Overclock.net › Forums › Graphics Cards › NVIDIA › [Official] NVIDIA GTX 1070 Owner's Club
New Posts  All Forums:Forum Nav:

[Official] NVIDIA GTX 1070 Owner's Club - Page 854

post #8531 of 8672
Quote:
Originally Posted by Yukss View Post
 
Quote:
1070 performance is adequate for most things, even if it requires you to reduce some settings from ultra to high on some titles, it is still a good experience.

I would wait 6 months and look at the gtx2070 that will perform about the same as a 1080ti does now or a new 2080 that should beat it if nvidia stay true to form.

yes, i guess we can always wait until next gens of cards but here the deals is to get a 1080 founders edition with waterblock at 450$ and try to sell mine at 380$ at least (it cost me 450$ less than a month ago)

 

3 months ago I would have said go for it. Now though, if it was me, I would be really disappointed if having spent just $450, I could get a new card for $400 that blows the card I just bought to the weeds.

 

You also mentioned that you have another spare 1070. If you did SLI with your existing 1070 you get 1080TI performance, for the most part and it costs you nothing extra


Edited by gtbtk - 5/3/17 at 7:15am
post #8532 of 8672
Quote:
Originally Posted by gtbtk View Post

3 months ago I would have said go for it. Now though, if it was me, I would be really disappointed if having spent just $450, I could get a new card for $400 that blows the card I just bought to the weeds.

You also mentioned that you have another spare 1070. If you did SLI with your existing 1070 you get 1080TI performance, for the most part and it costs you nothing extra

thanks again for your time, yes, i have a evga 1070 sc and a brand new asus 1070 strix oc, i did ran in sli for one day to test, but i did not like itm miss match lol, performance wise are monsters, but im running a 27" 1080p, waiting for a 2k monitor that i bought, sli is not an option because im more in to one gpu setup.. the "spare" gtx 1070 was for a second build, and i cannot sell both card because i promised the evga for a friend as a gift, so basically i have only one 1070 (the asus strix) for selling in order to get a 1080 or god knows what (no amd)
Yukss´s rig
(15 items)
 
  
CPUMotherboardGraphicsRAM
i7 3930k @4.6ghz HT 1.3 v EVGA X79 FTW gtx 1080 FE with EK waterblock 16gb Corsair Vengance 
Hard DriveHard DriveCoolingOS
OCZ vertex 3 240gb  Samsung 1tb Custom Watercooling Loop Windows 10 64bit 
MonitorKeyboardPowerCase
Samsung 27" 2ms 1920*1080 G.SKILL km570 mx Corsair HX 850w Thermaltake core p5 
MouseMouse PadAudio
Saitek R.A.T 5 Razer Goliathus speed SoundBlaster x-fi Titaniun 
  hide details  
Reply
Yukss´s rig
(15 items)
 
  
CPUMotherboardGraphicsRAM
i7 3930k @4.6ghz HT 1.3 v EVGA X79 FTW gtx 1080 FE with EK waterblock 16gb Corsair Vengance 
Hard DriveHard DriveCoolingOS
OCZ vertex 3 240gb  Samsung 1tb Custom Watercooling Loop Windows 10 64bit 
MonitorKeyboardPowerCase
Samsung 27" 2ms 1920*1080 G.SKILL km570 mx Corsair HX 850w Thermaltake core p5 
MouseMouse PadAudio
Saitek R.A.T 5 Razer Goliathus speed SoundBlaster x-fi Titaniun 
  hide details  
Reply
post #8533 of 8672
Quote:
Originally Posted by gtbtk View Post

The .950v point is the voltage point that also controls the video clock speed.  That is the reason why I said use that point. 

One thing that maybe you haven't paid any attention to is that thin line you can see on that pops out under the main curve. Understanding what that means will help you not spend ages scratching your head wondering why the curve never seems to be consistent. Keep in mind that the graphics card actually uses all the voltage points up to the level where it stops increasing, not just the highest point on the curve.

The way the AB curve works, is that it has a default set of voltage values are fixed in the bios. OC versions of the card have higher values baked in to the bios. The card uses those fixed values as the default curve. Any time you do any overclocking, you are not actually changing those values, instead, you are telling the card to apply an offset to each of the fixed values that you adjust. Using the slider and adding +100 in AB tells the card to offset every value in the range by and extra 100 on the y axis if the graph. If you use the slider and look at the graph, you will see that the thin line stays where the default curve used to be. 

