Haswell CPU Question

Message boards : Number crunching : Haswell CPU Question
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · Next

AuthorMessage
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14650
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1465347 - Posted: 17 Jan 2014, 9:17:29 UTC - in response to Message 1465280.  

(I can't figure out how to copy the chart they showed, but where (under heavy load) the 3770K was at 59C, the 4770K was at 78C)

This one?

ID: 1465347 · Report as offensive
Batter Up
Avatar

Send message
Joined: 5 May 99
Posts: 1946
Credit: 24,860,347
RAC: 0
United States
Message 1465413 - Posted: 17 Jan 2014, 16:11:14 UTC - in response to Message 1465347.  




That "MAXIMUM LOAD" is an artificial test that puts stress on a CPU one would never see in the real world.

Intel would not make a chip that could not run at "100% CPU utilization" with a thermal solution for a TDP of 84 watts. Haswell is not an "extreme" CPU; it is for the low power/low heat end of the desk-top PC world Intel and M$ is moving to.
ID: 1465413 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1465490 - Posted: 17 Jan 2014, 18:44:16 UTC

I am using a NZXT Respire T20 with my i5-4670K. With all 4 core cooking at full load Core 0, the hottest, stays in the 61-66ºC range. With a thermal limit of 100ºC I don't see this as an issue. After a 5 min cool down period I drop down to 23ºC w/o speed step enabled.

The i7-4770K has a slightly higher thermal junction limit of 105ºC.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1465490 · Report as offensive
Cruncher-American Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor

Send message
Joined: 25 Mar 02
Posts: 1513
Credit: 370,893,186
RAC: 340
United States
Message 1465523 - Posted: 17 Jan 2014, 20:11:18 UTC - in response to Message 1465347.  

Yes, that's the one.
How did you do that, Richard?
ID: 1465523 · Report as offensive
Cruncher-American Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor

Send message
Joined: 25 Mar 02
Posts: 1513
Credit: 370,893,186
RAC: 340
United States
Message 1465525 - Posted: 17 Jan 2014, 20:17:43 UTC - in response to Message 1465413.  




That "MAXIMUM LOAD" is an artificial test that puts stress on a CPU one would never see in the real world.

Intel would not make a chip that could not run at "100% CPU utilization" with a thermal solution for a TDP of 84 watts. Haswell is not an "extreme" CPU; it is for the low power/low heat end of the desk-top PC world Intel and M$ is moving to.


I wouldn't be so sure of that - after all, SETI uses the AVX instructions too, does it not?

Actually, you are right, in a sense. My brother-in-law used to work for DEC in the chip design area (Alphas, and such). They could write benchmarks that would just about liquefy the CPUs because they knew where the circuits were that were worked by each instruction, so they could make a benchmark that concentrated all its sound and fury in a very small part of the chip. In fact, they would try to outdo each other doing so. Ah, competition!
ID: 1465525 · Report as offensive
Cruncher-American Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor

Send message
Joined: 25 Mar 02
Posts: 1513
Credit: 370,893,186
RAC: 340
United States
Message 1465548 - Posted: 17 Jan 2014, 22:19:29 UTC
Last modified: 17 Jan 2014, 22:29:57 UTC

I turned off HT on my 4771.

The CPU temps dropped to about 70C, so I am happy with that. Also, CPU power usage has (according to Coretemp) dropped from 90-95 watts to 70-75 watts. Half a KW a day will save me about a dime, or $35/year.

I'm now running 3 of 4 cores for SETI, and I have the GPUs limited (via app_config.xml) to running <= 3 APs simultaneously on the 2 GPUs. Obviously, GPU computing won't be much affected, but I have lost 4 MB CPU threads, so I am curious to see the effect on RAC. I would estimate a drop of about 2K (~500/thread/day x 4 threads)

In other (forum) threads I have read that there is a Credit cost associated with HT, so perhaps it won't be that bad. We shall see!
ID: 1465548 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14650
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1465574 - Posted: 17 Jan 2014, 23:58:29 UTC - in response to Message 1465523.  

Yes, that's the one.
How did you do that, Richard?

I happen to be using Chrome as my browser, which may make it easier.

