"The Full Desktop" @ - Trinity + Mini-ITX @ $350

[Anonaru]

Why would 1.45V be healthy for the CPU part and unhealthy for the NB/GPU part? Is it simply because it's 23% more voltage than stock? Perhaps the reason the voltage was kept that low is for power consumption reasons, not longevity reasons. Don't forget that the Trinity series were designed primarily as mobile chips that had max 100W TDPs. What if 1.18V just happens to be the minimum voltage to be able to run DDR3-1600 and an 800MHz GPU core? As far as my testing has gone, it IS absolutely the minimum.

Its just as large(really, larger)of a jump as it is from 1.4 base to 1.55 on, say, a Deneb CPU is, and just as bad for longetevity. Why are you shocked in any way whatsoever that people feel like such a large NB jump is unhealthy for the processor? Really, a pound for pound increase would be goign from 1.4 to 1.65, which would be pretty damn suicidal.

I don't think its going to blow the gasket off the chip, no. However, I think it could definitly contribute to a notably shorter CPU life span

Edit: I have found that people tend not to want to go past 1.375 on NBv on this chip.

 

[visiblegorilla]

Sorry if my tone was bad. You gave a good response. No, I'm not shocked that a 23% voltage increase makes people nervous. Do you remember the Mobile Athlon XPs? They were simply the highest-quality samples of the chips that AMD manufactured at the time, selected for their ability to run at very low voltages, 1.2V if I recall. Those chips used the same exact architecture as all the other Athlon XPs which ran at, what, 1.4V? 1.45 maybe? Lol it's been awhile. The key thing is that you could buy a Mobile Athlon XP, put it in your desktop, and run it at 1.4V and get higher overclocks than with any of the not-so-special desktop variants. Knowing that, you would never say "Taking a 1.2V CPU and running it at 1.45V is suicide." They were undervolted from the factory.

EDIT: People are afraid of the unknown. APUs are a new concept and will inevitably upset our ideas about how overclocking works. Now there are digital 8-phase power converters where once 4-phase analog was state-of-the-art. Now we have integrated memory controllers (remember how important the Northbridge used to be?), LLC, dynamic voltages and clock rates... It's the march of progress, and people get uncomfortable, especially if they're successful under the status quo. There is so much misinformation about these Trinity chips, and I intend to show their potential. So what if that means running 1.45V to the Northbridge 24/7 in order to accomplish my goals? If I fry it, fine. I'll be the second person to do so (After TJ. Hey TJ! Have you gotten the two new A10s yet?). So relax! I'll do the 23% voltage increase on air cooling, and let you know when I smell burning silicon!

 

[BeepBeep2]

Why would 1.45V be healthy for the CPU part and unhealthy for the NB/GPU part? Is it simply because it's 23% more voltage than stock? Perhaps the reason the voltage was kept that low is for power consumption reasons, not longevity reasons. Don't forget that the Trinity series were designed primarily as mobile chips that had max 100W TDPs. What if 1.18V just happens to be the minimum voltage to be able to run DDR3-1600 and an 800MHz GPU core? As far as my testing has gone, it IS absolutely the minimum.

Let's take a look at the general behavior of AMD CPUs and Intel CPUs from the past:

AMD:
(Phenom II X4/X6) - 45nm CPU: 1.5-1.6v CPU fine on air/water. Much above 1.35-1.4v CPU-NB, and degradation will occur in a matter of days/weeks/months. Set 1.6v CPU-NB on air and your chip will insta-pop dead.

Llano - Not too much overclocker's information on these as overclocking is a bit useless...not much gain in clockspeed, power consumption through the roof, etc.

However, Llano and Bulldozer use a very similarly structured CPU-NB:

Bulldozer - 32nm CPU: 1.4-1.5v CPU fine, degradation may occur above this on air/water. Above 1.5v is generally specified as "unsafe", such as pushing 1.6v+ into Phenom II. Benchmarking for short periods, should be fine. CPU-NB much above 1.3v will result in degradation, I've already seen 4 CPUs in experienced hands do this. No reason to run above 1.3v anyway, because scaling stops 1.25-1.3v for most CPUs and temperatures go through the roof rather quickly.

Trinity and Piledriver use essentially the same IMC, with minor tweaks similar to the change from Phenom II X4 to X6.

Intel:
Sandy Bridge 32nm CPU: VCCSA and VCCIO are to be kept under ~1.1-1.25v at all times because it is useless to go further, while some run the CPU at 1.5v.
Ivy Bridge: VCCSA and VCCIO are kept 1.2v or less most of the time, while it is fine to run 1.4-1.45v on the CPU.

