Overclock.net › Forums › Graphics Cards › NVIDIA › ~~The GTX 670 Overclocking Master-Guide~~
New Posts  All Forums:Forum Nav:

~~The GTX 670 Overclocking Master-Guide~~ - Page 46

post #451 of 3357
Thread Starter 
Quote:
Originally Posted by General123 View Post

It maxes on its own at load.

That's not true in all cases. Some 670's seem to drop down to 1.162v under full-load instead of running at 1.175v if you don't manually max out the voltage slider. There's no reason not to just max the voltage slider though even if your 670 doesn't drop to 1.162v.
post #452 of 3357
Thread Starter 
Quote:
Originally Posted by ShotgunBFFL View Post

Ok Ty
Now I'm having 2 weird things happen, when I complete the heaven benchmark the gpu section says I'm running off my hd 3000 graphics, which is weird and idk why.
It's just a bug with Heaven. Even if you have the iGPU disabled in the bios, it still sometimes shows up in Heaven. Just ignore it.
Quote:
]Also, when I got to about 90-100 offset which brings me to 1245 or 1250, I crash in heaven with the error "D3D11AppWindow::swap_window(): device removed"
Edit: just did a successful run of heaven with +85 (1235 Mhz) and +200 Memory. Score was 1334
Edit2: just did a successful run of heaven with +85(1235 Mhz) and +275 Memory. Score was 1345
Is this really the OC limit of my 670... confused.gif

That might not be your maximum. Did you try going higher on the memory? Keep increasing the memory until you get artifacts (or crash) and then back down a bit from there. If you already did this, then yes, that's probably your maximum overclock.
post #453 of 3357
Quote:
Originally Posted by SeanPoe View Post

It's just a bug with Heaven. Even if you have the iGPU disabled in the bios, it still sometimes shows up in Heaven. Just ignore it.
That might not be your maximum. Did you try going higher on the memory? Keep increasing the memory until you get artifacts (or crash) and then back down a bit from there. If you already did this, then yes, that's probably your maximum overclock.

Yea I got to 300 on memory and the same error happened, i backed down to 275 and it was fine.

I was thinking i could get this thing to like 1300 =(

Also, since i have the msi pe isnt there supposed to be a voltage unlock thing coming so that we can go higher and possibly increase overclocks?
Power Edition
(18 items)
 
   
CPUMotherboardGraphicsRAM
i7 3610QM Clevo p151EMx Intel HM77 NVidia GTX 670m 16GB 1600Mhz DDR3 Samsung 4x4 
Hard DriveOptical DriveCoolingOS
Seagate Momentus 750GB 7200 RPM 16MB Cache TSSTcorp CDDVDW SN-208AB Sager NP9130 Internal Fans Windows 7 Ultimate 64 bit 
MonitorKeyboardPowerCase
15.6” 1920x1080 LED-Backlit Display Matte Finish Sager NP9130 Keyboard 8 Cell Smart Lithium-Ion Battery 76.96Wh Sager NP9130 
MouseAudio
Touchpad or Logitech G5 Onkyo Installed Speakers 
  hide details  
Reply
Power Edition
(18 items)
 
   
CPUMotherboardGraphicsRAM
i7 3610QM Clevo p151EMx Intel HM77 NVidia GTX 670m 16GB 1600Mhz DDR3 Samsung 4x4 
Hard DriveOptical DriveCoolingOS
Seagate Momentus 750GB 7200 RPM 16MB Cache TSSTcorp CDDVDW SN-208AB Sager NP9130 Internal Fans Windows 7 Ultimate 64 bit 
MonitorKeyboardPowerCase
15.6” 1920x1080 LED-Backlit Display Matte Finish Sager NP9130 Keyboard 8 Cell Smart Lithium-Ion Battery 76.96Wh Sager NP9130 
MouseAudio
Touchpad or Logitech G5 Onkyo Installed Speakers 
  hide details  
Reply
post #454 of 3357
Thread Starter 
Quote:
Originally Posted by Arizonian View Post

Might want to try backing down on power target offset. It may keep it from crashing without having to lower clocks. Try +125 and see if that changes anything.