Followed your link to the site: http://www.xbitlabs.com/articles/cpu/display/core-i7-4770k_12.html

Read the article until I found the chart I thought you might have been referring to.

Right-click on the graphic and choose 'Copy image URL': paste gives http://www.xbitlabs.com/images/cpu/core-i7-4770k/t3.png

Wrap that in [img] [/img] tags.

IE has 'properties' on the right-click menu - one of them is Address (URL), which you can select with the mouse and copy with Ctrl-C.
ID: 1465574 · Report as offensive
OTS
Volunteer tester

Send message
Joined: 6 Jan 08
Posts: 369
Credit: 20,533,537
RAC: 0
United States
Message 1465593 - Posted: 18 Jan 2014, 1:11:57 UTC - in response to Message 1465574.  

I have been using a Haswell CPU with no graphic capability (Intel Xeon E3-1230 v3 @ 3.30GHz) in a Linux server with HT enabled to crunch AP units when it wasn’t doing something else. Most of the time that has meant a 100% duty cycle for 8 SETI AP work units as reported by top as the non SETI load is very light.

When I first ran it I saw temperatures in the high 70sC which concerned me. As a result I wrote a bash script to check the temperature of all 4 cores every 10 seconds and email me if a single core reading exceeded 85C. In addition to emailing me it would also kill the boinc process if a single reading exceeded 90C.

What I found was that the script never killed boinc but about a half dozen times a day I would see a temperature reading that was above 85C for one reading. That was enough to make me replace the stock Intel cooler with a Noctua NH-U9B-SE2 cooler.

After I installed the new cooler the temperatures dropped to the mid 60s on average for a full CPU load and the highest the script has caught so far is 71C. That is something that seems more reasonable and I have reset the email warning to 75C. I figure that gives me enough time to clean out the dust bunnies or find out what else might be causing a rise in temperature before it creates a problem.

The bottom line is I am very happy with the way the Haswell CPU is performing doing SETI work with the non stock cooler. Of course it would be nice to receive a little more credit for that work, but that is another story ;-).
ID: 1465593 · Report as offensive
Cruncher-American Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor

Send message
Joined: 25 Mar 02
Posts: 1513
Credit: 370,893,186
RAC: 340
United States
Message 1465607 - Posted: 18 Jan 2014, 2:17:46 UTC - in response to Message 1465602.  

I have two identical computers doing SAH exclusively. Each computer has a Corsair H80 water cooled I7-4770K (Haswell) plugged into a Sabertooth Z87 mobo feeding two GTX780ti's. Running 5 CPU tasks and 4 GPU tasks simultaneously on each of them.

For the life of me, I can't figure out why one of them maintains a CPU temp of ~50C and the other maintains a CPU temp of ~60C.

I've checked and rechecked the cooler installation and added thermal compound on the hot one and nothing changes.

I've swapped processors and the hot processor stays in the same box.

I'm thinking I have a faulty water cooler. The mobo can't make a processor run hotter, can it? (EUFI settings are identical between them.)


Problem maybe with the MB sensor?
ID: 1465607 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1465743 - Posted: 18 Jan 2014, 12:47:57 UTC - in response to Message 1465602.  

I have two identical computers doing SAH exclusively. Each computer has a Corsair H80 water cooled I7-4770K (Haswell) plugged into a Sabertooth Z87 mobo feeding two GTX780ti's. Running 5 CPU tasks and 4 GPU tasks simultaneously on each of them.

For the life of me, I can't figure out why one of them maintains a CPU temp of ~50C and the other maintains a CPU temp of ~60C.

I've checked and rechecked the cooler installation and added thermal compound on the hot one and nothing changes.

I've swapped processors and the hot processor stays in the same box.

I'm thinking I have a faulty water cooler. The mobo can't make a processor run hotter, can it? (EUFI settings are identical between them.)

If you are reading the mobo CPU sensor when there could certainly be some variance giving you the different numbers.
If you are reading the CPU thermal sensors then I would suspect that one of the coolers is a bit off. I would first check the power supply outputs. Perhaps one is giving you 12.1v and the other 11.9v. That would cause the cooling systems to preform slightly differently.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1465743 · Report as offensive
Cruncher-American Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor

Send message
Joined: 25 Mar 02
Posts: 1513
Credit: 370,893,186
RAC: 340
United States
Message 1465785 - Posted: 18 Jan 2014, 14:21:52 UTC - in response to Message 1465548.  