Then look at 28nm/40nm and even 55nm GPUs - why are voltages such as 1.1-1.2v used, maximum ~1.3-1.4v for overclocking and 1.5-1.7v LN2 overclocking? Because the GPUs die much above that. Why would you assume the 32nm SOI GPU will take 1.45v with ease?

So, because of this information...
What logic would make you think 1.45v is safe 24/7 for the CPU-NB?

Generally, the rule of thumb too is that when you shrink a node, tolerable/useful voltages are lower too - why do you think 30nm DRAM runs at 1.35v stock vs 50nm DRAM that runs 1.5v?

 

[ivanlabrie]

I'd say Ivy chips can take 1.52v on air no prob...I've seen a few do 1.55v 24/7 (delidded on an h100 sub 70c load at 5.1ghz) with no degradation whatsoever.
Intel's max svid is listed as 1.52v same as max vcc in their spec sheets too...just sayin'

Nice info on the AMD camp, I'm a newb with those.

 

[Anonaru]

I'd say Ivy chips can take 1.52v on air no prob...I've seen a few do 1.55v 24/7 (delidded on an h100 sub 70c load at 5.1ghz) with no degradation whatsoever.
Intel's max svid is listed as 1.52v same as max vcc in their spec sheets too...just sayin'

Nice info on the AMD camp, I'm a newb with those.

I'd say 1.55 is pushing it. I know people can run it, but jeez there has to be a lot of electromitigation going on under there T_T

Shoot, I thought 1.390 was high on my 3930k-- I've seen people running signifcantly higher than this haha

 

[ivanlabrie]

I'd do 1.52v without breaking a sweat...delidded at least.

 

[BeepBeep2]

I'd do 1.52v without breaking a sweat...delidded at least.

Delidding is a little bit of a risk - if you have a bad mount you can kiss your CPU good bye because of hotspots on the die, hence the name heatspreader.
Overusing thermal paste is probably a good idea too if delidding.

Over at XtremeSystems, multiple tens of users have contributed to a Sandy Bridge thread talking about degradation, the general consensus over there is that to be on the safe side, use 1.4-1.45v as maximum for Sandy. A few users are running 1.5v+ with no issue at all doing crazy things, while others were using 1.45v and killing their CPUs, some are less tolerable than others.

Personally I have used up to 1.45v on the stock cooler (90-95c 2C/2T) and 1.55v with a Corsair H70 for benching, but it's really unnecessary and bad for longevity I think.

 

[visiblegorilla]

Nice info on the AMD camp...

Speaking of great info, here's an AMD white paper provided by RGone:

 

[ivanlabrie]

Interesting that they provide such numbers...wish Intel did the same but I doubt they really test beyond stock turbo clocks at all.

@BeepBeep2: Bad ihs mount is really unlikely if you remove the black glue which keeps the ihs far from the die, so no problem with that.
Just gotta have good technique after busting some pentium 4 (478) open with a razor.
And loads of tim is no good, better use a drop of liquid metal paste and spread it on the die, cause it behaves like mercury, so it won't creat air pockets. Perhaps add a thin layer on the ihs on both sides and spread it too.

 

[visiblegorilla]

Can you get lower temps on an air cooler by delidding one of these Trinity chips? Is it worth the risk compared to spending the extra money on a watercooled setup?

 

[ivanlabrie]

I doubt you can delid them, they're probably soldered like bulldozer/vishera.

 

[visiblegorilla]

What about using a grinder until you can just see silicon thru the aluminum? That's basically how it's done?

 

[Anonaru]

What about using a grinder until you can just see silicon thru the aluminum? That's basically how it's done?

At this point I'd probably just lap it over trying to grind the lid off x_x

 

[BeepBeep2]

It is possible to delid them if soldered, but you have to be really careful and will most likely end up in disaster even if you are successful.

To be honest, don't bother. AMD is using a new technique since late 2011 to lid the CPUs, the rubber adhesive/sealant is now offset 90 deg from how it used to be, and it is thinner making it a TON harder to delid them with a blade. Between burning yourself, trying to accurately control temperatures and no gain in temperatures, it's pointless.

Plus if you have a cooler that uses the stock bracket or has a fixed pressure (ie. anything but an LN2 pot or a waterblock), there will be an issue because the die is lower than the IHS...even with waterblocks, you can't get the pressure you need because the screws going through the backplate will not have room behind the motherboard.

Depending on how SMD's are sitting on top of the PCB too, it's easy to just crack one off without noticing until you open it completely...

 

[ivanlabrie]

Yeah, not worth it at all...on amd at least.

Any new updates on your tiny beast man?