mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Software

Reply
 
Thread Tools
Old 2020-08-20, 09:23   #595
mackerel
 
mackerel's Avatar
 
Feb 2016
UK

389 Posts
Default

They way I look at Windows CPU usage reporting is that it shows the thread usage time, which is not a linear relationship to core usage or throughput when HT is involved.

Consider a 4 core CPU with no HT. If you run at that 100%, all cores are used. Take the same CPU (same clocks) and turn on HT. If you run 4 core (using 4 threads), you get the same work done, but now Windows reports 50% usage. Now run 8 threads, you see 100% usage.

How much real work is being done at 100% compared to 50% indicated? It depends a lot on the software. Take Cinebench R15 or R20 on modern processors as an example of something that scales relatively well. Running that with 8 threads compared to 4 would give you around 30% more throughput (score). And this is one of the better cases. Prime95, for practical purposes, is close to 0% because it is able to extract the most performance out of the cores without using more than one thread per core. Seeing around 50% usage in Windows is already using 100% of your cores.
mackerel is online now   Reply With Quote
Old 2020-08-20, 10:20   #596
pepi37
 
pepi37's Avatar
 
Dec 2011
After milion nines:)

3×449 Posts
Default

The most relevant benchmark is time. Take one candidate and run it in 1*4 2*2 4*1. Record time, then turn HT on and do same but no use 1*8 2*4 4*2 and 8*1 configuration, and record time. Compare time and that is most accurate benchmark with most useful data
pepi37 is offline   Reply With Quote
Old 2020-08-22, 16:46   #597
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

19·241 Posts
Default

Quote:
Originally Posted by storm5510 View Post
what is it that I have done wrong all this time.
Take Task Manager's display too seriously; expect 4 cores to do the work of 8.

Even Intel with an ideal process mix expects no more than ~1.1 to 1.3 x, or 4.4 to 5.2 cores equivalent performance on your 4-core hardware, and that's the manufacturer's PR. What you show is 4 fully utilized cores and that's all you've got in the hardware, except for the HT duplication of registers etc. Which help with incidental other workloads while prime95 runs essentially full tilt on the 4 full cores actually present in hardware.
kriesel is offline   Reply With Quote
Old 2020-08-23, 00:01   #598
storm5510
Random Account
 
storm5510's Avatar
 
Aug 2009
U.S.A.

110010001102 Posts
Default

Quote:
Originally Posted by kriesel View Post
Take Task Manager's display too seriously; expect 4 cores to do the work of 8.

Even Intel with an ideal process mix expects no more than ~1.1 to 1.3 x, or 4.4 to 5.2 cores equivalent performance on your 4-core hardware, and that's the manufacturer's PR. What you show is 4 fully utilized cores and that's all you've got in the hardware, except for the HT duplication of registers etc. Which help with incidental other workloads while prime95 runs essentially full tilt on the 4 full cores actually present in hardware.
I understand what you are saying. The utilization percentage is misleading. I allow Prime95 to pick what it wants to use. Sometimes, one or more will appear as logicals. The associated physicals are not used. In total, never more than 4. 1 from each physical/logical pair.
storm5510 is offline   Reply With Quote
Old 2020-08-25, 01:53   #599
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

8,837 Posts
Default

Quote:
Originally Posted by storm5510 View Post
With everything above read, explain this (image attached).
That image shows a CPU which is used ONE HUNDRED percent. Not 50%.
LaurV is offline   Reply With Quote
Old 2020-08-25, 17:53   #600
storm5510
Random Account
 
storm5510's Avatar
 
Aug 2009
U.S.A.

2×11×73 Posts
Default

Quote:
Originally Posted by LaurV View Post
That image shows a CPU which is used ONE HUNDRED percent. Not 50%.
With a little guidance, I figured this out. There are four pairs, not eight individuals. A logical may look unused, but its paired physical is at max, or close t it. The reverse is also true. No pair can exceed 100%. I have also seen them mixed with other applications. A physical may be at 60% and its logical at 40%. Again, no more than 100%.
storm5510 is offline   Reply With Quote
Old 2020-08-26, 06:11   #601
LaurV
Romulan Interpreter
 
LaurV's Avatar
 
Jun 2011
Thailand

8,837 Posts
Default

Quote:
Originally Posted by storm5510 View Post
No pair can exceed 100%. I have also seen them mixed with other applications. A physical may be at 60% and its logical at 40%. Again, no more than 100%.
That's not true either. Each "logical" core can go to 100% independent of the other, and a pair together can reach 200%, hihi. I can easily make my computer (and any computer) show 100% CPU usage just by launching few copies of the "usual" programs (word or excel running some macros, uvision running a compilation, or any multi-thread video editor or cad program I use daily, or just opening 100 threads of "isprime()" in pari/gp with some large number like 50 digits or so).

