Overclock.net › Forums › Graphics Cards › NVIDIA › [Official] NVIDIA GTX 1080 Ti Owner's Club
New Posts  All Forums:Forum Nav:

[Official] NVIDIA GTX 1080 Ti Owner's Club - Page 1427

post #14261 of 15190
Quote:
Originally Posted by Beagle Box View Post

Maybe that's true for Classifieds, but I think what you actually found was that the fastest memory speeds are found 50 ticks apart on the slider.

I've worked with two GTX 1080s (Zotac and MSI) and once I found a 'sweet spot' in memory speed, I'd find another at precisely +50 or -50 from that point. Neither card's sweet spots ended in '0' or '5'.

That makes sense too. I wonder why every +50 or -50 are stronger than numbers in-between. An overclock is still an overclock. So it shouldn't really matter. But for some reason it does.
Nick-PC
(9 items)
 
  
CPUMotherboardGraphicsRAM
I7 7700K ASUS Z270 ROG MAXIMUS IX APEX NVIDIA GeForce GTX 1080Ti G.SKILL Trident Z 3600 C16 
CoolingOSMonitorPower
EK KIT P360 (Custom Water Loop) Microsoft Windows 10 Home, 64 bit ASUS PG278Q 850 Bronze 'be Quiet POWER ZONE' 
Case
Corsair Obsidian 750d Airflow 
  hide details  
Reply
Nick-PC
(9 items)
 
  
CPUMotherboardGraphicsRAM
I7 7700K ASUS Z270 ROG MAXIMUS IX APEX NVIDIA GeForce GTX 1080Ti G.SKILL Trident Z 3600 C16 
CoolingOSMonitorPower
EK KIT P360 (Custom Water Loop) Microsoft Windows 10 Home, 64 bit ASUS PG278Q 850 Bronze 'be Quiet POWER ZONE' 
Case
Corsair Obsidian 750d Airflow 
  hide details  
Reply
post #14262 of 15190
Quote:
Originally Posted by nuno_p View Post

How?

What you do is set the curve so that the highest voltage is at the frequency you want. It'll boost up to that voltage and frequency and no farther, because you have the curve fall off at that point.

I don't even bother w/the custom voltage curve anymore though, I just add an offset to the frequency and extend the voltage range.
post #14263 of 15190
Quote:
Originally Posted by nrpeyton View Post

That makes sense too. I wonder why every +50 or -50 are stronger than numbers in-between. An overclock is still an overclock. So it shouldn't really matter. But for some reason it does.

Different timing straps or a clock crossing boundary that has a different latency at different different memory multipliers.
Primary
(15 items)
 
Secondary
(13 items)
 
In progress
(10 items)
 
CPUMotherboardGraphicsRAM
5820K @ 4.2/3.5GHz core/uncore, 1.175/1.15v Gigabyte X99 SOC Champion (F22n) Gigabyte AORUS GTX 1080 Ti (F3P) @ 2025/1485, 1... 4x4GiB Crucial @ 2667, 12-12-12-28-T1, 1.34v 
Hard DriveHard DriveHard DriveCooling
Plextor M6e 128GB (fw 1.06) M.2 (PCI-E 2.0 2x) 2x Crucial M4 256GB 4x WD Scorpio Black 500GB Noctua NH-D15 
OSMonitorKeyboardPower
Windows 7 Professional x64 SP1 BenQ BL3200PT Filco Majestouch Tenkeyless (MX Brown) Corsair RM1000x 
CaseMouseAudio
Fractal Design Define R4 Logitech G402 Realtek ALC1150 + M-Audio AV40 
CPUMotherboardGraphicsRAM
X5670 @ 4.4/3.2GHz core/uncore, 1.36 vcore, 1.2... Gigabyte X58A-UD5 r2.0 w/FF3mod10 BIOS Sapphire Fury Nitro OC+ @ 1053/500, 1.225vGPU/1... 2x Samsung MV-3V4G3D/US @ 2000, 10-11-11-30-T1,... 
RAMHard DriveHard DriveHard Drive
1x Crucial BLT4G3D1608ET3LX0 @ 2000, 10-11-11-3... OCZ (Toshiba) Trion 150 120GB Hyundai Sapphire 120GB 3x Hitachi Deskstar 7k1000.C 1TB 
CoolingOSPowerCase
Noctua NH-D14 Windows 7 Pro x64 SP1 Antec TP-750 Fractal Design R5 
Audio
ASUS Xonar DS 
CPUMotherboardGraphicsRAM
i7-6800K @ 4.3/3.5GHz core/uncore, 1.36/1.2v ASRock X99 OC Formula (P3.10) GTX 780 (temporary) 4x4GiB Crucial DDR4-2400 @ 11-13-12-28-T2, 1.33v 
Hard DriveHard DriveCoolingOS
Intel 600p 256GB NVMe 2x HGST Travelstar 7k1000 1TB Corsair H55 (temporary) Windows Server 2016 Datacenter 
PowerCase
Seasonic SS-860XP2 Corsair Carbide Air 540 
  hide details  
Reply
Primary
(15 items)
 
