Overclock.net › Forums › Intel › Intel CPUs › Need Help with 2500K 4.7 GHz - Failing large FFTs
New Posts  All Forums:Forum Nav:

Need Help with 2500K 4.7 GHz - Failing large FFTs

post #1 of 10
Thread Starter 
Setup

MB - ASROCK P67 Extreme4 (bios 1.60)
CPU - i5-2500K
RAM - GSkill 4x4GB 1333 (2x F3-10666CL9D-8GBRL packs)
Heatsink - Scyth Mugen 2 Rev B

Side note : I upgraded to this from an E6750, which I spent a lot of time working on a 3.8GHz OC, so that's where my "recent" OC experience is from . . .


Here's the problem I am having when testing with Prime95.

I can run a "small in place" test a remain stable at 4.8GHz. This is the toughest test on the CPU as it keeps all the data in on chip cache. My max temperatures hit 80*C on the hottest core (I'm ok with this number for now The Mugen 2 is a PITA to mount so I'll try remonting it when I have a friend to help with an extra set of hands. Either way, this is max temp on the cores and the CPU is stable . . . no problems )

When I try to run a large FFT with RAM usage I get errors pretty quickly. CPU temperature in this mode tops out at about 70*C.
Sometimes I will get a BSOD, and sometimes a single Prime 95 thread will get an error at which point that thread stops, the temperature drops, and the other 3 threads continue with no problems.


I have tried a couple of things to verify operation of the RAM.
*At 4.5 GHz I don't get this problem so it's not the RAM.
*At 4.7 Ghz I have tried increasing the timing settings and even droping the clock (without decreasing the timing settings) down to 1066, but I still have the exact same problems.

This leads me to believe that there is a problem with the memory controller or the interface between the CPU cores and the memory controller.

I've tried enabling higher voltage PPL (not sure why not would be needed to fix this problem as I am not having a problem with the core frequency)

I've tried adjusting each of the following voltages ~ +10% independently add as a group from the 4.7GHz "small FFT stable" setting
CPU Core Voltage
PCH Voltage
CPU PLL Voltage
VTT Voltage
VCCSA Voltage

However, I still get the same fault (BSOD or failed prime95 thread) on a large FFT test.


1) Does anyone that knows more about the P67 chipset than I do have any guesses about what could be failing here? (such information would point to the correct solution)
I can tell you it's
NOT the CPU cores
NOT the physical RAM
NOT that the Memory BUS.

2) Should I try moving BCLK down to say 94MHz and bump the multiplier up to 50 (this will maintain 4.7GHz) but at lower clocks on everything outside of the core.

3) The last though I have is the bump of the Vcore at 4.6GHz a bit to give the same power output on the core. This should let me know if its just a temperature problem.

4) Any other bright ideas or should I just thank intel for the 4.6GHz chip and optomize there?


-dan
post #2 of 10
What BSOD are you getting? Most likely if you are just getting failed workers you need more voltage for your chip. - once you start moving past a certain point, you need much more voltage than ~10% - try bumping it several notches, you shouldn't need to bump the other voltages so just set them to Auto
The Laptop
(6 items)
 
  
CPUGraphicsRAMHard Drive
Intel i7 2860QM AMD Radeon HD 6770M Corsair Vengence 1600MHz Samsung 840 Pro 256GB 
OSMonitor
Mac OS X 10.8 1680 x 1050 anti-glare high resolution  
  hide details  
Reply
The Laptop
(6 items)
 
  
CPUGraphicsRAMHard Drive
Intel i7 2860QM AMD Radeon HD 6770M Corsair Vengence 1600MHz Samsung 840 Pro 256GB 
OSMonitor
Mac OS X 10.8 1680 x 1050 anti-glare high resolution  
  hide details  
Reply
post #3 of 10
Thread Starter 
Why would I need more v-core?


Its stable at 4.7 GHz with small FFTs (much harder on the core than large FFTs because the core spends a lot of time shutdown waiting for data from ram during the large FFT test).

But . . . I did try increasing the core by 10% (that's right a whole .15V) and it make no difference.


-dan
post #4 of 10
Quote:
Originally Posted by dparkermit View Post
Why would I need more v-core?


Its stable at 4.7 GHz with small FFTs (much harder on the core than large FFTs because the core spends a lot of time shutdown waiting for data from ram during the large FFT test).

But . . . I did try increasing the core by 10% (that's right a whole .15V) and it make no difference.


-dan
What psu do you have?

