I recently interviewed the CEO of a company that's making a very sophisticated CPU cooling product soon to be reviewed here. His point of view on CPU stress testing is that it is totally artificial and has very little to do with real world conditions: there are almost no desktop processes that requires a CPU to run continuously at 100% load for longer than a few minutes at a time. Yet every hardware review site tests HS by stressing the CPU long term -- 20 minutes, half an hour, even an hour or longer. The stress on a CPU in real-world usage by typical users, power or otherwise, is far more dynamic, jumping up and down constantly.
This was precisely my PoV before I began testing heatsinks and publishing reviews. SPCR HS reviews utilize continuous 100% CPU load testing because it is expected, because no clear dynamic CPU testing procedure exists.
WE also know that the same dynamic load conditions apply to the PSU. The static full power testing integrated into SPCR reviews is probably useful as a gauge of whether the unit delivers its rated power, but asking it to do that for more that a few minutes is probably totally unnecessary.
Any thoughts/comments on how/whether these concepts should impact SPCR's current testing procedures on HS & PSU?
Dynamic PC loads
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
Most computer tasks involve a lot of user interaction, but how about tasks you start before the coffee break? For instance DivX compression and batch MP3 compression? Or screen savers that does scientific work and sends the result over internet?
Not sure, but I would think there are some tasks that will run the processor at max speed for many minutes, and it is nice to know the heatsink will handle it.
Max load at the PSU is probably less likely to happen? But if you defrag the harddrive, burn a CD and play a speedy 3D game at the same time it might happen
Not sure, but I would think there are some tasks that will run the processor at max speed for many minutes, and it is nice to know the heatsink will handle it.
Max load at the PSU is probably less likely to happen? But if you defrag the harddrive, burn a CD and play a speedy 3D game at the same time it might happen
I disagree on the useless-ness of static loading. While it's true that few users will ever run their CPU at 100% load for hours on end it does accurately represent the worst case scenario for the processor.
Testing at levels beyond what a product can be expected to experience is standard in virtually every form production/engineering testing done anywhere. Automakers test drive cars in Death Valley, Washing Machine Makers run their machines on continious loops, Window manufacturerers expose their windows to 250mph winds, etc, etc. It's only by pushing the product beyond its normal operating range can you exaggerate the subtle differences that exist between similar designs.
And you must remember that some users do run their machines at 100% load for long periods. I have a 15 machine render farm that spends most of its time turned off, but when its on each machine runs at near peak load for days, even weeks at a time. If I bought a heatsink whose rating was based on the "normal" usage of only a few minutes of 100% load time what would happen to my machines after 250 hours of loading?
Testing at levels beyond what a product can be expected to experience is standard in virtually every form production/engineering testing done anywhere. Automakers test drive cars in Death Valley, Washing Machine Makers run their machines on continious loops, Window manufacturerers expose their windows to 250mph winds, etc, etc. It's only by pushing the product beyond its normal operating range can you exaggerate the subtle differences that exist between similar designs.
And you must remember that some users do run their machines at 100% load for long periods. I have a 15 machine render farm that spends most of its time turned off, but when its on each machine runs at near peak load for days, even weeks at a time. If I bought a heatsink whose rating was based on the "normal" usage of only a few minutes of 100% load time what would happen to my machines after 250 hours of loading?
-
MikeC
- Site Admin
- Posts: 12285
- Joined: Sun Aug 11, 2002 3:26 pm
- Location: Vancouver, BC, Canada
- Contact:
I'd like to see you do that. Don't think it is possible in any Windows version I know.But if you defrag the harddrive, burn a CD and play a speedy 3D game at the same time it might happen
Russ, I never said useless -- just that static 100% load testing for long periods is not -- generally -- the way most people use their PCs most of the time. I think your ownership of a 15 PC farm makes you an unusal user, to say the least! It renders you... umm... how about abnormal?