Secondary
(13 items)
 
In progress
(10 items)
 
CPUMotherboardGraphicsRAM
5820K @ 4.2/3.5GHz core/uncore, 1.175/1.15v Gigabyte X99 SOC Champion (F22n) Gigabyte AORUS GTX 1080 Ti (F3P) @ 2025/1485, 1... 4x4GiB Crucial @ 2667, 12-12-12-28-T1, 1.34v 
Hard DriveHard DriveHard DriveCooling
Plextor M6e 128GB (fw 1.06) M.2 (PCI-E 2.0 2x) 2x Crucial M4 256GB 4x WD Scorpio Black 500GB Noctua NH-D15 
OSMonitorKeyboardPower
Windows 7 Professional x64 SP1 BenQ BL3200PT Filco Majestouch Tenkeyless (MX Brown) Corsair RM1000x 
CaseMouseAudio
Fractal Design Define R4 Logitech G402 Realtek ALC1150 + M-Audio AV40 
CPUMotherboardGraphicsRAM
X5670 @ 4.4/3.2GHz core/uncore, 1.36 vcore, 1.2... Gigabyte X58A-UD5 r2.0 w/FF3mod10 BIOS Sapphire Fury Nitro OC+ @ 1053/500, 1.225vGPU/1... 2x Samsung MV-3V4G3D/US @ 2000, 10-11-11-30-T1,... 
RAMHard DriveHard DriveHard Drive
1x Crucial BLT4G3D1608ET3LX0 @ 2000, 10-11-11-3... OCZ (Toshiba) Trion 150 120GB Hyundai Sapphire 120GB 3x Hitachi Deskstar 7k1000.C 1TB 
CoolingOSPowerCase
Noctua NH-D14 Windows 7 Pro x64 SP1 Antec TP-750 Fractal Design R5 
Audio
ASUS Xonar DS 
CPUMotherboardGraphicsRAM
i7-6800K @ 4.3/3.5GHz core/uncore, 1.36/1.2v ASRock X99 OC Formula (P3.10) GTX 780 (temporary) 4x4GiB Crucial DDR4-2400 @ 11-13-12-28-T2, 1.33v 
Hard DriveHard DriveCoolingOS
Intel 600p 256GB NVMe 2x HGST Travelstar 7k1000 1TB Corsair H55 (temporary) Windows Server 2016 Datacenter 
PowerCase
Seasonic SS-860XP2 Corsair Carbide Air 540 
  hide details  
Reply
post #14264 of 15190
Interesting that the clock number ending affecting perceived performance is mentioned - I swear my 1080ti runs smoother at clock additions ending in 75.
+75, +175 , +375 etc.
Pretty confident it's not placebo either. . .
post #14265 of 15190
Quote:
Originally Posted by WillG027 View Post

Interesting that the clock number ending affecting perceived performance is mentioned - I swear my 1080ti runs smoother at clock additions ending in 75.
+75, +175 , +375 etc.
Pretty confident it's not placebo either. . .

