good job of keeping all cores busy at all times. Apparently sometimes it does while other times it does not thus the problem comes and goes inexplicably. If I understand those posts correctly there isn't much one can do about it other than learn to live with it.
ID: 90153 · Report as offensive
Steveplanetary

Send message
Joined: 14 Feb 19
Posts: 5
United States
Message 90154 - Posted: 16 Feb 2019, 2:22:52 UTC - in response to Message 90153.  

I suspect the problem might return and eventually go away again. I don't have any GPUs so the following info is just hearsay but there might be something to it. I recall seeing posts at LHC@home mentioning that on hosts running CPU only projects alongside GPU projects the scheduler doesn't do a very good job of keeping all cores busy at all times. Apparently sometimes it does while other times it does not thus the problem comes and goes inexplicably. If I understand those posts correctly there isn't much one can do about it other than learn to live with it.

You're certainly right that if the problem is the scheduler there's nothing users can do about it. I've received a lot of help from MAGIC Quantum Mechanic and he has eight computers, I believe. He's done alpha and beta testing for LHC. On at least one of those computers he has run LHC CPU tasks and Einstein@Home GPU tasks concurrently for many years, and he says he's never had a problem.

My new memory arrives tomorrow. I'll just hope for the best. Thanks for the post.
ID: 90154 · Report as offensive

Message boards : Questions and problems : Problem with CPU benchmark

Copyright © 2019 University of California. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.