X99 and XMP VCCSA voltages

[Schoolofmonkey]

Now I'm a little confused about acceptable VCCSA voltage on a 5820k.

The guys over at another overclocking forum says a lower VCCSA is better and less stress on the processors IMC.

But if I use the XMP setting on my ram it will set the VCCSA to 1.25v, from what I was told the default IMC voltage is 0.850v, and it's not designed for higher voltages (higher than 2133mhz ram).

So under that advice I lowed the VCCSA to the most stable voltage I could (1.050v) but I had to increase my ram voltage from 1.2v to 1.3v to maintain stability. (G-Skill 2666Mhz DDR4).

Is this correct, will my CPU last longer and less risk of dying with a lower VCCSA, or is the Ram's XMP settings safe.
I'm not overclocking my ram at all, but I do have a respectable 4.3Ghz@1.19v overclock on the CPU (Cache is 3.3Ghz@1.060v due to not having a OC socketed board).

Is this true, I'm getting a lot of conflicting answers from other people.

[Gofspar]

dont fucking touch VCCSA..

 

Let XMP do the work

[Schoolofmonkey]

dont fucking touch VCCSA..

 

Let XMP do the work

 

See some of the big wigs over at Overclock.net say different, so I'm confused, I never touched the VCCSA on my 4790k and I overclocked that to 4.5Ghz with my ram using XMP..

[runit3]

Now I'm a little confused about acceptable VCCSA voltage on a 5820k.

The guys over at another overclocking forum says a lower VCCSA is better and less stress on the processors IMC.

But if I use the XMP setting on my ram it will set the VCCSA to 1.25v, from what I was told the default IMC voltage is 0.850v, and it's not designed for higher voltages (higher than 2133mhz ram).

So under that advice I lowed the VCCSA to the most stable voltage I could (1.050v) but I had to increase my ram voltage from 1.2v to 1.3v to maintain stability. (G-Skill 2666Mhz DDR4).

Is this correct, will my CPU last longer and less risk of dying with a lower VCCSA, or is the Ram's XMP settings safe.

I'm not overclocking my ram at all, but I do have a respectable 4.3Ghz@1.19v overclock on the CPU (Cache is 3.3Ghz@1.060v due to not having a OC socketed board).

Is this true, I'm getting a lot of conflicting answers from other people.

What RAM kit are you using (model number)? I've got 3000MHz G.Skill's and they (or any other RAM kit) has never touched Sys Agent, despite forcing a 125 strap.

 

Whatever the people on OCN were saying about VCCSA wasn't exactly wrong, it is certainly more stressful on the components that it controls, but that can be said about every voltage setting. I would guess the instability they site would be coming from the PCIe portion of the argument and not necessarily the IMC. I don't even think my board color flags System Agent until it is over 1.3v (I can double check tomorrow).

 

I don't know who told you that the IMC wasn't designed for anything higher than .85v, but that's completely ridiculous. OTB mine was running at 1.05v, currently running at 1.165v, and had it all the way up to 1.25v when I was RAM troubleshooting. I have yet to see a serious OC running at anything less than 1.1v for VCCSA on X99.

 

Bottom line, you're going to have to bump VCCSA, RAM, or both voltages to get stability if you're trying to hit 2666. The best advice would be to enable XMP and forget about it as long as the system is stable. If you wanted to get an absolute answer regarding the appropriate VCCSA the best people to ask would be G.Skill, as they vetted the profiles for their kits.

[Schoolofmonkey]

What RAM kit are you using (model number)? I've got 3000MHz G.Skill's and they (or any other RAM kit) has never touched Sys Agent, despite forcing a 125 strap.

 

 

Ram kist I'm using is G.Skill Ripjaws 4 F4-2666C15Q-16GRR, this does push up the system agent voltage to a offset of +0.36v (1.21v).

Being 2666Mhz it still runs on the default 100 strap.