If you use the individual voltage points on the curve to overclock, this is where it can get confusing. Even though all the points on the curve appears to move around the one you just moved, they actually are not. Even though the thick curve appears to move, the Afterburner application is just relocating the where it is representing the 0 offset on screen. Intuitively, I think most people would assume that the 0 offset point would remain fixed on the screen but it doesn't.

The thin line that you can see is actually there to indicate how much offset that any given voltage point has from the 0 level that is moving all around the screen. If the thin line ends up above the main curve line, it is showing you that at that voltage point the offset is negative. Sometimes you can move the point and hit apply and it will look like the curve didnt move. but if you pay attention to the thin line, that did move and it shows that the offset was actually changed.

This curve is only there to illustrate what I am saying and is not a usable curve for your graphics card. It took me more than 6 months of head scratching before I worked out what the curve was actually telling me and understand why it all seemed so inconsistent.

This is what a curve should look like if the 0 offset remained fixed on the screen and you made positive and negative adjustments. It the way I think everyone assumes the curve works, but it doesn't in Afterburner. The curve adjustment in precision XOC is ugly and clunky, but, the 0 baseline is actually flat along the bottom of the histogram, making it much easier to understand what you are doing if you manually adjust curves. 




This is actually the exact same curve after you hit the apply button. See how it looks nothing like what a normal person would assume it should look like given the settings in the first screencap? The thin line though, being above and below the thick line does actually show you where points have been offset in a positive direction and where points have been offset in a negative direction. The thin line is actually 0.  

 



If you look at your curve, the only points that are getting any offset from default are the ones between .900 and .950. Everything else is running at stock. There is more performance that you can get out of your card while still running at a lower voltage than default.

What I would suggest, is that you increase every voltage point from the 0.800v point up to 0.950v. If you want to limit the cards voltage to .950v, don't adjust anything above that point. The highest level for each point will be dependent on your card but I would try adding +200 mhz to each of those low range point first and test for stability. If it is stable, increase them all by +25 and test again. if not reduce each point by 25 and test. when you get the .8 to .95v points adjusted and stable, then increase the .950v point up to 2025mhz if you are still below that level on the curve and see how that works for you.

If you are really keen, you can adjust each point individually as they are all likely to have different maximum offset values but the gains are probably not worth the extra time to fine tune it and maybe see a 1fps improvement in some games. 

Thanks for the detailed explanation of the curve, things indeed make quite a bit more sense now! I ran a few tests with 2000mhz @ 0.900v and 2000mhz @ 0.950v and I indeed lost quite a bit performance:

FFXIV Stormblood Benchmark:

Stock:

Score: 12935
Average Frame Rate: 91.601

0.900v:

Score: 12987
Average Frame Rate: 91.750

0.950v:

Score: 12971
Average Frame Rate: 91.861


Unigine Heaven Benchmark 4.0:

Stock:

FPS:
91.2

Score:
2296

Min FPS:
9.0

Max FPS:
192.1

0.900v:

FPS:
91.3

Score:
2300

Min FPS:
31.8

Max FPS:
188.6

0.950v:

FPS:
95.5

Score:
2405

Min FPS:
31.5

Max FPS:
196.4

3DMark Firestrike:

Stock: 13364
0.900v: 13247
0.950v: 13720

That odd results are from some benchmarks jumping all over the states at times so I have to figure out the lower clocks individually. Setting the lower states to +200 across the board sadly doesn't work for me as some games start to crash then as soon as they land at some lower values as soon as videos play or I'm in a menu. I think I'll keep the 0.900v curve in a separate profile though as it gives me around stock performance with vastly lower temps which will be useful for summer when I feel 0.950v will also start to creep up a bit more. For now 0.950v never went higher than 74°C for me at good fan-speeds but I like to have headroom for these 30°C+ days.
Edited by Skyblaze - 5/3/17 at 1:04pm
post #8534 of 8672
Broke my FireStrike record again biggrin.gif

http://www.3dmark.com/3dm/19701453

Any beastly 1070´s around ?
post #8535 of 8672
Quote:
Originally Posted by Skyblaze View Post
 
 
Thanks for the detailed explanation of the curve, things indeed make quite a bit more sense now! I ran a few tests with 2000mhz @ 0.900v and 2000mhz @ 0.950v and I indeed lost quite a bit performance:

FFXIV Stormblood Benchmark:

Stock:

Score: 12935
Average Frame Rate: 91.601

0.900v:

Score: 12987
Average Frame Rate: 91.750

0.950v:

Score: 12971
Average Frame Rate: 91.861


Unigine Heaven Benchmark 4.0:

Stock:

FPS:
91.2

Score:
2296

Min FPS:
9.0

Max FPS:
192.1

0.900v:

FPS:
91.3

Score:
2300

Min FPS:
31.8

Max FPS:
188.6

0.950v:

FPS:
95.5

Score:
2405

Min FPS:
31.5

Max FPS:
196.4

3DMark Firestrike:

Stock: 13364
0.900v: 13247
0.950v: 13720

That odd results are from some benchmarks jumping all over the states at times so I have to figure out the lower clocks individually. Setting the lower states to +200 across the board sadly doesn't work for me as some games start to crash then as soon as they land at some lower values as soon as videos play or I'm in a menu. I think I'll keep the 0.900v curve in a separate profile though as it gives me around stock performance with vastly lower temps which will be useful for summer when I feel 0.950v will also start to creep up a bit more. For now 0.950v never went higher than 74°C for me at good fan-speeds but I like to have headroom for these 30°C+ days.

I only suggested 200 as a starting point. The amount that you can increase these cards clock speeds is dependent on the silicon itself and what the factory set the default clocks at. A reference clocked card can add many more points to an overclock than say a Gigabyte Xtreme Gaming card can simply because the Gigabyte card has already had an overclock applied that is close to the silicons limits in the factory.

 

On my GamingX I can take the lower points up to about +150 but the points around 1.0-1.025v don't really like being increased more than about +75 and then from 1.043v it is ok with about an extra 100. If i tune CPU PLL voltages, I can tune that dip in performance out, but CPU temps start increasing. While there is certainly variation in silicon chips performance, I really think that silicon lottery gets blamed more than it should because motherboard voltage combinations, ram speed and latency cpu clocks all impact How well your GPU performs.

 

I Have experimented with it but I do not run my card undervolted as a general rule. I have settled to run my card at the default voltage and overclock with the curve that has a hump at .950 dips after that and then boosts up again to peak at 1.063v.

 

When I first got the card, my first instinct was to add +100 to the voltage slider and run the card a 1.093V all the time and the card performs OK. It will let me run the card stable at 2126Mhz.  At default voltages, I can only run the card stable at 2076Mhz but I discovered that it runs almost 10 deg cooler and I get similar Firestrike and Heaven results. http://www.3dmark.com/fs/11822144 this is a recent FS result running on an i7-2600. Unfortunately the z68 MB died and I not found a reasonably priced replacement yet.

 

1070 love memory bandwidth and will increase your Benchmark scores significantly. Clock your memory as fast as you stably can, that should probably be somewhere around +500 to +600. The card may run the memory faster, but gpu memory errors start creeping in and the performance drops off again. OCCT is a good tool that can test graphic card memory overclocks.

 

Make sure that, unless you use it, you also disable the windows GameDVR service running on Win 10 by default. It kills GPU benchmark performance.

post #8536 of 8672
So what's are recommended, stable OC settings on the Gaming X (base voltage)?
I ask because I want a starting point for Oc'ing mine.
post #8537 of 8672
Quote:
Originally Posted by ParlyShary View Post

So what's are recommended, stable OC settings on the Gaming X (base voltage)?
I ask because I want a starting point for Oc'ing mine.

 

You could try starting at:

 

Voltage 0

Powerlimit 126

Temp limit 93

core clock +100 or +75

memory clock +500

 

adjust core clock in + or - 25 increments to get close and then dial it in in +/- 12 increments to fine tune. smaller adjustments wont make any changes to reported values.

 

Create a fan curve. I don't mind the 100% fan noise from my GamingX so for benchmarking my fans are at 30% at idle keeping temps at around 30-35 deg, start increasing at 40 deg and get to 100% at about 50. That will keep temps about as low as they can go and keep your clocks up. If you use the zerofan option, it will idle at 59 deg and get hotter from there.

 

Don't be afraid of increasing the voltage slider, it wont hurt the card. The 100 value is not mV, it is a percentage offset adjusting the highest point in the curve between 1063mV and 1093mV. the extra voltage will let you run the card faster.

