High Vdimm and default or low Vcore = possible troubles on newer A64s

http://www.xtremesystems.org/forums/showthread.php?t=79509&page=1&pp=25

The following is a quote by Tony (of OCZ) from the above linked thread. It relates to cpus dying when they are operating in environments of high Vdimm but either default or otherwise low Vcore. It is worth looking into and perhaps heeding the warning.

Quote:

Guys this is most important

If you run a high vdimm and a low vcore with E die etc you are running the risk of killing the cpu's. So if you have VX etc and you are not raising the vcore on your cpu you stand a chance of pumping a lot of current into the memory controller and causing permenant damage.

Im seeing people running $1k cpu's at default vcore and BH5 at 3.4V, this is a sure way to kill the cpu and is NOTHING to do with an issue on the board. If you are going to run high vdimm run a higher vcore also, and keep your CPU's ok.

End of quote

This should be of particular interest to those running Opterons since one of their major attraction is the ability to achieve high Overclocks with low Vcore.

And to think that all along we thought we were doing our cpus a favor by keeping the Vcore as low as possible!

If I read this correctly, in the thread he refers to having gotten this info directly from AMD so I wouldn't be quick to discount it.

wow thats evry interesting, my ram is 2.75v tho and If I wasnt using 2gigs I would be on TCCD only.

Based on what has been discussed in that thread, you should be OK. It appears that the issue is still possible with TCCD and similar ram, provided you are running the cpu low enough...its the "gap" between the two settings that creates the problem. How large a gap is the real question and I don't know that it has been firmly established yet at what point there is a proble.

Clearly Winbond at 3.4v and an Opteron at 1.4v is a candidate for trouble.

Reefa_Madness said:

Based on what has been discussed in that thread, yo should be OK. It appears that the issue is still possible with TCCD and similar ram, provided you are running the cpu low enough...its the "gap" between the two settings that creates the problem. How large a gap is the real question and I don't know that it has been firmly established yet at what point there is a proble.

Clearly Winbond at 3.4v and an Opteron at 1.4v is a candidate for trouble.

Click to expand...

yeah I guess any vdimm over 3.0 while using 1.4 vcore is a candidate one would imagine.

Yeah, 3.4 to 1.45 is one hell of a difference in voltages, and in that small space (something like, what, 90mm^2 now?) is really asking for it. The step down from RAM data/address bus -> CPU internal should be kept as low as is possible. This may minimize the likelihood of a failure.

Hrm, I always thought CPU and RAM used their own voltage lines.

The reason for all of this is still a little murky for me, but that is why I made the comment about this stuff originating from AMD (per Tony).

Very interesting. Let us see how this actually works.

Tony mentions "Diodes". My guess is these are Zener Diodes he is referring to. A zener will not operate properly if it isn't fed adequet current. Zener's need current biasing for them to work properly. Another post says :

Martin Cracauer said:

Half of the RAM voltage ("Vdimm") goes into the CPU via the memory controller (Vtt/Vref). So if you have memory at 3.6 volts, 1.8 volts arrive at the CPU. The CPU protects itself from that voltage via diodes. On one side of the diodes is the CPU core voltage ("Vcore"), on the other side is the Vtt (half of Vdimm).

Click to expand...

Just so that everybody understands how one can burn up a Zener, here are a few basics that would prove useful. (Taken from NEC's datasheets)

A Zener diode has the characteristic that when a voltage is applied in the reverse direction, it changes to low impedance when the voltage reaches a certain value, at which point the current starts flowing rapidly. The principle consists of exploiting the following breakdown phenomena that occur when the PN junction direction is reversed.

(1) Zener breakdown occurring through quantum mechanical tunnel effect
(2) Avalanche breakdown caused by electrons occurring in high electric field and avalanche-like increase in electron holes.
Let us assume these two breakdown phenomena are characterized by a threshold of 5 to 6 V. The Zener breakdown is dominant when the voltage is under 5 to 6 V, and the avalanche breakdown is dominant when the voltage is higher than 5 to 6V.

The key here is "Temprature". If your temps are under control, the Zener will not break down. After all, the Zener is designed to operate in the breakdown region, right? The zener diode is sensitive to temperature. Excessive temperature will destroy a zener diode, and because it both drops voltage and conducts current, it produces its own heat in accordance with Joule's Law (P=IE). If this heat is not dissipated fast enough, you will cook your Zener.

Another problem would be to "Undervolt" the Zener. In this case, it will STOP regulating the voltage and sometimes dangerously high currents can flow through your CPU. I'm not really sure about all of this, but to me it seems that this issue can be alleviated if the CPU supply is divorced from VDRAM. Also, I'm not sure if this is a problem with ALL motherboards or with DFI alone.

On a related note, our poor friend mikeguva lost 3 CPU's... Maybe a connection could be made here?

http://www.xtremesystems.org/forums/showthread.php?t=79187&page=1&pp=25

Supposedly, I'm in the safe regimn 1.45 x 113% , BH-5@ ~ 2.68V. Nothing bad has happened so far, so I must be O.K. Not much data is available at this point. It would be interesting to know if the CPU's died a slow death or instantaneously and if adequet cooling can prevent it? My guess is Thermal runaway would take some time, but it also depends on the types of diodes AMD use.

thanks reefa, i'm glad i found this thread. very good info to know for my upcoming 64 build.

s-n? so if i understood correctly (and i probably didn't)... having vdimm @ 3.2v would put 1.6v into the cpu's mem controller, thus you would want vcore at or close to 1.6v to minimize problems?

just more solid proof to stay away from high voltage memory and to not use the 4 volt jumper even if it did work.