 

My system is stable with a VCCSA of 1.05v, Ram voltage of 1.3v (over the XMP's 1.2v), still using the XMP's ram timings but set the command rate to 1T.

Passes 3 hours of linux's stressapptest, 2 hour Realbench, 2000% HCI Mem Test, AIDA64 (for cache and CPU), 10 passes of x265 (2x Overkill, Pmode, 64, 4k, Normal process).

I'd consider than pretty stable, and have been using the settings for 2 days now.

 

Couple of the guys mentioned the default IMC voltage being 0.85v, anything over can degrade the chip, and XMP over compensate with higher VCCSA voltages.

 

I have noticed more with X99 overclocking that if you mess with one voltage you're messing with them all, it's more in depth that when I overclocked my 4790k..lol

 

 

I've just seen so many mixed opinions over there on VCSSA settings I thought I'd ask outside the "loop".

[mao91]

 

 

I have noticed more with X99 overclocking that if you mess with one voltage you're messing with them all, it's more in depth that when I overclocked my 4790k..lol

 

 

And I chose X99 for my first build (and, consequently, to learn overclocking on). 

 

OT: Following this thread.

[runit3]

Ram kist I'm using is G.Skill Ripjaws 4 F4-2666C15Q-16GRR, this does push up the system agent voltage to a offset of +0.36v (1.21v).

Being 2666Mhz it still runs on the default 100 strap.

 

My system is stable with a VCCSA of 1.05v, Ram voltage of 1.3v (over the XMP's 1.2v), still using the XMP's ram timings but set the command rate to 1T.

Passes 3 hours of linux's stressapptest, 2 hour Realbench, 2000% HCI Mem Test, AIDA64 (for cache and CPU), 10 passes of x265 (2x Overkill, Pmode, 64, 4k, Normal process).

I'd consider than pretty stable, and have been using the settings for 2 days now.

 

Couple of the guys mentioned the default IMC voltage being 0.85v, anything over can degrade the chip, and XMP over compensate with higher VCCSA voltages.

 

I have noticed more with X99 overclocking that if you mess with one voltage you're messing with them all, it's more in depth that when I overclocked my 4790k..lol

 

 

I've just seen so many mixed opinions over there on VCSSA settings I thought I'd ask outside the "loop".

lol your issues center around 1T vs 2T. Unless you're trying to crack some benchmark scores, command rate is going to do nothing for your RW performance. One sure way to crash out any kit is to tighten up to 1T if it's rated for 2T, you're asking to the kit to drop from two first access clock cycles down to one. You will have to increase voltage or slack out the primary timings, no way around that.

 

VCCSA was an afterthought from Sandy on because dual channel and speeds hovering around 2400MHz max didn't require any additional VCCSA voltage, same was true for Haswell Z97(the exceptions would be the people chasing world records for RAM speed).  In the pre-Sandy days when the two settings were coupled as QPI/VTT it mattered because nearly everyone was BCLK OCing (if you think X99 has a metric shit ton of voltage settings check out a screen grab from an X58 BIOS). X99 is a different beast altogether, VCCSA and VCCIO actually matter again in terms of OCing with much higher speeds for memory and quad channel support.

 

IMO anything up to 1.2-1.25v VCCSA is perfectly safe for any X99 chip. I find it bizarre that G.Skill enabled a 1.21v for a 2666 kit, but there would be no harm in using it. If you've found a stable 1T OC on the RAM, VCCSA is under 1.4v, RAM is under 1.5v, congrats you did well, and have nothing to worry about

[Schoolofmonkey]

IMO anything up to 1.2-1.25v VCCSA is perfectly safe for any X99 chip. I find it bizarre that G.Skill enabled a 1.21v for a 2666 kit, but there would be no harm in using it. If you've found a stable 1T OC on the RAM, VCCSA is under 1.4v, RAM is under 1.5v, congrats you did well, and have nothing to worry about

 

Right, that's the best insight on VCCSA I've had anywhere, thanks heaps for that 

 

It doesn't seem to matter if I run on 1T or 2T, system is happy with 1T until I try clock speeds over 4.3Ghz, then things get sporadic.