I would estimate a drop of about 2K (~500/thread/day x 4 threads)

In other (forum) threads I have read that there is a Credit cost associated with HT, so perhaps it won't be that bad. We shall see!


Slight correction: the amount of credit I get per CPU thread is about 1000/day when HT. So I should show a drop of around 4K after turning off HT, if there is no effect.
ID: 1465785 · Report as offensive
Batter Up
Avatar

Send message
Joined: 5 May 99
Posts: 1946
Credit: 24,860,347
RAC: 0
United States
Message 1465826 - Posted: 18 Jan 2014, 16:56:01 UTC

There is a video of a guy removing the heat spreader from a Haswell and reapplying a thermal compound between the cores and heat spreader. He had one core running hot, this fixed the problem. So the problem was Intel and indirectly AMD for no longer competing.

http://www.youtube.com/watch?v=XXs0I5kuoX4
ID: 1465826 · Report as offensive
Profile Michael W.F. Miles
Avatar

Send message
Joined: 24 Mar 07
Posts: 268
Credit: 34,410,870
RAC: 0
Canada
Message 1467548 - Posted: 23 Jan 2014, 4:41:17 UTC

With my machine AMD Phenom 2 1100T it always ran hot.
What I did was simple.
Take off the side cover and place a small desk fan to blow in.
Cool as a cucumber now.

Intels run HOT.
Amd will run hot and the max temp is fairly low. 65C is about it.
The fan beside my computer works great.
Now temps at full load running 45-50C

Cheap and easy but you must clean often

Just a thought

Michael Miles
Team New Mexico
ID: 1467548 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13722
Credit: 208,696,464
RAC: 304
Australia
Message 1467571 - Posted: 23 Jan 2014, 5:36:48 UTC - in response to Message 1467548.  

Intels run HOT.
Amd will run hot and the max temp is fairly low. 65C is about it.

It depends on the CPU & the load as to what runs hot & what doesn't.
The present Haswells run hotter than previous CPUs due to the change in the casing, but they still run at similar temperatures to equivalent AMD CPUs since they use about half the power.
Grant
Darwin NT
ID: 1467571 · Report as offensive
Cruncher-American Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor

Send message
Joined: 25 Mar 02
Posts: 1513
Credit: 370,893,186
RAC: 340
United States
Message 1468301 - Posted: 24 Jan 2014, 16:50:53 UTC

OK - I've had a few days to see what going from HT on my I7-4771 to non-HT. So far, things are a little bumpy because of the flood of APs a couple of days ago.
I will post again here after the next 10 days or so, to see how much RAC has been lost.

At present, "Account" -> "Computers on this account" is showing about a 6.3K difference in Avg. Credit (before dropping HT it was about 5K). So not much loss, but I need a bit more stability before making any pronouncements.
ID: 1468301 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1468428 - Posted: 24 Jan 2014, 21:05:54 UTC

On my i7-860 machines I was seeing a RAC reduction of only about 4-7% when running 4 tasks at a time vs the machines still running 8 at a time. I left HT on, but was just running less tasks.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1468428 · Report as offensive
Cruncher-American Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor

Send message
Joined: 25 Mar 02
Posts: 1513
Credit: 370,893,186
RAC: 340
United States
Message 1468515 - Posted: 25 Jan 2014, 2:32:44 UTC - in response to Message 1468428.  

On my i7-860 machines I was seeing a RAC reduction of only about 4-7% when running 4 tasks at a time vs the machines still running 8 at a time. I left HT on, but was just running less tasks.


The problem is that when a real core is made into 2 by HT, they are not fully independent (they share some of the hardware from the original real core). So certain kinds of code cannot process in parallel. So to some extent, how an app behaves on HT vs. real is somewhat app dependent. If you leave HT on, and two HT cores are on the same real core, you may still get the degradation, even if you are running only 4 threads on a HT CPU that can run 8 (or 12) HT threads.
ID: 1468515 · Report as offensive
Previous · 1 · 2 · 3 · Next

Message boards : Number crunching : Haswell CPU Question


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.