Do you have any examples of where this worked for other people without causing a throttle? When i tested this out, all that happened was I'd get throttled by power (which would then dynamically reduce my overclock), only making my unstable overclock stable because it was actually reducing my overclock. For example, if i was at 1300MHz with a +150MHz offset (unstable) and then reduced the power-target a bit, it would stabilize my overclock but only because i was getting throttled down to ~1250Mhz (though my offset was still set at 150). I found i could get more performance by just keeping the power target maxed and then manually reducing my overclock a bit. So instead of being at 1300MHz and 105% power target, i would set it at 1275MHz and 121% power target (this resulted in a performance increase). I have a hunch that in less-demanding applications (that don't tap out the power-draw causing a power-related throttle) that you might still be unstable without that throttle happening. So manually reducing your overclock would not only result in more performance in most cases but would also eliminate the possibility of instability when that power-related throttle doesn't happen.

This is just what I've observed with my own 670 anyways, things might be different for other people's 670s. If reducing the power target does indeed result in stability and increased performance for other people, I'll be sure to update the guide to reflect this new information. thumb.gif
post #455 of 3357
Quote:
Originally Posted by General123 View Post

It maxes on its own at load.

Was defective i rmaed with NEwegg got one of these got tired of AMDs buggy drivers.. This thing is alittle slower then my 7970 was but meh no driver headaches ftw..
    
CPUMotherboardGraphicsGraphics
i7 3770k 5.2ghz  GIGABYTE G1 Assassin Sniper 3 7950 with Alphacool block 7950 w/ Alphacool Block 
GraphicsRAMHard DriveCooling
7950 w/ Alphacool block G.SKILL Trident X 2500mhz 2x4gb SanDisk Extreme 240gb sata III 6gbs x5 SwiftTech Apongee Drive II 
CoolingCoolingCoolingOS
Monsta 360 86mm thick rad X 4 MCP 35X Pump x2 400ML Frozen Q Fusion res x3 Windows 7 Ultimate 64 bit 
MonitorKeyboardPowerCase
CatLeap 2560x1440 + Acer hn274h bmiiid 120hz 3D Logitech G19 Corsair Ax1200 watts  Mountain Mods Ascension 
MouseAudio
Logitech M570 Pioneer SP-BS21 Bookshelf Loudspeakers + Lepai Amp 
  hide details  
Reply
    
CPUMotherboardGraphicsGraphics
i7 3770k 5.2ghz  GIGABYTE G1 Assassin Sniper 3 7950 with Alphacool block 7950 w/ Alphacool Block 
GraphicsRAMHard DriveCooling
7950 w/ Alphacool block G.SKILL Trident X 2500mhz 2x4gb SanDisk Extreme 240gb sata III 6gbs x5 SwiftTech Apongee Drive II 
CoolingCoolingCoolingOS
Monsta 360 86mm thick rad X 4 MCP 35X Pump x2 400ML Frozen Q Fusion res x3 Windows 7 Ultimate 64 bit 
MonitorKeyboardPowerCase
CatLeap 2560x1440 + Acer hn274h bmiiid 120hz 3D Logitech G19 Corsair Ax1200 watts  Mountain Mods Ascension 
MouseAudio
Logitech M570 Pioneer SP-BS21 Bookshelf Loudspeakers + Lepai Amp 
  hide details  
Reply
post #456 of 3357
Quote:
Originally Posted by Arizonian View Post

Might want to try backing down on power target offset. It may keep it from crashing without having to lower clocks. Try +125 and see if that changes anything.
Let us know what clocks your getting. Offset values vary and clocks reached with Kepler Boost.
May want to set up a fan profile. 1C-1% example 50C - 50% fan speed. Mine is more aggressive benching.