Do you have llc voltage enabled?, also short and long duration set @ around 200-250

Core current limit 200 ?

Your voltage may be fine but your cpu is not getting enough current
Chewy's Chomper
(14 items)
 
  
CPUMotherboardGraphicsRAM
Intel I7 6700k Asis Z170 Deluxe MSI GTX 1080 Gaming x 16GB Corsair vengence DDR4 3200 
Hard DriveCoolingOSMonitor
Samsung M2 256gb nvme Custom loop, 360+120 rads Windows 10 64bit Asus 28" MG28UQ 4K 
PowerCase
Seasonic X series 750w gold Corsair 400c 
  hide details  
Reply
Chewy's Chomper
(14 items)
 
  
CPUMotherboardGraphicsRAM
Intel I7 6700k Asis Z170 Deluxe MSI GTX 1080 Gaming x 16GB Corsair vengence DDR4 3200 
Hard DriveCoolingOSMonitor
Samsung M2 256gb nvme Custom loop, 360+120 rads Windows 10 64bit Asus 28" MG28UQ 4K 
PowerCase
Seasonic X series 750w gold Corsair 400c 
  hide details  
Reply
post #5 of 10
Thread Starter 
Quote:
Originally Posted by Chewy View Post
What psu do you have?

Do you have llc voltage enabled?, also short and long duration set @ around 200-250

Core current limit 200 ?

Your voltage may be fine but your cpu is not getting enough current
Yes LLC is enabled and properly compensated

I'll try increasing the power limit when I get home, but I don't see what good that would do. The system is running just fine in a higher power mode. Its failing in a lower power mode that indicates that the problem is not in the the cores but in one of the interfaces.

-dan
post #6 of 10
You *need* more vcore, no ands, ifs ors or buts.
There's more to a CPU than just the cores and cache now. The cpu's have an IMC, so it's the exact opposite that you saw on the Core 2 chips. Back then, you could pass large FFT/Blend all day, but have small FFT fail, because the northbridge is on the motherboard; the CPU thus had alot less work to do.

The CPU has to use its decode engines when accessing RAM, not just the CPU cache, and that puts more stress on the entire CPU. Run Prime blend...you'll get lower temps and most likely the same crashes.

Hyperthreading puts even MORE of a stress on those engines than your large FFT. I can do 5 ghz 1.39v with HT disabled just fine, but enable HT and I need 1.43v...
SB Rig
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7-2600k Gigabyte P67A-UD5 B3 R9 290X 16GB 
Hard DriveOptical DriveOSMonitor
WD 2 TB, WD 1 TB, 250gb Liteon ihbs212 blu-ray burner/reader XP+W7 Benq XL2720Z 
KeyboardPowerCaseMouse
Ducky Shine 69/Year of the Goat/Shine 5 Seasonic Platinum X1000 Corsair 760T Logitech G502 with r0ach approved ™ sensor 
Mouse Pad
Puretrak Talent 
  hide details  
Reply
SB Rig
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7-2600k Gigabyte P67A-UD5 B3 R9 290X 16GB 
Hard DriveOptical DriveOSMonitor
WD 2 TB, WD 1 TB, 250gb Liteon ihbs212 blu-ray burner/reader XP+W7 Benq XL2720Z 
KeyboardPowerCaseMouse
Ducky Shine 69/Year of the Goat/Shine 5 Seasonic Platinum X1000 Corsair 760T Logitech G502 with r0ach approved ™ sensor 
Mouse Pad
Puretrak Talent 
  hide details  
Reply
post #7 of 10
Thread Starter 
Quote:
Originally Posted by Falkentyne View Post
You *need* more vcore, no ands, ifs ors or buts.
There's more to a CPU than just the cores and cache now. The cpu's have an IMC, so it's the exact opposite that you saw on the Core 2 chips. Back then, you could pass large FFT/Blend all day, but have small FFT fail, because the northbridge is on the motherboard; the CPU thus had alot less work to do.

The CPU has to use its decode engines when accessing RAM, not just the CPU cache, and that puts more stress on the entire CPU. Run Prime blend...you'll get lower temps and most likely the same crashes.

Hyperthreading puts even MORE of a stress on those engines than your large FFT. I can do 5 ghz 1.39v with HT disabled just fine, but enable HT and I need 1.43v...
Thanks Falkentyne - I appreciate you intelligent and thorough response.

I can try raising Vcore more, but after going from 1.36 -> 1.50 had no effect, I didn't think bumping it more would do any good. My assumption was that it must have been something related to the memory controller, ring bus, or "system agent".

