jonnyGURU Forums

jonnyGURU Forums (http://www.jonnyguru.com/forums/index.php)
-   Testing Methodology Discussion (http://www.jonnyguru.com/forums/forumdisplay.php?f=15)
-   -   Intel's Haswell testing methodology (http://www.jonnyguru.com/forums/showthread.php?t=10253)

jonnyGURU 05-25-2013 05:59 PM

Intel's Haswell testing methodology
 
So I've seen a lot of companies posting that their group regulated PSUs are "Haswell ready". Sure.. anything is Haswell ready if you disable C7 sleep state in the BIOS, but that doesn't meet Intel's definition of Haswell ready, so it's sort of misleading.

If any of you reviewers with load testers want to duplicate Intel's testing for Haswell:

Put 0.1A on +12V1.
Put 0.05A on +12V2.
If the PSU has a single +12V rail, 0.15A load on the +12V does the same thing.
Put the maximum load on the +3.3V and +5V.
The way I calculate this is to take the maximum and divide by 4.15 and then divide by 2.
For example: Max combined for +3.3V and +5V is 150W. (150/4.15)/2=18A per rail.

A pass is when all voltages are within 5%. Typically, the PSU, if group regulated, will go beyond 12.6V.

I've found that even if the load on the non-primary rails is a more realistic lower load, like 5A per rail, the PSU's voltages will be in spec, but if the load suddenly changes, like a drop in the +3.3V and +5V or a sudden rise in the +12V, the PSU may shut off. This emulates going into or coming out of sleep state.

Good luck!

ferky 05-26-2013 09:14 AM

Quote:

Originally Posted by jonnyGURU (Post 98134)
So I've seen a lot of companies posting that their group regulated PSUs are "Haswell ready". Sure.. anything is Haswell ready if you disable C7 sleep state in the BIOS, but that doesn't meet Intel's definition of Haswell ready, so it's sort of misleading.

Basically, if there is an average VGA, couple of HDD and fans, installed in a system ... you've got a couple of amps pulled on a +12V rail.

They probably used that as an average user PC, but exactly that part could be misleading. What I mean is - if a user were to use an CPU integrated GPU, little or no fans and a SSD. This scenario could lead to some problems in a C6/C7 CPU state.

Quote:

If any of you reviewers with load testers want to duplicate Intel's testing for Haswell:

Put 0.1A on +12V1.
Put 0.05A on +12V2.
If the PSU has a single +12V rail, 0.15A load on the +12V does the same thing.
Put the maximum load on the +3.3V and +5V.
The way I calculate this is to take the maximum and divide by 4.15 and then divide by 2.
For example: Max combined for +3.3V and +5V is 150W. (150/4.15)/2=18A per rail.
Is there an official Intel specification published on the web? Or is this it, what you wrote (max W on minor rails and 0.05A on +12V rail)?

I ask this because it really sound exaggerated to put max W on the minor rails while the +12V is chilling @0.6W :)

Quote:

A pass is when all voltages are within 5%. Typically, the PSU, if group regulated, will go beyond 12.6V.
BTW there really is no group regulated PSU on the market that could pass this one (even without shutting down)?

Quote:

I've found that even if the load on the non-primary rails is a more realistic lower load, like 5A per rail, the PSU's voltages will be in spec, but if the load suddenly changes, like a drop in the +3.3V and +5V or a sudden rise in the +12V, the PSU may shut off. This emulates going into or coming out of sleep state.
Very interesting stuff. Is there a pattern on what could separate the non shut off group regulators, from the shutting off ones?

jonnyGURU 05-26-2013 10:07 AM

There's no public document. Only internal, so nothing I can share.

Yes, no group regulated PSU would pass Intel's testing.

Not sure why the PSU shut off. I had a CX600 on the tester and wanted to see how high I could get the +3.3V and +5V without having the +12V go out of spec.

Test 1 was programmed for just the +12V load and nothing on the +3.3V and +5V. Test 2 put 5A on the +3.3V and +5V. I then used the up arrow on the tester to increase the load on the +3.3V and +5V. When I hit 5.6A, the +12V hit +12.61V. I then clicked back over to test 1 to take the load back down. When I did this, the PSU shut off.

crmaris 05-26-2013 10:37 AM

thanks for sharing this info with us Jonny. I will modify my CL1 test. Instead of applying 2 A load at +12V (1 A for each one of the two true +12V rails of my fixture) I will apply the loads you wrote. This way the CL1 test from highly unrealistic will become really useful :D

crmaris 05-26-2013 10:49 AM

What about 5VSB? Zero load there? Not that it plays any role but just asking how they do it there in Intel :)

Philipus II 05-26-2013 11:41 AM

be quiet! claims the Straight Power E9 are Haswell-ready. Seeing your defintion here I can hardly imagine full haswell support according to the intel testing. The situation for the other Aurum based psus (Coolermaster Silent Pro Gold 450/550, FSP Aurum, EVGA) should be the same.

jonnyGURU 05-26-2013 01:00 PM

Quote:

Originally Posted by Philipus II (Post 98142)
be quiet! claims the Straight Power E9 are Haswell-ready. Seeing your defintion here I can hardly imagine full haswell support according to the intel testing. The situation for the other Aurum based psus (Coolermaster Silent Pro Gold 450/550, FSP Aurum, EVGA) should be the same.

Let's just say if I posted the results of the different PSUs I've tested with the Haswell crossload that claim pass the testing that people would think I have an axe to grind. That's why it's better that this information is out there and someone that doesn't work in the industry do the testing.

crmaris 05-26-2013 03:21 PM

1 Attachment(s)
My first victim:
Fractal Design Tesla R2 1000 W -> passed (as it should of course).

jonnyGURU 05-26-2013 03:51 PM

Oh.. you asked about +5VSB. Nothing on +5VSB either.


All times are GMT -4. The time now is 02:05 PM.

Powered by vBulletin® Version 3.8.7
Copyright ©2000 - 2017, vBulletin Solutions, Inc.
Copyright © JonnyGURU.com 2000 -