All done temps have been no issue hot as i get is 62c ish.. im at 127% power +87 on core and +700 on the mem
    
CPUMotherboardGraphicsGraphics
i7 3770k 5.2ghz  GIGABYTE G1 Assassin Sniper 3 7950 with Alphacool block 7950 w/ Alphacool Block 
GraphicsRAMHard DriveCooling
7950 w/ Alphacool block G.SKILL Trident X 2500mhz 2x4gb SanDisk Extreme 240gb sata III 6gbs x5 SwiftTech Apongee Drive II 
CoolingCoolingCoolingOS
Monsta 360 86mm thick rad X 4 MCP 35X Pump x2 400ML Frozen Q Fusion res x3 Windows 7 Ultimate 64 bit 
MonitorKeyboardPowerCase
CatLeap 2560x1440 + Acer hn274h bmiiid 120hz 3D Logitech G19 Corsair Ax1200 watts  Mountain Mods Ascension 
MouseAudio
Logitech M570 Pioneer SP-BS21 Bookshelf Loudspeakers + Lepai Amp 
  hide details  
Reply
    
CPUMotherboardGraphicsGraphics
i7 3770k 5.2ghz  GIGABYTE G1 Assassin Sniper 3 7950 with Alphacool block 7950 w/ Alphacool Block 
GraphicsRAMHard DriveCooling
7950 w/ Alphacool block G.SKILL Trident X 2500mhz 2x4gb SanDisk Extreme 240gb sata III 6gbs x5 SwiftTech Apongee Drive II 
CoolingCoolingCoolingOS
Monsta 360 86mm thick rad X 4 MCP 35X Pump x2 400ML Frozen Q Fusion res x3 Windows 7 Ultimate 64 bit 
MonitorKeyboardPowerCase
CatLeap 2560x1440 + Acer hn274h bmiiid 120hz 3D Logitech G19 Corsair Ax1200 watts  Mountain Mods Ascension 
MouseAudio
Logitech M570 Pioneer SP-BS21 Bookshelf Loudspeakers + Lepai Amp 
  hide details  
Reply
post #457 of 3357
Quote:
Originally Posted by SeanPoe View Post

Also, if you swap the better performing card to the first slot in your SLI-array, then you will see a small performance boost too (less than 1%).
Yes, that's exactly what you'll need to do.
... Did you check the graphs to see if you're indeed being throttled?

First slot assuming the lane closest to the CPU?

What would it look like on the graph if I was throttled by the power at 100-115% when power slider is maxed at 145%? I merely suggested that because the temps were fine but it was unstable.

BTW thanks for suggesting me to loop heaven for 2 hours to test for stability...I thought about it but it sound like it defeated the purpose of your 30 minute guide biggrin.gif

So If I should test for stability for 2 hours...what's to say that let's say I follow the quick guide, find a close to max GPU clock then starting playing around with the mem and find it crashing often...how would one pinpoint the error to the new slight memory offset or the close to max GPU offset if they do not let the close to max GPU offset run in loop for hours? I mean it seems like it would be a prerequisite to confirm for stability for GPU offset before playing with mem offset (if I read the quick guide correctly it only says to do a 7 min heaven loops for each adjustment)
Edited by twitchyzero - 7/18/12 at 9:05pm
     
CPUMotherboardGraphicsGraphics
Intel Core i5-3570K @ 4.2GHz Gigabyte Z77X-UD5H-WB EVGA GTX 670 FTW 2GB @ 1.3/7.1GHz 1.21V EVGA GTX 670 FTW 2GB @ 1.3/7.1GHz 1.21V 
RAMHard DriveHard DriveOptical Drive
Samsung Green DDR3 4X4GB @ 2.1GHz Samsung 850 Pro 2TB Samsung SpinPoint F4 EcoGreen 2TB HDD LG GH24NS72 DVD Rewriter 
CoolingOSMonitorMonitor
Swiftech H240-X Liquid Cooler Microsoft Windows 10 Pro 64-bit Dell Ultrasharp U2312HM E-IPS Dell Ultrasharp U2311H E-IPS 
MonitorKeyboardPowerCase
Dell Ultrasharp U2311H E-IPS Logitech G15 Gaming Keyboard Rev 2 Seasonic X-750 80+ Gold 750W Rev 3 Corsair Carbide 500R Arctic White 
MouseMouse PadAudioAudio
Alienware TactX Gaming Mouse 5000DPI SteelSeries QcK Gaming Mousepad Edifier M3300SF 2.1 Speaker Set 54W RMS Logitech G930 7.1 Wireless Gaming Headset 
AudioAudioOtherOther
Sennheiser HD 650 Open-Ear Headphones Creative Sound Blaster E5 High-Res USB DAC/ Por... Logitech F710 Wireless Gamepad  Thrustmaster T-Flight Hotas X Flight Stick 
CPUGraphicsRAMHard Drive
Intel Core m7-6Y75 vPro Intel Graphics HD 515 DDR3L-1600MHz 8GB Samsung 960 Pro 1TB  
OSMonitorOther
Windows 10 Pro x64 Sharp IGZO 4K UHD touch Dell x Wacom active stylus 
  hide details  
Reply
     