I can get the system stable with a 1.050v VCCSA, but ram voltages need to be increased.

 

It surprised me and the guys from OC.net that the XMP is setting it so high, going off HWMonitor it was a offset of 0.36v, it's 0.51v if I set the BCLK to 100.5Mhz..lol..

I've never messed with the VCCIO voltages, I did ask if I needed to, was told there's no need to and just to leave it at the default.

 

I'm actually enjoying tinkering with the X99 platform, I've learnt a heck of a lot more than I knew before overclocking mainstream stuff.

Trying to push the CPU higher, but I think I lost the silicone lottery, I need 1.25v for 4.4Ghz and 1.29v for 4.5Ghz...  :rolleyes:

[runit3]

Right, that's the best insight on VCCSA I've had anywhere, thanks heaps for that 

 

It doesn't seem to matter if I run on 1T or 2T, system is happy with 1T until I try clock speeds over 4.3Ghz, then things get sporadic.

I can get the system stable with a 1.050v VCCSA, but ram voltages need to be increased.

 

It surprised me and the guys from OC.net that the XMP is setting it so high, going off HWMonitor it was a offset of 0.36v, it's 0.51v if I set the BCLK to 100.5Mhz..lol..

I've never messed with the VCCIO voltages, I did ask if I needed to, was told there's no need to and just to leave it at the default.

 

I'm actually enjoying tinkering with the X99 platform, I've learnt a heck of a lot more than I knew before overclocking mainstream stuff.

Trying to push the CPU higher, but I think I lost the silicone lottery, I need 1.25v for 4.4Ghz and 1.29v for 4.5Ghz...  :rolleyes:

As you increase the number of clock cycles per second it will correspond with an increase in calls on the RAM, if it's set to 1T it's ramped to once per cycle, instability is pretty expected. I've never been able to get mine stable at 1T w/out some pretty drastic increases in all voltages, and it just isn't worth it for chasing benchmarks.

 

VCCIO is pretty unnecessary unless you're teetering on stability and don't/can't touch core voltage or Input voltage, I've always bumped mine a bit. Seems to have minimal impact on temps.

 

1.25v and 1.29v for 4.4/.5 is pretty damn good. I had to crank mine to 1.45v at 4.5. Just got the new 5960X from Intel, we'll see how this one lands in the lotto.

[Schoolofmonkey]

1.25v and 1.29v for 4.4/.5 is pretty damn good. I had to crank mine to 1.45v at 4.5. Just got the new 5960X from Intel, we'll see how this one lands in the lotto.

 

Everyone keeps saying that, but then you see the same guys chasing the J batch which is claimed to do 4.4Ghz on 1.2 volts.  :wacko:

Personally I'm fine with 4.3Ghz, I haven't actually noticed any day to day different betwen that and 4.5Ghz other than extra heat.

Yes it shows up in Cinebench but we aren't running Cinebench all the time..

 

I did drop the ram back to 2T, I still needs 1.3v with the VCCSA at 1.050v, though I am using a MSI x99a Gaming 7 which doesn't have a OC socket.

 

I've never had a BSOD while dialling in my OC, it's always been hard locks or hard reboots (oc failed message), the Gaming 7 won't throw a BSOD, generally the rule of thumb is a hard lock is RAM or cache (been buying Asus boards for years, had none in stock when I built my x99 rig), not with a MSI board, they don't even have LLC settings, its controlled by the vdroop settings which are all in %, found 95% to be the most stable, allowing for about 30mv drop, though again I keep getting told a 50mv drop is recommended, but in doing that I create instabilities.

But again I things are so different on my MSI board compared to their Asus boards, so you have to take the info and turn it into MSI speak..  :lol:

 

Update:

Went back to the XMP profile to show you what voltages it sets, more stable using the XMP at 4.4Ghz than manual settings.