SN,

Well thanks for putting that in layman's terms.... . I feel better now.

Darn glad I chose to go into accounting, as calculating the cost to produce the projected number of widgets required to meet customer demands over the next production cycle is just more up my alley.

As long as I don't fry my next cpu I'll be happy.

And Rattle...as long as you feed your cpu over 1.8v, then you can run that BH-5 at 3.6v without any issues. Sounds doable to me.

Well, reading the first few lines you wrote, I'm glad to do Physics instead of accounting

Lets take a quick peek at the Athlon64 voltage regulation standards suggested by AMD to see if its the MoBo maker's fault or not?
Some clues I found in THIS document:

The AMD Athlon processor PLL is powered by VCCA on the processor module. CCA is derived by a resistor divider powered by VCC_SRAM from the motherboard. The processor PLL does not lock if VCCA is not high enough for the processor logic to switch for some period of time before PWROK asserts. The recommended minimum time before PWROK assertion is 5μs.

Click to expand...

The motherboard has the following two voltage regulators:

The AMD Athlon processor core voltage regulator
(VCC_CORE)
This regulator uses 5V and provides a processor core voltage determined by the voltage ID (VID). A typical core voltage is 1.6V. The 5V is provided by the silver box power supply. Current motherboards use 5V to power the VID resistor network.



SRAM core voltage regulator (VCC_SRAM)T
This regulator either passes 3.3V from the power supply or 2.5V. The voltage is determined by VCC2SEL, which identifies if the SRAM requires a core voltage of 2.5V or 3.3V.

Although the system clock generator is powered from the 3.3V supply, the outputs are open drain and pulled up to the AMD Athlon processor core voltage. Therefore, the system clock does not really toggle until the processor core voltage has ramped.

A resistor network on the processor module provides VCCA from VCC_SRAM. VCCA nominally should be 2.5V and therefore, the network may divide down from 3.3V.

Click to expand...

Do the above two paragraphs tell us that drawing on the VDram line is the recommended technique in the Power ON sequence? I don't know anything about this. Maybe somebody who does (Moto, emboss or yeha) can tell us what is going on?

Another interesting whitepaper:
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/22651.pdf

Where is the Zener located on the motherboard?

Interesting information Nade.

Reefa_Madness said:

SN,

Well thanks for putting that in layman's terms.... . I feel better now.

Darn glad I chose to go into accounting, as calculating the cost to produce the projected number of widgets required to meet customer demands over the next production cycle is just more up my alley.

As long as I don't fry my next cpu I'll be happy.

And Rattle...as lonng as you feed your cpu over 1.8v, then you can run that BH-5 at 3.6v without any issues. Sounds doable to me.

Click to expand...

LOL 1.8 hahahaha

no lower the voltage the better plus I'm on air hehe

IWasHungry said:

Where is the Zener located on the motherboard?

Interesting information Nade.

Click to expand...

No zeners on the motherboard. The Zeners are in the CPU. I'm not sure if they are outside or inside the CPU Die. Voltage is shaped by Voltage regulators (LM3xx) series, on the motherboard. They are a lot better than bare Zeners. I think there is a precision voltage divider network on the MoBo which sources power to the CPU. Not sure where though.

I'm more interested in knowing WHY, CPU needs SDRAM voltage sourced to it? If I were to speculate, I'd say this has something to do with how the POST sequence is handled. Comeon guys, somebody tell me what is going on?

Why does the A64 need SDRAM voltage to it? Onboard RAM controller.

I suppose so. If you look at the tech doc. There is an initial timing signal from the SDRAM, which isnecessary for POST. I still don't see why VZener =(VCore ~VDRAM) ?

im scared..........

lol im scared too.. seeing as i own a DFI, and im running a Venice @ 1.36V and my VDIMM @ 3.2V rofl.. infact, im very worried and im wondering if ive already done some major damage >_< if i set the VDIMM to 2.5V (stock) my pc isnt stable because the RAM now cant even keep up 200mhz @ stock timings!.. what should I do???

Careface*

Careface said:

lol im scared too.. seeing as i own a DFI, and im running a Venice @ 1.36V and my VDIMM @ 3.2V rofl.. infact, im very worried and im wondering if ive already done some major damage >_< if i set the VDIMM to 2.5V (stock) my pc isnt stable because the RAM now cant even keep up 200mhz @ stock timings!.. what should I do???

Careface*

Click to expand...

If the PC3200 won't run DDR400 at default timings and voltage, then it is clearly a prime candidate for an RMA (if still under warranty) or an outright replacement. With the price of a gig of value ram being so low I would certainly not risk this cpu or any future cpus by continuing to use generic ram requiring 3.2v to run sub 200MHz speed.

My point is not to be critical, but to simply state the obvious which is that the ram is clearly defective and in need of replacement.