i7 needs clean stable +5 and 3.3 - its not all about the +12

B

BillParrish

Supreme [H]ardness
Joined
Aug 25, 2006
Messages
7,519
My ongoing investigation finds the Uncore portion of an i7 draws it's power from the +5 from the main motherboard connector for the onboard memory controllor and power from the 3.3V for the QPI.

I was reading this http://www.lostcircuits.com/mambo//index.php?option=com_content&task=view&id=44&Itemid=1&limit=1&limitstart=2

and it occured to me to wonder how it would affect power supplies for new i7 builds or upgrades, (somewhat astonished some power supply makers did not jump on their products with solid +5 as "i7 friendly" or some such rubbish, Guess its a bit over the heads of most marketing types. :D )



Intel recommends for their own X58 board (and of course the do not consider any OCing when stating their specs/recommendations)

The power supply must comply with the indicated parameters of the ATX form factor
specification.
• The potential relation between 3.3 VDC and +5 VDC power rails
• The current capability of the +5 VSB line
• All timing parameters
• All voltage tolerances
For example, for a system consisting of a supported 130 W processor (see section 1.4
on page 14 for a list of supported processors), 1 GB DDR3 RAM, one high end video
card, one hard disk drive, one optical drive, and all board peripherals enabled, the
minimum recommended power supply is 460 W. Table 24 lists the recommended
power supply current values.
Table 24. Recommended Power Supply Current Values
Output Voltage 3.3 V 5 V 12 V1 12 V2 -12 V 5 VSB
Current --------22 A 20 A 16 A 16 A 0.3 A 1.5 A


My point is that it now seems that it is no longer all about the +12. Most high performance supply's should have sufficient +5 and +3.3 but I have found a few less powerful/expensive ones that are marginal or do not. While Intel recommends at least 20A on the +5 and 22A for the +3.3 for their own X58 board I would want more and I would want it to be clean and just as important would want to know how did loading up the +12 affect the voltage regulation and noise on the +5 and +3.3

I highly recommend that people recommending supplies for i7 systems go back and look at the quality (noise, ripple, regulation) of the +5 and 3.3 of a supply before recommending it for an i7 machine.

Disclaimer: As mentioned my investigation is not complete as there are no new VRD specs from Intel and Motherboard manuf do not bother to provide sufficient technical details of there board designs but Intels own specs and reviews around the web where i7 CPU power requirements that where determined only by measuring the current flowing through the 4 or 8 pin aux 12V power that feeds the CPU VRD (aka VRM) are wildly incorrect supports my preliminary conclusions. There also seem to be a lot more components not near the cpu on X58's that look like VRMs than I am used to seeing.
 
will the OCZ elitexstream 1000w comply with all your listed standards?
 
Off hand I would think so, thats way up there on the quality/power ladder, will look to find a solid review, back in a bit.

Well thats interesting and makes my point for me.

The OCZ website does not have the current capacity listed for the +5 and 3.3. not that I could quickly find anyway.
http://www.ocztechnology.com/products/power_management/ocz_elitexstream_power_supply

And the first review I hit, did not even load the 3.3 (note the little green square over to the right that says "0 watts" despite talking about their custom power supply testing setup.
http://www.overclock3d.net/reviews.php?/power_supply/ocz_elitexstream_1000w_atx_psu/4

So at first glance, "I dont know" and sharply points out how recently (and for good reason) all anyone talks about is the +12. The game has changed.

Looking again for more info on that supply, you got me interested, hopefully I will find what is needed to make a good judgement otherwise that supply might become the poster child for what is not right about published information about supplies for i7's.


ahhh this is better, if true this shows its pretty decent supply, I am looking at Test 3 100% load (3.3/5V weighted) which is perfrect for this discussion, good load regulation, plenty of amps and that is the 800W version.

http://www.bit-tech.net/hardware/200...eam-800w-psu/6

(but still no noise and ripple results).



So, got to do other things, but that 1000W OCZ should be just fine, I had no reason to think it would not be but it was fun to help point out how hard it is to find the info to make a good decision.

I am more concerned about supplys in the 500-600 watt range that might be skimping on the 3.3 and +5 when the +12 is heavly loaded like with an SLI or Xfire setup. Top of the line supplies like that OCZ should not be a problem.
 
Vmem is usually generated from +5V, that's true.
The CPU Vtt VRM, as far as I know, supply the power from +12V, not +5V or +3.3V. I confirmed this after viewing several designs of Vtt VRM of X58 boards.
It's not all about the +12V, but the other two rails won't be loaded very much. Both the 3 rails are filtered by VRM's input filtering caps/inductors, before generating the chip voltages.
 
TravisCHEN said:
Vmem is usually generated from +5V, that's true.
The CPU Vtt VRM, as far as I know, supply the power from +12V, not +5V or +3.3V. I confirmed this after viewing several designs of Vtt VRM of X58 boards.
It's not all about the +12V, but the other two rails won't be loaded very much. Both the 3 rails are filtered by VRM's input filtering caps/inductors, before generating the chip voltages.
Click to expand...

It is true that Vtt is +12. I dont think you read datasheet or the link I provided. Vttd for an I7 that powers the QPI and I/0 in an i7can draw 28A (per Intel datasheet) at full load in stock configurations from the 3.3 over and above the normal 3.3v loads found on C2D systems. That is a significant new 3.3 load. That number is a max and so might be a bit overstated but the datasheet does not give typical values for current. Your point about filtering is correct. My point is that some recent supplies in the midrange power levels might come up lacking on the 3.3v rail when OCing faced with this additional 3.3v wattage requirement as all the emphasis in the last 2 years has been pumping up the +12. Normally cross loading has not been an issue, with medium powered supplies I think it now has to be looked at more closely.

However it occurs to me I did not look at what moving the memory controller to the cpu did for 3.3v current consumption on the MCH ( I guess its more accurate to call it a northbridge now :D ) and if it was using any 3.3V wattage that is now eliminated from total system draw. I will leave that for someone else.
 
BillParrish said:
It is true that Vtt is +12. I dont think you read datasheet or the link I provided. Vttd for an I7 that powers the QPI and I/0 in an i7can draw 28A (per Intel datasheet) at full load in stock configurations from the 3.3 over and above the normal 3.3v loads found on C2D systems. That is a significant new 3.3 load. That number is a max and so might be a bit overstated but the datasheet does not give typical values for current. Your point about filtering is correct. My point is that some recent supplies in the midrange power levels might come up lacking on the 3.3v rail when OCing faced with this additional 3.3v wattage requirement as all the emphasis in the last 2 years has been pumping up the +12. Normally cross loading has not been an issue, with medium powered supplies I think it now has to be looked at more closely.

However it occurs to me I did not look at what moving the memory controller to the cpu did for 3.3v current consumption on the MCH ( I guess its more accurate to call it a northbridge now :D ) and if it was using any 3.3V wattage that is now eliminated from total system draw. I will leave that for someone else.
Click to expand...

Link to datasheet that says the i7 will pull 28A @3.3v from the 3.3v rail?
 
Paul_Johnson said:
Link to datasheet that says the i7 will pull 28A @3.3v from the 3.3v rail?
Click to expand...

Not a link to what you asked.

This however is a link to perhaps an explanation of some of misunderstanding:

http://www.lostcircuits.com/mambo//...task=view&id=44&Itemid=1&limit=1&limitstart=2

About half way down the page is a diagram.

I wonder if 7-UP had something to do with calling the additional circuits inside the CPU the “Uncore”. As it happens the page linked it #3 of 5, the entire article is worth a read.
 
You must log in or register to reply here.
Back
Top