Simple question for overclockers

[William Gaatjes]

I have once read that the memory controller of phenom II processors do not like it when you change the voltage from 1.5 Volt to 1.65 or 1.7 Volts.
is this true ? How does the memory controller respond to increased voltage with respect to reliable operation and endurance ?



I want to combine an athlon II with Geil memory of 1333 DD3 ram CL6 6-6-20.

I know the stock maximum memory speed for DDR3 is 1066 MHz with the Athlon X2. I will reuse a proven idea i had, to buy fast memory modules for a given clockspeed and run them on a lower clockspeed to get tighter timings with a minimal voltage increase. I did this with my current system and it has proved to work for a couple of years now. The increase in price justified the responsive nature of my current system.

My idea was to get the Geil memory and run it on 1066 with a minimal voltage increase and test to see how much i can shave of the timing settings. After endurance testing that would be the specific timings that i would use.

 

[Zap]

I think you have it confused with Core i7. Intel has stated that memory voltages above around 1.65v can damage the CPU. Anecdotal evidence suggests that it is the difference in voltage between the Uncore and memory that does the harm, so some people run higher Uncore voltages along with higher than 1.65v on memory.

Something similar happened with Athlon 64. I remember back in the socket 939 days and the old DFI Street forums that some people speculated it was the difference between core and memory voltage that killed stuff. People tested the theory by lowering CPU voltage and raising memory voltage, and a few CPUs died in the process.

I will reuse a proven idea i had, to buy fast memory modules for a given clockspeed and run them on a lower clockspeed to get tighter timings with a minimal voltage increase. I did this with my current system and it has proved to work for a couple of years now. The increase in price justified the responsive nature of my current system.

You can also put the money towards a faster CPU or more RAM, or something else. IMO getting that last bit of performance out of memory comes pretty much last (as long as you don't have painfully slow memory) after all the other upgrades and tweaks. Heck, these days the money may be better spent on an SSD. Indeed if you want your normal Windows usage to be super responsive, likely a cheap dual core with the slowest RAM kit (as long as sufficient amount of RAM) with an SSD would feel more responsive than the fastest Core i7 with the fastest and lowest latency RAM... running off a normal hard drive.

 

[Arkaign]

1.7V on PhII!!?!?!? I don't think it's just going to be the memory controller that's upset with you.

edit : oh, maybe you mean the actual ddr3 memory voltage. Nevermind 🙂 lol

 

[William Gaatjes]

I think you have it confused with Core i7. Intel has stated that memory voltages above around 1.65v can damage the CPU. Anecdotal evidence suggests that it is the difference in voltage between the Uncore and memory that does the harm, so some people run higher Uncore voltages along with higher than 1.65v on memory.

Something similar happened with Athlon 64. I remember back in the socket 939 days and the old DFI Street forums that some people speculated it was the difference between core and memory voltage that killed stuff. People tested the theory by lowering CPU voltage and raising memory voltage, and a few CPUs died in the process.

That is something i can entirely understand. Many complex logic chips like cpu's but also more and more microcontrollers have a core voltage and an io voltage. Since in the case of a cpu the i/o has to be fast, there are limited options to build that "bridge" between the core and the i/o. One side of the bridge runs at a lower level then the other side of that bridge. I am not planning on raising the NB voltage of the athlon II. I am just interested in the finding out if the athlon II can handle an increase from 1.5 Volt to a maximum of 1.7 Volt. Although i am planning to never set it that high. It wil likely be 1.55V or 1.6V. Maybe i can leave it 1.5Volt. I did some digging around the internet but i was not really satisfied with the results i found. I still have this mental itch.

You can also put the money towards a faster CPU or more RAM, or something else. IMO getting that last bit of performance out of memory comes pretty much last (as long as you don't have painfully slow memory) after all the other upgrades and tweaks. Heck, these days the money may be better spent on an SSD. Indeed if you want your normal Windows usage to be super responsive, likely a cheap dual core with the slowest RAM kit (as long as sufficient amount of RAM) with an SSD would feel more responsive than the fastest Core i7 with the fastest and lowest latency RAM... running off a normal hard drive.

With the amount of memory i will have most programs run from memory anyway. Even with my current system with 1GB of memory i hardly need the HDD. It's only when opening or closing a program. Files are just not that big enough to cause a delay. I guess the HDD cache takes care of that.

Although i value your opinion, an SSD is not an option. For as when i would buy one it would be an industrial model and those are to expensive to justify the in my experience limited speed-up. I did my research to SSD. For now i will stay with a HDD.

Although i do not want to start an SSD/HDD flame war this is my opinion as written as a response on another poster :

Indeed. But fortunately there is a positive scenario.
Although most flash media have a data retention on average of 10 years, it does not mean that flash or eeprom can hold data longer... It depends on the process used and the materials. For all those SSD drives and usb flashsticks , flash writes need to be fast. The issue is, you want to write fast and be more prone to errors or you want to your data to be save and have low write speeds. Reading data to has it's influence but less then high write speeds.
I myself do not bet on using flash media as a backup device. I have tried it as a test next to a common HDD. And the flash device failed. And it is not the only time i have seen this. Many people will come with calculations of possible failure. But i have reality on my side. Low density flash is almost indestructible , high density flash memory is certainly not.

There is an endurance failure and a data retention failure.
The gate oxide wears out when used. The gate oxide of the floating gate accumulates electrons during every write /erase action. After a while a 1 or 0 written will still give the same result back. And the insulator is not a perfect insulator, leaking electrons over time. This means that the charge on a function floating gate is leaking away. This all get's much worse when the process get's smaller. When a microcontroller with 256 kB of flash memory can retain it's data for 25 years i believe it can. But a high density flash memory of a few Gigabyte build on a modern process ? No, not unless you are willing to pay for physical reliability. With wear leveling algorithms and error detection/correction algorithms you can do only so much. Until all programs and windows itself can be modified to write only to specified mass storage as of my choosing, i do not need an SSD. For every temp file written there must be an option to where that file will be written.