CPUMotherboardGraphicsGraphics
Intel Core i5-3570K @ 4.2GHz Gigabyte Z77X-UD5H-WB EVGA GTX 670 FTW 2GB @ 1.3/7.1GHz 1.21V EVGA GTX 670 FTW 2GB @ 1.3/7.1GHz 1.21V 
RAMHard DriveHard DriveOptical Drive
Samsung Green DDR3 4X4GB @ 2.1GHz Samsung 850 Pro 2TB Samsung SpinPoint F4 EcoGreen 2TB HDD LG GH24NS72 DVD Rewriter 
CoolingOSMonitorMonitor
Swiftech H240-X Liquid Cooler Microsoft Windows 10 Pro 64-bit Dell Ultrasharp U2312HM E-IPS Dell Ultrasharp U2311H E-IPS 
MonitorKeyboardPowerCase
Dell Ultrasharp U2311H E-IPS Logitech G15 Gaming Keyboard Rev 2 Seasonic X-750 80+ Gold 750W Rev 3 Corsair Carbide 500R Arctic White 
MouseMouse PadAudioAudio
Alienware TactX Gaming Mouse 5000DPI SteelSeries QcK Gaming Mousepad Edifier M3300SF 2.1 Speaker Set 54W RMS Logitech G930 7.1 Wireless Gaming Headset 
AudioAudioOtherOther
Sennheiser HD 650 Open-Ear Headphones Creative Sound Blaster E5 High-Res USB DAC/ Por... Logitech F710 Wireless Gamepad  Thrustmaster T-Flight Hotas X Flight Stick 
CPUGraphicsRAMHard Drive
Intel Core m7-6Y75 vPro Intel Graphics HD 515 DDR3L-1600MHz 8GB Samsung 960 Pro 1TB  
OSMonitorOther
Windows 10 Pro x64 Sharp IGZO 4K UHD touch Dell x Wacom active stylus 
  hide details  
Reply
post #458 of 3357
Thread Starter 
Quote:
Originally Posted by twitchyzero View Post

First slot assuming the lane closest to the CPU?
Yup.
Quote:
What would it look like on the graph if I was throttled by the power at 100-115% when power slider is maxed at 145%? I merely suggested that because the temps were fine but it was unstable.
BTW thanks for suggesting me to loop heaven for 2 hours to test for stability...I thought about it but it sound like it defeated the purpose of your 30 minute guide biggrin.gif
So If I should test for stability for 2 hours...what's to say that let's say I follow the quick guide, find a close to max GPU clock then starting playing around with the mem and find it crashing often...how would one pinpoint the error to the new slight memory offset or the close to max GPU offset if they do not let the close to max GPU offset run in loop for hours? I mean it seems like it would be a prerequisite to confirm for stability for GPU offset before playing with mem offset (if I read the quick guide correctly it only says to do a 7 min heaven loops for each adjustment)

This is exactly correct, and the reason the long guide exists. In 99% of cases, you can resolve a crash by just reducing the GPU offset and in 99% of cases you can fix artifacts by reducing the memory offset. That's why i mention this in the last step of the short guide:
Quote:
Originally Posted by SeanPoe View Post

If you crash in any of these tests, then reduce your core offset by 2 until you stop crashing
If you see artifacts, then reduce your memory offset by 5 until you stop seeing artifacts
In the spirit of keeping the short method actually short, this is the best suggestion i can make as it holds true to almost every single 670; It also allows someone the majority of people to skip the 2 hour long heaven runs. biggrin.gif
Edited by SeanPoe - 7/18/12 at 11:38pm
post #459 of 3357
Quote:
Originally Posted by SeanPoe View Post