Perhaps changing the test procedure is not called for, but it is definitly worthwhile implementing a quiet vs cool switch in many systems. Our general approach is to use high capacity HS with low speed fans; an easily accessed fan switch/control is the obvious answer for long intensive CPU loading. Something like my simple 5/12V switch. A dynamically temp-sensitive system like the AOpen SilentTek is probably ideal. (if it works right -- looks like gekkani is having some issues with it: http://forums.silentpcreview.com//viewtopic.php?t=3470)
Case in point: I am running a P4-2.8 in the system I am typing on now. It has a Zalman 6500B-CU w/a Panaflo at 5-6V. Set low enough I can't hear it. Normally, it runs perfectly fine, with high CPU temps in the 50s. A couple days ago, I spent a couple hours ZIPping & archiving to CD 7 gigs of data from one of the 2 Barracudas on this PC. The ZIPing process on large chunks of data took many minutes & I was running it almost continuously for the whole 2+ hours. CPU temps quickly hit >70C, where I have an alarm set. I ended up having to open up the case & set the fanmate1 controller on the Panaflo up to full (and still got 62-65C). The point is that it's the first time in months that fan has seen more than 6-7V.
Anyway... speaking of fan controllers, I have in hand a Sunbeam Rheobus. 4 separate fan controls, each capable of 20W! Except for the annoying super bright LEDs, it seems to work very well. No weird PWM pulsing on fans, no buzzing or noise fron the unit itself. Feels & looks good (except for said LEDs). Will keep it under observation/use for a while longer before I post a review, but for $20 at SVC, it looks like a winner.
Last edited by MikeC on Wed Feb 26, 2003 10:56 am, edited 1 time in total.
Mike,
why don't you adjust the fan benchmarking a bit .. Just keep the current Idle/100% ratings but also add a few extras. Such as typing a text in word while being on MSN and listening to an mp3 or something along those lines. (You might of course want to pick the same mp3 and same programme and same text all the time).
- Bram.
why don't you adjust the fan benchmarking a bit .. Just keep the current Idle/100% ratings but also add a few extras. Such as typing a text in word while being on MSN and listening to an mp3 or something along those lines. (You might of course want to pick the same mp3 and same programme and same text all the time).
- Bram.
MikeC wrote:I think your ownership of a 15 PC farm makes you an unusal user, to say the least! It renders you... umm... how about abnormal?
I think abnormal is an understatement
I've thought for a long time that every fan in a computer system should be adjustable according to load. It simply doesn't make sense to have a fan that's designed to cool under max load running at that speed when the load is much less.MikeC wrote:Perhaps changing the test procedure is not called for, but it is definitly worthwhile implementing a quiet vs cool switch in many systems. Our general approach is to use high capacity HS with low speed fans; an easily accessed fan switch/control is the obvious answer for long intensive CPU loading. Something like my simple 5/12V switch.
Even easier than a switch would be a simple thermo-control for each fan (Like the one I use on my system's 120mm exhaust fan). It's cheap ( $2.00 worth of Radio-shack components) and best of all its automatic. No forgetting to flip a switch at the right time.
Isnt Intel going to change form factors soon? I think they are? When they do, they should take airflow, noise, fan control ect. into consideration. If they do they should have the motherboards and case standard changed around because I think there are alot of problems with the current design that could be improved upon. There should be some type of fan control for every fan. Either built into the case or the mobo.
But having the fan speed control on the motherboard would require that any device you want to monitor the temperature of would have to report it to the motherboard. That limits the usefullness of it.
With individual temp sensors you can thermally control the fans on your PSU, Hard drives, video card, CPU, sound card, case temps, etc etc etc. All individually, without software.
With individual temp sensors you can thermally control the fans on your PSU, Hard drives, video card, CPU, sound card, case temps, etc etc etc. All individually, without software.
I have 2 main problems with mobo based controls:
1. They require every piece of hardware you want to monitor to be compliant. How many PSU's, GPU's, Sound cards, etc. have you seen that can report temps to a mobo?
2. Every fan then has to run through the mobo. Most motherboards only have 2 or 3 fan headers. I'm of the school of thought that the less amp load you can run through the mobo, the better.
1. They require every piece of hardware you want to monitor to be compliant. How many PSU's, GPU's, Sound cards, etc. have you seen that can report temps to a mobo?
2. Every fan then has to run through the mobo. Most motherboards only have 2 or 3 fan headers. I'm of the school of thought that the less amp load you can run through the mobo, the better.
-
Ralf Hutter
- SPCR Reviewer
- Posts: 8636
- Joined: Sat Nov 23, 2002 6:33 am
- Location: Sunny SoCal
MikeC - I'm in the "keep the 100% stress test" camp too. I really think "worst case scenario" testing is definitely the way to go if you value a stable computer. I think this is especially important if you do 3D gaming, audio/video encoding or other CPU-intensive work.
