PDA

View Full Version : 0.25v-1.52v is NOT the correct VID range for sandy bridge.



Falkentyne
02-01-2011, 09:47 PM
1.52v is the maximum VID that is expressed in binary, that the CPU can accept a signal from, using the VRD 12 specification (possibly WRONG, see below). I believe someone said that the vcore the chip requests at auto voltage is the voltage it requests to run at at that frequency, based on that specification.

This has NOTHING to do with "Absolute maximum" voltage.

If you remember on the c2q 45nm chips, maximum VID was 1.3625 while absolute maximum voltage was 1.45v. Notice max VID is LESS than absolute max vcore. On the datasheet, the VID went up to 1.600v, but this was simply for the sake of completeness when entering binary bits as 1 or 0.

On 45nm i7, maximum VID was 1.35v while absolute maximum voltage was 1.55v. (proof here: http://download.intel.com/design/pro...hts/320834.pdf )

Now on sandy bridge, maximum VID is 1.52v, while absolute maximum voltage is...(unknown).

Now I honestly don't know why the maximum VID is 1.52v here. This doesn't make sense. Both previous processor families (C2Q/wolf series and core i7) had max vid around 1.35v.

I suggest you guys look at this carefully.

Go back to the core i7 chart for 45nm.
Notice in the vid CHART (3 graphs), the maximum VID possible that is shown in binary is 1.600v ?

But the maximum VId allowed for the processor is 1.35v? (minimum is 0.825v)

Now look at the SB chart here:
http://download.intel.com/design/pro...hts/324641.pdf

Both the CHART and the vid range ENDS at 1.52v, and starts with an ungodly low voltage of 0.25v. . This does NOT happen with previous processors, except for the binary VID chart, which is not a proper VID range for the processors. Something doesn't match up correctly here. Almost seems like someone made a design error making the documents, and made the VID range match the chart, which is clearly not what is supposed to happen. For proof, look at the 45nm I7 vid range: it starts at 0.825v at minimum. A processor can actually RUN at this voltage. But Sandy Bridge starts at 0.250. Good luck making any processor operate at that voltage, even under LN2. Someone clearly just copied the Binary field for vid range from the chart, and no one actually put the VALID proper vid range for the processor. (Willing to bet its about 0.825v-1.35v).

I'll bet anyone hard cold money that 1.52v is NOT the "max vid" for this processor. Unless you want to prove to me that 1.60v was the max vid under valid vid range, for the i7 45nm. Anyone care to take me up on that?

Unlike some of you guys, I don't have higher ups over at Intel, so me contacting them may do nothing...

Johnny87au
02-01-2011, 10:37 PM
yes yourr right it isnt the max VID, this has been discussed several times..

Skratch
02-01-2011, 10:43 PM
Im at stock voltage and 4ghz so far.if I can get 4.5 ghz at 1.3 ill be happy.

unclewebb
02-01-2011, 10:43 PM
The 2nd Generation Intel Core datasheet shows the VID table starting at 0.00000. That value is used when the CPU enters the C6 sleep state.

The 45nm Core 2 mobile processors like many of the T8000 and T9000 series had this spec:

VID=0.650-0.859 V [C4]
VID=0.600-0.850 V [DC4]
VID=0.350-0.700 V [C6]

The very low VID values are used when the CPU is in some of the various C States. The table in the Core 2 mobile datasheet also has VID values all the way down to zero volts.


On the datasheet, the VID went up to 1.600v, but this was simply for the sake of completeness when entering binary bits as 1 or 0.

The Core 2 desktop went up to 1.6125 and the mobile VID goes up to 1.5000. These values work correctly. That's part of the Intel spec and motherboards should be able to deliver that voltage to the CPU if requested.

scook9 needed every bit of that VID value to get this record setting performance out of his QX9300.

http://ro.hwbot.org/forum/showthread.php?t=8055

You can adjust CPU-Z so it reports VID instead of actual core voltage on the Core 2 desktops and it shows the full amount on an Extreme CPU too.

http://img269.imageshack.us/img269/1479/videxample.png

I don't see that the VID values in the latest datasheets are much different than previous generations in terms of accuracy.
They have to be accurate so motherboard manufacturers can design their boards correctly.

Falkentyne
02-02-2011, 12:23 AM
Thank you very much for your reply, UB:

But Unclewebb:
Can you explain this for me?

Why does the vid chart for the 45nm core i7 go from (i forgot, super low value here) up to 1.600v, but the vid range shown on the table in the PDF is 0.825-1.35v? Why not 0 (I believe some charts started negative) up to 1.60V ?

What is the purpose of the "Vid range" of 0.825-1.35v there?

Is this what USED to be called "functional voltages" (aka "your CPU can expect full reliability and longevity when used at this range"?)

And why is that much smaller range gone from sandy bridge? Along with the "absolute maximum" voltage? I guess that's what I wish I could understand. (I'm sorry for sounding like a total idiot here; I know you know far more about this stuff than me....I'm just a joe sixpack overclocker and gamer :P)

I think that's what we really need now, so we can avoid burning up our chips...

unclewebb
02-02-2011, 11:05 AM
I think Intel probably removed the Absolute Maximum Voltage spec for Sandy Bridge because people on XS tend to use that number as a starting point before they do some serious overclocking. :D

I don't have time at the moment to pour through the Intel docs and read between the lines to try and understand what the Intel numbers really mean. I think what you are really looking for is a magic number which tells you how much voltage you can get away with without having to worry about your CPU degrading. That's impossible to say. Temperature also plays a part in this so what a person can get away with when water cooled can be different than someone using air cooling. Load is also a factor. If you play games that mostly only load a couple of cores, you can probably get away with more MHz and voltage compared to running LinX 24/7 with SP1.

Table 7-5 of the datasheet shows this:


Unless otherwise noted, all specifications in this table are based on estimates and simulations or empirical data. These specifications will be updated with characterized data from silicon measurements at a later
date.

About all you can do is wait 6 months and see how much voltage users are able to get away with and see if Intel updates their datasheet with some voltage numbers based on real world testing. If you burn up a 2600K, at least it's not as expensive as burning up a 980X.

Ket
02-02-2011, 11:23 AM
Guesstimating maximum safe 24/7 voltage range isn't that hard if based on the nanometer manufacturing process;

90nm - 1.65v
65nm - 1.55-1.6v
45nm - 1.45-1.55v
32nm - 1.38-1.45v

Not really rocket science. :shrug:

To back those numbers up a bit;

A64 3500+ I ran at 1.6v 2.9GHz, still alive today
E4400 I ran at 1.55v 3.2GHz, still alive today
Q8400 I ran at 1.47v 3.6GHz, still alive today