Personal computing discussed
Moderators: renee, Flying Fox, morphine
Vhalidictes wrote:Yes, I had a very minor overvolt for my i7-920, but that didn't help in the end. After 5 years at 3.0GHz, I was losing about 100MHz a month every month. I eventually retired the CPU and switched over to another 920 for about a year, after which it showed the same problem, even though that CPU hadn't been overclocked before.
It seems to be an age thing as best as I can tell. You only get so much OC time before a CPU starts having issues.
jackbomb wrote:Were older CPUs less sensitive to overclocking?
DPete27 wrote:I'd wager it's a combination of other factors also.
Such as: How many hours has your computer actually been running at 4.4GHz? 4 years of 2 hours per week is a lot different than 4 years at 20 hours per week.100 hours per week....etc etc.
Also, chip lottery can play a role. I would agree that your OC doesn't seem to be overly aggressive, but maybe your chip has some weak links.
jihadjoe wrote:Yes the bigger process would make older CPUs more resilient against electromigration. The degradation is caused by material getting moved around, and a bigger node means there's more stuff used for each feature making them harder to damage.
jihadjoe wrote:My Kentsfield is invincible, lol!
Not gonna make the same claim about Sandy Bridge, but I do have an i7-3820 that's been doing 4.5GHz 24/7 since the day it was built in 2012. Old process tech ftw.
Vhalidictes wrote:... I can't imagine that most 14nm processors would last longer than 2 years even at reasonable voltages.
DancinJack wrote:Vhalidictes wrote:... I can't imagine that most 14nm processors would last longer than 2 years even at reasonable voltages.
This makes no sense. I don't know what are "reasonable voltages" in your mind, but if 2 years were an expected lifetime from "reasonable voltages" Intel would be replacing damn near every CPU they sell.
Vhalidictes wrote:Bleeding-edge overclocks/overvolts was what I was talking about.
I'm sure that at default settings most CPUs last at least 10 years.
synthtel2 wrote:All five of my last five desktop CPUs have had issues (some minor) of one sort or another (voltages stock/stock/stock/1.27/1.33). I'm not exactly a strong believer in CPU reliability. Two of those (the 1.27 and 1.33, both Haswell on the same mobo) degraded to paperweight status from their OCs, despite not being very heavily stressed chips.
It's true 1.165V is a bit low for that, especially considering the light load patterns. Have you confirmed lately that thermals are good? When setting up the overclock, did you leave much margin for degradation, or was 1.165V pretty close to the edge?
Chrispy_ wrote:Giving a very small sample size, but one that is still useful statistically, we added 24 nodes (3 stacks of 8 nodes) to our renderfarm when the 4790K came out.
These CPUs are cooled well and perform 4C/8T software rendering at the stock 4GHz for about two thirds of their power-on time. The remaining third of their power-on time is idle and they are running 24/7/365.
So, these chips have had a fairly hard life, but probably an easier life than if they belonged to a consumer who folds or cryptomines. I've replaced three processors due to electromigration failure in the last six months. Two were memory controller degradation, in that the CPUs couldn't complete a clean Memtest x86 pass in a testbench machine with a board and RAM that were known-good. The most recent failure was clockspeed degredation, requiring higher LLC to even POST, and crashing on Prime95 unless it was run at under 3GHz.
So they all ran at stock for 3 years, and then 12.5% of them died from electromigration. Those machines will be replaced with Ryzen or Coffelake in Q2 2018 but I'm curious to see what the final death rate is for that rendergroup once we hit 4 years. I'm also worried about the Ryzen7 parts because unlike Intel, ALL of the die is 14nm FinFET. Intel's 14nm is actually not 100% 14nm, so there is a much higher transistor density on the AMD CPUs. Smaller processes are more susceptible, and Ryzen is the smallest, most dense design to date.
DrDominodog51 wrote:If the node was an issue, Ivy Bridge would have had the same issue. Ivy Bridge, however, was fairly voltage tolerant (once delided).
It is my understanding that Haswell and Broadwell were fragile CPU architectures. My favorite theory is that the FIVR is degrading before the rest of the CPU and gradually has worse voltage regulation over time.
jackbomb wrote:And yeah, 1.165v was right at the edge. It never was able to do 4.4GHz at only 1.160v.
mcarson09 wrote:So who was actively cooling their VRMs and ram while they were overclocking and had their cpus fail?
Kougar wrote:But I would assume servers wouldn't run that setting outside of Intel's spec, certainly not as default at any rate?
Waco wrote:Kougar wrote:But I would assume servers wouldn't run that setting outside of Intel's spec, certainly not as default at any rate?
Servers won't, but he's talking about desktop boards unless I'm mistaken.
mcarson09 wrote:Waco wrote:Kougar wrote:But I would assume servers wouldn't run that setting outside of Intel's spec, certainly not as default at any rate?
Servers won't, but he's talking about desktop boards unless I'm mistaken.
Certain 2 Processors have an overclocking feature. The asus z10pe-d8 ws and z10pe-d16 ws being amoung those boards. Supermicro has what is calls a hyper-speed setting which is indirect control OCing.
Waco wrote:mcarson09 wrote:Waco wrote:Servers won't, but he's talking about desktop boards unless I'm mistaken.
Certain 2 Processors have an overclocking feature. The asus z10pe-d8 ws and z10pe-d16 ws being amoung those boards. Supermicro has what is calls a hyper-speed setting which is indirect control OCing.
I don't consider those server boards, and further, they don't accept non-Xeon CPUs unless I'm mistaken.
Supermicro is a blacklisted vendor at many institutions for good reason.
mcarson09 wrote:You are mistaken a lot. You can use I7s that fit the sockets in cpu0 only.