I run run Prime95 and Memtest86 on every system I build, whether it's for me or a customer. I like knowing that things will run fine under the worst conditions possible.
If you had run Prime95 on your P4 2.8 rig you would have known what would happen to your temps if you put the CPU under load for a while. Yes, I suppose you could squeeze a few less dB out of your rig if you knew you were mostly going to be using it for non-CPU Intensive tasks but if that were the case, why not just go with a Via C-3 or undervolted Celeron in the first place? If you feel you need a powerful CPU like a top-end P4, you should test it and make sure it will work correctly under load. Even if you're thinking of running a reobus you should still stress test. Without doing that you still don't know what will happen if you load your CPU for any significant amount of time. You might even find that you can quite safely run your "high" reobus settings at something less than 12V, but you'd never know if you didn't stress test it.
Look at my main system. It's a P4 2.66 OCed to 3.0Ghz. I mainly use it for gaming and video stuff. I could tweak things here and there to make it run a little cooler but then it would run hotter under load. The temps I get when I've been playing a game or video encoding are within 2-4°C of the max Prime95 load temps I get during my stress testing. I figure that extra 2-4°C is a little insurance so I just suck up the few extra dB and feel a little more comfortable.
The whole stress testing thing just seems like good insurance to me, but I guess it boils down to how stable you want your system to be.
I run run Prime95 and Memtest86 on every system I build, whether it's for me or a customer. I like knowing that things will run fine under the worst conditions possible.
If you had run Prime95 on your P4 2.8 rig you would have known what would happen to your temps if you put the CPU under load for a while. Yes, I suppose you could squeeze a few less dB out of your rig if you knew you were mostly going to be using it for non-CPU Intensive tasks but if that were the case, why not just go with a Via C-3 or undervolted Celeron in the first place? If you feel you need a powerful CPU like a top-end P4, you should test it and make sure it will work correctly under load. Even if you're thinking of running a reobus you should still stress test. Without doing that you still don't know what will happen if you load your CPU for any significant amount of time. You might even find that you can quite safely run your "high" reobus settings at something less than 12V, but you'd never know if you didn't stress test it.
Look at my main system. It's a P4 2.66 OCed to 3.0Ghz. I mainly use it for gaming and video stuff. I could tweak things here and there to make it run a little cooler but then it would run hotter under load. The temps I get when I've been playing a game or video encoding are within 2-4°C of the max Prime95 load temps I get during my stress testing. I figure that extra 2-4°C is a little insurance so I just suck up the few extra dB and feel a little more comfortable.
The whole stress testing thing just seems like good insurance to me, but I guess it boils down to how stable you want your system to be.
Some of us run scientific computing workloads that saturate the CPU for long periods
of time, and we prefer quiet computers too.
Please don't omit the fully-loaded tests -- they provide information relevant to workloads
that do occur in the real world, even if they aren't representative of typical pc usage.
of time, and we prefer quiet computers too.
Please don't omit the fully-loaded tests -- they provide information relevant to workloads
that do occur in the real world, even if they aren't representative of typical pc usage.
-
jamoore9
- Friend of SPCR
- Posts: 152
- Joined: Thu Jan 30, 2003 5:58 am
- Location: Fairfax, Virginia, USA
I'm in this camp. I participate in some distributed computing endeavors on my home PC while I'm at work. They can be pretty rough on the CPU, and I'm not home to oversee the machine and manage its heat. Which is why I pay attention to max load tests and Prime95 torture tests. I also found an article at xBit labs on the built in temperature control fail-safes on P4 processors that was truly fascinating. This is one reason that I won't use an 80mm fan on my CPU HS, won't 5V it (7V min), and use thermally controlled case fans. If things heat up while I'm not there, the case fans kick up, and I know that my CPU will stay cooler than the failure temp. Check out that article here: http://www.xbitlabs.com/cpu/p4-temp/bkh wrote:Some of us run scientific computing workloads that saturate the CPU for long periods
of time, and we prefer quiet computers too.
I'm pretty interested in the Rheobus, though. I could ostensibly turn up the CPU fan when I leave in the morning (since I won't be home to be annoyed by the sound), then turn it down when I get home. That would be very nice. Let us know how it goes, MikeC!