1366 X58 Xeon 5650

DragonQ · 14 Feb 2015 at 21:29

Ugh, this Asus throttling makes overclocking so difficult. With a BLCK limit somewhere at 200 MHz, the only way to get over 4 GHz is to use the 22x multiplier, but that means it will also go to 23x if only a few cores are being used. 200x23 = 4.6 GHz and that's a big jump from 4.0 GHz.

So 20x200 seems stable on my rig now, but if I just bump the multiplier to 22x and use Task Manager to make IBT only use 4 cores, it insta-crashes (even at VCore = 1.35 V). So I bumped down the BLCK to 191 and did the same thing and (so far) it seems stable. The 4 cores being used for IBT ramp up to 23x (well ~22.7 on average, I assume it's constantly switching between 22x and 23x) and the rest sit at 22x.

If I can get 191x22 stable (looks good so far) then it'll run at 3.8 GHz when really stressed (Asus throttling to 20x) but otherwise 4.2 Hz when using lots of cores or 4.4 GHz when using a few. The question is whether it's better to choose 200x20 for a flat 4.0 GHz or 191x22 for a 3.8 GHz / 4.2 GHz / 4.4 GHz combo depending on load. :rolleyes:

Maybe I can get it to work at 195 MHz (i.e. 3.9 GHz / 4.3 GHz / 4.5 GHz) but we'll see.

kylew_is_back · 15 Feb 2015 at 14:59

DragonQ@

Why not just set the multiplier to 20x and BLCK 215MHz with uncore below 3GHz, and ram multiplier 6x and qpi 18x, this way it will be also easier to get it stable, since you only have to test one multiplier.

DragonQ · 15 Feb 2015 at 15:10

kylew_is_back said:
DragonQ@

Why not just set the multiplier to 20x and BLCK 215MHz with uncore below 3GHz, and ram multiplier 6x and qpi 18x, this way it will be also easier to get it stable, since you only have to test one multiplier.

210 MHz isn't stable so 215 MHz won't be.

kylew_is_back · 15 Feb 2015 at 15:20

DragonQ said:
210 MHz isn't stable so 215 MHz won't be.

of course it won't be stable unless you lower the uncore, ram multiplier and qpi.

DragonQ · 15 Feb 2015 at 16:05

kylew_is_back said:
of course it won't be stable unless you lower the uncore, ram multiplier and qpi.

I'm already using:

- Highest RAM multiplier that is <=1600 MHz
- Lowest QPI multiplier
- 2x DRAM speed for Uncore

There is no 18x QPI multiplier on my motherboard, 36x is the lowest.

kylew_is_back · 15 Feb 2015 at 16:40

DragonQ said:
I'm already using:

- Highest RAM multiplier that is <=1600 MHz
- Lowest QPI multiplier
- 2x DRAM speed for Uncore

There is no 18x QPI multiplier on my motherboard, 36x is the lowest.

Set your ram at 1290MHz with Uncore 2580MHz and see if it is stable, set VTT 1.35v.

As for the QPI try not to go above 4GHz, with 18x215 will give you 3870MHz QPILink.

DragonQ · 15 Feb 2015 at 17:02

kylew_is_back said:
Set your ram at 1290MHz with Uncore 2580MHz and see if it is stable, set VTT 1.35v.

As for the QPI try not to go above 4GHz, with 18x215 will give you 3870MHz QPILink.

You clearly aren't listening to me. This is not stable, even at 1.35 V:

BLCK = 210 MHz
DRAM = 1260 MHz (lowest available)
Uncore = 2520 MHz
QPI = 7.56 GT/s (lowest available)

DragonQ · 15 Feb 2015 at 23:26

OK now I'm starting to think my whole setup is just buggered. Even if I can get IBT stable, games are not. Even with VCore = 1.35 V and VTT = 1.35 V using BCLK = 182 MHz, which should be super-easy. For example, over the past few hours TF2 has crashed half a dozen times. Every time the same thing happens: all of the players and most of the objects turn invisible, then the application just closes with no explanation. I've tried updating GPU drivers, checking file integrity, etc.

I have no idea what would cause such an issue - I suppose disabling the overclock completely and running a RAM test would be a good next step.

P0tenc · 16 Feb 2015 at 10:18

DragonQ said:
I have no idea what would cause such an issue - I suppose disabling the overclock completely and running a RAM test would be a good next step.

Everything stock and run the game again also to rule out some driver bug or other issue. What PSU are you using?

P0tenc · 16 Feb 2015 at 10:22

On another note I have stepped back my previous OC a little as 4.4Ghz was not completely stable at 1.375v and I am not willing to push more volts for a tiny bump in speed.

4.3Ghz is stable at 1.375v but I have now dropped to 4.2Ghz (20x 210 turbo disabled) which seems to only need 1.33v. Also dropped the QPI to only 1.22v.

Still testing

I really miss offset voltage and adaptive voltage

Stu · 16 Feb 2015 at 10:22

DragonQ said:
OK now I'm starting to think my whole setup is just buggered. Even if I can get IBT stable, games are not. Even with VCore = 1.35 V and VTT = 1.35 V using BCLK = 182 MHz, which should be super-easy. For example, over the past few hours TF2 has crashed half a dozen times. Every time the same thing happens: all of the players and most of the objects turn invisible, then the application just closes with no explanation. I've tried updating GPU drivers, checking file integrity, etc.

I have no idea what would cause such an issue - I suppose disabling the overclock completely and running a RAM test would be a good next step.

Very frustrating! I would also expect 1.35v on vcore and VTT to be enough for a moderate overclock.