In that case, all bars in task manager/performance, either logical or "illogical" cores (in my computer, I believe all cores are illogical, because they run very slow, and you can not logically reason with them! ) will be glued to the ceiling, and never drop a pixel.

Windoze fubars that task manager up completely, in the sense that, if HT is enabled in bios, it believes your CPU has 2n cores (where n is the number of physical cores). Therefore, if you run n threads of a task (like P95) with HT disabled, on a system with n physical cores, it will always show a 50% occupancy, because it sees only n cores running from 2n he assumes you have. But this doesn't mean you can not occupy it "100%" (please see the quotes!) if you want. To continue on the analogy made by other user before, you have a highway with 8 lanes, but there is a bridge on it with 4 lanes only. Imagine a police checkpoint on the bridge, but not all cars are checked, only the yellow ones. Most of the programs will run different colors of cars, red, and yellow and pink and green, orange, and purple and blue... So they still can keep the highway busy while not feeling much on the bottle neck on the bridge. When you run those programs in your computer, they will each push cars over that bridge from time to time, according with their priorities, when cars from different lanes try to "merge" in a single band on the bridge in the same time, the task with higher priority will push its car first. But there are always gaps between the cars on lanes, and gaps between the cars on the bridge too, because the programs do other things too, not only computing, they access peripherals, wait for you to type key or move the mouse, etc. These things don't push cars on the bridge (i.e. don't need CPU resources). That is why the 4 lanes on the bridge could, in theory, be shared by 8 (and more) lanes (tasks) on the highway, and yet, the traffic won't slow down. This is how HT was born, the CPU makers wanted to use those "dead times" spaces between cars, and give more work to the police checkpoint on the bridge, so they allow cars from more lanes to merge on the bridge, and split back each on its own lane, after.

What windoze does (well, not exactly, but to keep the analogy), it has no idea about the bridge, and it only measures the traffic on each of the 8 lanes. With all colors of cars, the bridge won't slow the traffic, and Windoze will believe that the highway is busy enough. Now, P95 only runs yellow cars. They run at low priority, to make sure they don't slow the traffic, so any time they meet another car at the merging point, they will make space for that another car, but on the other hand, they will immediately fill all the spaces between other cars when there is no merging conflict. This way, all 4 bands on the bridge go at max speed, and the police checkpoint works full time, but yet, it will appears for windows that the traffic on the highway is only 50% of what the highway can support, in spite of the fact that the police on the bridge works at full capacity. On the other hand, running two copies of P95, or one copy with HT enabled, is like pushing yellow cars on all 8 lanes, the poor policemen are 100% full anyhow with 4 lanes, and you won't get faster output. Contrarily, things will slow down as the drivers of the yellow cars start yelling at each-other and compete for an earlier place in line at the checkpoints. Some will even get out of the cars and start fighting, like in Russian videos with "road rage" on youtube.

This doesn't mean you can't cheat it, in the sense that you can only push green cars through the bridge, without letting any spaces between them, so the police can safely sleep and do no work, but the highway is full, and no yellow cars (lower priority) will ever pass the bridge. In that case you will see 100% CPU occupancy, but yet, the CPU doesn't really do any work, and P95 just waits. This is quite easy to do, and people used this trick to write "messages" on task manager's screen, or even play Doom (search youtube for "Max Holt 896").

Last fiddled with by LaurV on 2020-08-27 at 09:28
LaurV is offline   Reply With Quote
Old 2020-08-26, 07:26   #602
retina
Undefined
 
retina's Avatar
 
"The unspeakable one"
Jun 2006
My evil lair

169C16 Posts
Default

Quote:
Originally Posted by storm5510 View Post
What I have is an i7. Four physical cores and four logical cores.
Not really. You have four computing cores, fed by eight logical cores; two logical cores per one computing core.

Everything is physical; all eight logical core instruction feeders are physical, all four computing cores are physical.

Logical is not the opposite of physical.

[/rant]
retina is online now   Reply With Quote
Old 2020-08-26, 14:37   #603
storm5510
Random Account
 
storm5510's Avatar
 
Aug 2009
U.S.A.

2·11·73 Posts
Default

Okay, I am going to let this go and take everyone's word for what they have written. The last time I studied CPU architecture was in the Computer Dark Ages, the late 1980's. It is working properly and that is all I really need to know.

I thank you all for your time and efforts.
storm5510 is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Prime95 version 29.2 Prime95 Software 71 2017-09-16 16:55
Prime95 version 29.1 Prime95 Software 95 2017-08-22 22:46
Prime95 version 26.5 Prime95 Software 175 2011-04-04 22:35
Prime95 version 25.9 Prime95 Software 143 2010-01-05 22:53
Prime95 version 25.8 Prime95 Software 159 2009-09-21 16:30

All times are UTC. The time now is 22:49.

Wed Oct 21 22:49:32 UTC 2020 up 41 days, 20 hrs, 1 user, load averages: 1.59, 1.74, 1.73

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

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.
A copy of the license is included in the FAQ.