You are not seeing things. What you are seeing is the effect of memory straps (i.e. latencies) on Bandwidth.

If my clockspeed is aligned with my strap (for me it's any factor of 10) the bandwidth difference is at least 50GB/s vs misaligned.

The catch here is the straps get retrained at the beginning of every single Windows 10 boot (the IMC trains itself according to parameters like bus heat, VRAM heat, IMC heat, electrical factors etc etc). To find the aligned strap, I have to constantly re-test any VRAM OC because the strap alignment is +-10 from the original value of last boot.

I just run the CUDA VRAM Bandwidth test (originally designed to test the 970 4gb array) and if it spits out 495GB/s then I know I'm aligned. Otherwise, it shows 440GB/s.
Edited by Dasboogieman - 10/3/17 at 3:01am
post #14266 of 15190
Thanks for that ^^ DasBoogieman.

It's both interesting and useful.
post #14267 of 15190
Quote:
Originally Posted by Dasboogieman View Post

You are not seeing things. What you are seeing is the effect of memory straps (i.e. latencies) on Bandwidth.

If my clockspeed is aligned with my strap (for me it's any factor of 10) the bandwidth difference is at least 50GB/s vs misaligned.

The catch here is the straps get retrained at the beginning of every single Windows 10 boot (the IMC trains itself according to parameters like bus heat, VRAM heat, IMC heat, electrical factors etc etc). To find the aligned strap, I have to constantly re-test any VRAM OC because the strap alignment is +-10 from the original value of last boot.

I just run the CUDA VRAM Bandwidth test (originally designed to test the 970 4gb array) and if it spits out 495GB/s then I know I'm aligned. Otherwise, it shows 440GB/s.

Can you go into more detail on this?

For example if your clock speed is exactly 2000mhz and you have set VRAM to exactly 8ghz(assuming 8000mhz is a possible strap) you will have perfect alignment with full VRAM speed.
If you set 2012 mhz with 8ghz vram, your VRAM bandwidth drops down to 440gbs.

Is that correct?
post #14268 of 15190
Quote:
Originally Posted by Agent-A01 View Post

Can you go into more detail on this?

For example if your clock speed is exactly 2000mhz and you have set VRAM to exactly 8ghz(assuming 8000mhz is a possible strap) you will have perfect alignment with full VRAM speed.
If you set 2012 mhz with 8ghz vram, your VRAM bandwidth drops down to 440gbs.

Is that correct?

Yes that is correct. Which side it swings to can vary between boots (I haven't done detailed testing but I suspect the main culprit for the repeated VRAM retraining is the operating temperature variances).

You can test it yourself with these two applications:

Use the NVIDIA Profile inspector to change the P2 CUDA VRAM speed throttle to force the VRAM to run at full speed in CUDA applications.
Then use the VRAM bandwidth test to parse the realtime bandwidth.

nvidiaProfileInspector.zip 129k .zip file
vRamBandWidthTest.zip 120k .zip file
post #14269 of 15190
Quote:
Originally Posted by Dasboogieman View Post

Yes that is correct. Which side it swings to can vary between boots (I haven't done detailed testing but I suspect the main culprit for the repeated VRAM retraining is the operating temperature variances).

You can test it yourself with these two applications:

Use the NVIDIA Profile inspector to change the P2 CUDA VRAM speed throttle to force the VRAM to run at full speed in CUDA applications.
Then use the VRAM bandwidth test to parse the realtime bandwidth.

nvidiaProfileInspector.zip 129k .zip file
vRamBandWidthTest.zip 120k .zip file

So what VRAM clock are you running at?

I'm using +499 which equates to 460-470 GB/s.

I forced P2 off in cuda, VRAM shows max clock; core does not max out.
post #14270 of 15190
Why would a 2070 beat 1080ti,1080 was barely faster than 980ti when launched.
And knowing how nvidia likes to release the potato GPU-s first and then months later release the good cards like 1080ti, so they can ofcourse milk everyone, with titans.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: NVIDIA
Overclock.net › Forums › Graphics Cards › NVIDIA › [Official] NVIDIA GTX 1080 Ti Owner's Club