Let's start thinking about other options? Have you got the latest BIOS? Have you disabled CPU Spread Spectrum? Make sure Turbo is turned off (at least for now). Try increasing uncore by one notch (i.e. the next number down after 2xDRAM). For CPU try odd number multipliers (e.g. many say x19 and x21 can be more stable than x20. It might be worth playing around with the CPU Differential Amplitude (800mV) and the Clock Skew (try Delay 100ps on both)... I got some extra stability on my 920 using these settings, but, to be honest, they shouldn't be needed for the settings you are using right now. Check the RAM timings - when set to AUTO, my board seems to change the timings on its own when I make changes to BCLK.

P0tenc · 16 Feb 2015 at 11:34

Has anyone who replaced an i7 chip with a x56xx tried to use ECC ram on it? I don't think anyone has tried this.

I am pondering buying a dirt cheap single ECC module to just try it. The x56xx IMC does support it, but the mainboard officially doesn't (it doesn't support the CPU officially either). I am just wondering if the mainboard needs to support it in any way since the memory controller is on the CPU.

Stu · 16 Feb 2015 at 12:06

P0tenc said:
Has anyone who replaced an i7 chip with a x56xx tried to use ECC ram on it? I don't think anyone has tried this.

I am pondering buying a dirt cheap single ECC module to just try it. The x56xx IMC does support it, but the mainboard officially doesn't (it doesn't support the CPU officially either). I am just wondering if the mainboard needs to support it in any way since the memory controller is on the CPU.

Why would you want to use ECC for a home PC?

P0tenc · 16 Feb 2015 at 12:08

I might be able to get it very cheap.

Stu · 16 Feb 2015 at 12:42

Is it unbuffered or registered. If registered, then I very much doubt it will work; you might get lucky with unbuffered. Also, some ECC memory is 1.35v, which your mobo may not like.

However, I can't find a definitive answer for you, but I'm not sure it is worth the risk. I've seen mention additional check bit traces between the memory controller and DIMM slots, which may not be included in the Rampage III... I'm not sure what happens if you load ECC into a non-compatible boards, whether they don't work at all, or they perform like non-ECC - there is mixed opinion around the web, and it appears to depend on the motherboard.

These threads are useful reading:
http://hardforum.com/showthread.php?t=1624480
http://forums.evga.com/ECC-RAM-Question-m615228.aspx

Goody_GTI · 16 Feb 2015 at 12:44

DragonQ said:
OK now I'm starting to think my whole setup is just buggered. Even if I can get IBT stable, games are not. Even with VCore = 1.35 V and VTT = 1.35 V using BCLK = 182 MHz, which should be super-easy. For example, over the past few hours TF2 has crashed half a dozen times. Every time the same thing happens: all of the players and most of the objects turn invisible, then the application just closes with no explanation. I've tried updating GPU drivers, checking file integrity, etc.

I have no idea what would cause such an issue - I suppose disabling the overclock completely and running a RAM test would be a good next step.

Have you run any other stability tests instead if IBT ?

P0tenc · 16 Feb 2015 at 13:02

Stu said:
Is it unbuffered or registered. If registered, then I very much doubt it will work; you might get lucky with unbuffered. Also, some ECC memory is 1.35v, which your mobo may not like.

However, I can't find a definitive answer for you, but I'm not sure it is worth the risk. I've seen mention additional check bit traces between the memory controller and DIMM slots, which may not be included in the Rampage III... I'm not sure what happens if you load ECC into a non-compatible boards, whether they don't work at all, or they perform like non-ECC - there is mixed opinion around the web, and it appears to depend on the motherboard.

These threads are useful reading:
http://hardforum.com/showthread.php?t=1624480
http://forums.evga.com/ECC-RAM-Question-m615228.aspx

Thanks, I forgot about the voltage, too many unknowns. Think I will skip.

DragonQ · 16 Feb 2015 at 13:35

P0tenc said:
Everything stock and run the game again also to rule out some driver bug or other issue. What PSU are you using?

Stu said:
Very frustrating! I would also expect 1.35v on vcore and VTT to be enough for a moderate overclock.

Let's start thinking about other options? Have you got the latest BIOS? Have you disabled CPU Spread Spectrum? Make sure Turbo is turned off (at least for now). Try increasing uncore by one notch (i.e. the next number down after 2xDRAM). For CPU try odd number multipliers (e.g. many say x19 and x21 can be more stable than x20. It might be worth playing around with the CPU Differential Amplitude (800mV) and the Clock Skew (try Delay 100ps on both)... I got some extra stability on my 920 using these settings, but, to be honest, they shouldn't be needed for the settings you are using right now. Check the RAM timings - when set to AUTO, my board seems to change the timings on its own when I make changes to BCLK.

PSU is 700 W, easily enough. RAM timings are manually set to 8-8-8-24, which is what the sticks are rated at. DRAM voltage is 1.64 V.

I ran memtest overnight and everything passed, then I restored defaults in the BIOS. I will play the game again tonight and see if I get the same issue. I've ruled out dodgy GPU drivers and will hopefully rule out an unstable overclock tonight. That'd still leave many things: faulty CPU, faulty motherboard, faulty GPU, application bug...

I'm hoping it's a bug in the game because no other applications seem to be having issues (including IBT and Prime95) - there are forum posts about exactly what I'm experiencing from 2013 which ended up in Valve releasing a patch. However, I can't see any recent posts about it. I will post in the Valve forums if I continue to experience the issue at stock speeds.

Combat Fighter · 16 Feb 2015 at 14:19

Nice to see this thread still rocking. . .

Sold my X5675 over the weekend. Was a good chip for the money.

X58 was fun and wont be forgotten.

My X99 does rock though!

Ricimer · 16 Feb 2015 at 14:23

Memtest is rubbish for finding RAM issues, what timing does it pick on automatic? I have turbo and EIST disabled on a P6TD. You may need to increase PCI-E and IOH voltage a tiny bit too. How many RAM modules?