I tried raising the Memory Controller and "system agent" voltage ~15% already, but again that had no effect.

If vcore is the only way to stabilize at 4.7 I'm just never going to get there. Can't tolerate 20-25% more power with my current temps for 1 bin gain. Well I suppose I could, but it would be pushing 90* at that point under Core Load.

Fortunately I don't have to worry about HT issues on the 2500K
post #8 of 10
try using blend test, it's been recommended as the best tester for sandybridge.
it looks alive
(17 items)
 
Sandy Transformed
(69 photos)
 
CPUMotherboardGraphicsRAM
i5 2500K [5ghz 1.44v 24/7] Asus Maximus IV Extreme SLI GTX 460 @ 920/1840/4100 [1.11v] 4GB Dominator GT @ 2133mhz CL8 [1.65v] 
Hard DriveHard DriveCoolingOS
Intel X25-M G2 80GB 3x Samsung F3 500GB RAID 0 EK Supreme HF Full Copper Windows 7 Ultimate 64bit 
MonitorPowerCaseAudio
LG42SL9500 42" 1080p [Surround Soon] XFX XXX Edition 650w Xigmatec Elysium JVC HA-RX700 [foam mod] 
  hide details  
Reply
it looks alive
(17 items)
 
Sandy Transformed
(69 photos)
 
CPUMotherboardGraphicsRAM
i5 2500K [5ghz 1.44v 24/7] Asus Maximus IV Extreme SLI GTX 460 @ 920/1840/4100 [1.11v] 4GB Dominator GT @ 2133mhz CL8 [1.65v] 
Hard DriveHard DriveCoolingOS
Intel X25-M G2 80GB 3x Samsung F3 500GB RAID 0 EK Supreme HF Full Copper Windows 7 Ultimate 64bit 
MonitorPowerCaseAudio
LG42SL9500 42" 1080p [Surround Soon] XFX XXX Edition 650w Xigmatec Elysium JVC HA-RX700 [foam mod] 
  hide details  
Reply
post #9 of 10
Without knowing the BSOD code your getting it is hard to say what your problem is... 9 times out of 10 yes it is the Vcore

BSOD Codes
0x124 = add/remove vcore or QPI/VTT voltage (usually Vcore, once it was QPI/VTT)
0x101 = add more vcore
0x50 = RAM timings/Frequency add DDR3 voltage or add QPI/VTT
0x1E = add more vcore
0x3B = add more vcore
0xD1 = add QPI/VTT voltage
“0x9C = QPI/VTT most likely, but increasing vcore has helped in some instances”
0X109 = add DDR3 voltage
0x0A = add QPI/VTT voltage



EDIT
And yes with Sandy you need to be using Blend FFT
The Raven
(16 items)
 
  
CPUMotherboardGraphicsGraphics
i7-2600K Gigabyte GA-P67A-UD5-B3 EVGA GTX 570 SC EVGA GTX 570 SC 
RAMHard DriveOptical DriveCooling
16GB G.SKILL Ripjaws X 1866 Samsung 840 Pro  iHAS324 - Lite-On DVD-RW Noctua NH-D14 
OSMonitorMonitorKeyboard
Windows 10 ASUS VN248 ASUS VN248 Logitech G510 
PowerCaseMouse
XFX 850W BE SILVERSTONE RV02B-EW Logitech MX518 
  hide details  
Reply
The Raven
(16 items)
 
  
CPUMotherboardGraphicsGraphics
i7-2600K Gigabyte GA-P67A-UD5-B3 EVGA GTX 570 SC EVGA GTX 570 SC 
RAMHard DriveOptical DriveCooling
16GB G.SKILL Ripjaws X 1866 Samsung 840 Pro  iHAS324 - Lite-On DVD-RW Noctua NH-D14 
OSMonitorMonitorKeyboard
Windows 10 ASUS VN248 ASUS VN248 Logitech G510 
PowerCaseMouse
XFX 850W BE SILVERSTONE RV02B-EW Logitech MX518 
  hide details  
Reply
post #10 of 10
Thread Starter 
Quote:
Originally Posted by munaim1 View Post
try using blend test, it's been recommended as the best tester for sandybridge.
Why would I use the blend test when I KNOW ITS NOT STABLE ALREADY???

seriously, you don't have to comment, but if you are going to comment you can at least read the post first . . .
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Intel CPUs
Overclock.net › Forums › Intel › Intel CPUs › Need Help with 2500K 4.7 GHz - Failing large FFTs