Do you have any examples of where this worked for other people without causing a throttle? When i tested this out, all that happened was I'd get throttled by power (which would then dynamically reduce my overclock), only making my unstable overclock stable because it was actually reducing my overclock. For example, if i was at 1300MHz with a +150MHz offset (unstable) and then reduced the power-target a bit, it would stabilize my overclock but only because i was getting throttled down to ~1250Mhz (though my offset was still set at 150). I found i could get more performance by just keeping the power target maxed and then manually reducing my overclock a bit. So instead of being at 1300MHz and 105% power target, i would set it at 1275MHz and 121% power target (this resulted in a performance increase). I have a hunch that in less-demanding applications (that don't tap out the power-draw causing a power-related throttle) that you might still be unstable without that throttle happening. So manually reducing your overclock would not only result in more performance in most cases but would also eliminate the possibility of instability when that power-related throttle doesn't happen.
This is just what I've observed with my own 670 anyways, things might be different for other people's 670s. If reducing the power target does indeed result in stability and increased performance for other people, I'll be sure to update the guide to reflect this new information. thumb.gif

First, great thread for discussion on 670's over clocking. A lot of great info. wink.gif

I don't have a 670, own a 680. Over clocking Kepler is similar across the series it seems from what I've been reading. I suggested to lower the power target as I've found I still max 1.175v whether PT offset was +135 or +125 on my 690. It did play a small role in lowering Kepler boost enough to maintain stability without lower the base Core over clock.

See for me with max Power Target I'd get GPU #1 1189 GPU #2 1214 but instability would occur when maxed for a length of time. Lowered PT down and now stable @ GPU #1 1175 & GPU 1201. When I did back down Core and kept PT maxed I only found it to decreased my Heaven & 3DMark11 bench scores. So it actually didn't improve performance to keep PT maxed with lower Core clock.

In short, main goal is to get highest over clock before hitting the magic 70C temp that will throttle the card 13 MHz Core clocks. It can be a balancing act between offsets for optimal results.

Thought Hokies83 could try that just to see if in fact it did give him stability. At that point some testing / trials on lowering Core while keeping Power Target maxed vs lowering power target with higher core gives better results. I found it to be the latter for my 24/7 over clock.

I know experiences and approaches can vary on over clocking Kepler. Seems there are in deed some variance between vendors.
     
  hide details  
Reply
     
  hide details  
Reply
post #460 of 3357
Thread Starter 
Quote:
Originally Posted by Arizonian View Post

First, great thread for discussion on 670's over clocking. A lot of great info. wink.gif
I don't have a 670, own a 680. Over clocking Kepler is similar across the series it seems from what I've been reading. I suggested to lower the power target as I've found I still max 1.175v whether PT offset was +135 or +125 on my 690. It did play a small role in lowering Kepler boost enough to maintain stability without lower the base Core over clock.
See for me with max Power Target I'd get GPU #1 1189 GPU #2 1214 but instability would occur when maxed for a length of time. Lowered PT down and now stable @ GPU #1 1175 & GPU 1201. When I did back down Core and kept PT maxed I only found it to decreased my Heaven & 3DMark11 bench scores. So it actually didn't improve performance to keep PT maxed with lower Core clock.
In short, main goal is to get highest over clock before hitting the magic 70C temp that will throttle the card 13 MHz Core clocks. It can be a balancing act between offsets for optimal results.
Thought Hokies83 could try that just to see if in fact it did give him stability. At that point some testing / trials on lowering Core while keeping Power Target maxed vs lowering power target with higher core gives better results. I found it to be the latter for my 24/7 over clock.
I know experiences and approaches can vary on over clocking Kepler. Seems there are in deed some variance between vendors.

Thanks for sharing your experience. I'll be sure to add your advice to lower the power target as a way to troubleshoot an unstable overclock in the troubleshooting section of the guide.

I think i perhaps misinterpreted what you meant in your your previous post. I thought you meant to leave the core offset alone and only reduce the power target as a way to regain stability (which as i described above, would only make the overclock stable because of the power-related throttle that occurs). I didn't know you meant to reduce both the offset and power target in tandem, which makes a lot more sense, and also wouldn't have the problem i described above as my 'hunch'.
Edited by SeanPoe - 7/18/12 at 11:54pm
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: NVIDIA
Overclock.net › Forums › Graphics Cards › NVIDIA › ~~The GTX 670 Overclocking Master-Guide~~