 

Once you have dialed in settings overclocking on the slider, there is probably more performance to be found using the curve to overclock, but you are best off getting the slider oc sorted first.

post #8538 of 8672
Quote:
Originally Posted by Minium View Post

Broke my FireStrike record again biggrin.gif

http://www.3dmark.com/3dm/19701453

Any beastly 1070´s around ?

That's a great Graphics score. :specool:

 

I'm tempted to point you to this one http://www.3dmark.com/fs/10295423 :devil-smi but it was a borked driver run.

 

There are some 1070s around that will do 22000+ graphics score. Unfortunately I haven't found a way to get mine there yet.

 

This is one of the best I have managed with an i7-2600 http://www.3dmark.com/fs/11532231

 

 

 

 

 

The best I have managed to get min is 21500

post #8539 of 8672
Quote:
Originally Posted by Minium View Post

Broke my FireStrike record again biggrin.gif

http://www.3dmark.com/3dm/19701453

Any beastly 1070´s around ?

You have Galax/KFA HOF 980ti card, running with 1636mhz/8280mhz effective, not even 980ti in general can do that.
However current highest pro-overclocker 980ti under LN2 could do +1900mhz corespeed. http://www.3dmark.com/fs/6825897 28,829 Graphic scores.

Might be 1070 HOF owner has capability to reach yours, or some lucky people in the world probably could reach that who knows, not every single 1070s owner are at OCN.

You can search results at 3dmark.com, as far as I can find single 1070 about +22.4xx graphic scores.

btw did you only run the benchmark? no stable test/stress test/games?

I can reach my highest scores +21k, but I turn it down, coreclock too high would result crashes during stress tests, and also the memory I turn it down to +600mhz because it's start artifacting at +700mhz in stress tests. Achieving high scores/clocks but no perfect stable it's not my goals in overclocking.
Edited by zipper17 - 5/4/17 at 1:53am
PC Gaming 4 Life
(11 items)
 
  
CPUMotherboardGraphicsRAM
Intel Core i5 3570k @4.7GHZ Biostar TZ77XE4 Galax GTX 1070 EXOC 2x8 Team Vulcan 19200 @2400 CL11 DDR3 
Hard DriveHard DriveOptical DriveCooling
Seagate 500GB, Seagate 2TB Crucial MX100 256GB Lite On 24x sata CM Hyper 212 EVO  
OSMonitorPower
Win 10 Pro 64 bit  Dell U2515H Seasonic ss 650 At 
  hide details  
Reply
PC Gaming 4 Life
(11 items)
 
  
CPUMotherboardGraphicsRAM
Intel Core i5 3570k @4.7GHZ Biostar TZ77XE4 Galax GTX 1070 EXOC 2x8 Team Vulcan 19200 @2400 CL11 DDR3 
Hard DriveHard DriveOptical DriveCooling
Seagate 500GB, Seagate 2TB Crucial MX100 256GB Lite On 24x sata CM Hyper 212 EVO  
OSMonitorPower
Win 10 Pro 64 bit  Dell U2515H Seasonic ss 650 At 
  hide details  
Reply
post #8540 of 8672
Quote:
Originally Posted by zipper17 View Post

You have Galax/KFA HOF 980ti card, running with 1636mhz/8280mhz effective, not even 980ti in general can do that.
However current highest pro-overclocker 980ti under LN2 could do +1900mhz corespeed. http://www.3dmark.com/fs/6825897 28,829 Graphic scores.

Might be 1070 HOF owner has capability to reach yours, or some lucky people in the world probably could reach that who knows, not every single 1070s owner are at OCN.

You can search results at 3dmark.com, as far as I can find single 1070 about +22.4xx graphic scores.

btw did you only run the benchmark? no stable test/stress test/games?

I can reach my highest scores +21k, but I turn it down, coreclock too high would result crashes during stress tests, and also the memory I turn it down to +600mhz because it's start artifacting at +700mhz in stress tests. Achieving high scores/clocks but no perfect stable it's not my goals in overclocking.
Its not even a galax Its a gigabyte 989Ti xtreme gaming
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: NVIDIA
Overclock.net › Forums › Graphics Cards › NVIDIA › [Official] NVIDIA GTX 1070 Owner's Club