Overclocking with newbie (me) 970G1 Gigabyte

[Pablo1990]

Hello
I was thinking about overclocking my GPU so i read plenty of topics, watched hours of youtube videos.

I have Gigabyte Geforce GTX970 G1 Gaming GPU and Be Quiet Pure Power 630 Wat PSU with +80 Bronze.
It gives a lot of power , and should be sufficient so I turned up the GPU and CPU future.

So i started from raising up power limit to 112%.... and i get core clock +190 stable with +25 mV more Core Voltage.

So i started raising memory. From GPU-Z i read that i have Samsung memory so i thinked that +500 will be reachable.
I was wrong. With +450 i get crash when runing Furmark. So i get +400 stable.

I wanna ask, did my memory isn't that good, or i need to raise voltage or sth? Also i see in furmark that core clock isnt 1519 as GPU-Z shows, but i have 1355? Why?

It's my first overclock so if you can please explain me this as simple as you can

Best Regards
Pablo

 

[JackNaylorPE]

I have **observed** nVidia cards throttle OCCT but not Furmark....not since 5xx series anyway.

http://nvidia.custhelp.com/app/answers/detail/a_id/2857

By observed I mean, with the power limiter set say at 110%, I routinely see power draw exceeding that value (112 - 113).

I first noticed this while running OCCT on a lappie with a 675M ... asked vendor for a suggesting to check ability of lappie cooling system to sustain high head loads, and they suggested Furmark.... unlike w/ OCCT, I was able to put 100% loads on the GPU whereas OCCT did not. ... but that was 2 laptops ago.

I thot throttling today was strictly via thermal detection these days.

Can't see how a reputable site like techpowerup would continue to use it in all of their testing if this was still the case.

https://www.techpowerup.com/reviews/MSI/GTX_980_Ti_Gaming/28.html

Maximum: We use Furmark Stability Test at 1280x1024, 0xAA. This results in a very high no-game power-consumption reading that can typically only be reached with stress-testing applications. We report the highest single reading after a short startup period. Initial bursts during startup are not included as they are too short to be relevant.

Interesting that the Furmark test here hit 279 watts under Furmark and 281 watts under Metro 2033.

If it' supposed to detect Furmark and throttle, it's clearly not working. If nVidia was detecting Furmark, I don't see it's how it's feasible that Gigabytes 980 Ti could hit 359 watts... more than 100 watts (44%) over nvidia's stated TDP of 250 watts.

http://www.techpowerup.com/reviews/Gigabyte/GTX_980_Ti_G1_Gaming/28.html

Nvidia is doing so many things these days to limit card voltage (both legally and physically) that it's pretty much impossible to break a card. Even with custom BIOSs, and dual pump cooling systems, I have not been able to do much past normal air cooling OCs.

On my wc'd 780s (39C cores / mid 50s VRM / AISC = 77% / 80%)) under Furmark, I didn't do so well. A custom BIOS got me higher core speeds, but benchmark fps (firestrike, heaven, valley) was lower at the higher cores and temps rose so I went back to stock. I managed a 26% OC under all the standard benchmarks..... BF4, Metro 2033 and Witcher 3 ver 1.08 (1.06 was fine) don't like it tho.

 

[Vellinious]

I have **observed** nVidia cards throttle OCCT but not Furmark....not since 5xx series anyway.

http://nvidia.custhelp.com/app/answers/detail/a_id/2857

By observed I mean, with the power limiter set say at 110%, I routinely see power draw exceeding that value (112 - 113).

I first noticed this while running OCCT on a lappie with a 675M ... asked vendor for a suggesting to check ability of lappie cooling system to sustain high head loads, and they suggested Furmark.... unlike w/ OCCT, I was able to put 100% loads on the GPU whereas OCCT did not. ... but that was 2 laptops ago.

I thot throttling today was strictly via thermal detection these days.

Can't see how a reputable site like techpowerup would continue to use it in all of their testing if this was still the case.

https://www.techpowerup.com/reviews/MSI/GTX_980_Ti_Gaming/28.html

Maximum: We use Furmark Stability Test at 1280x1024, 0xAA. This results in a very high no-game power-consumption reading that can typically only be reached with stress-testing applications. We report the highest single reading after a short startup period. Initial bursts during startup are not included as they are too short to be relevant.

Interesting that the Furmark test here hit 279 under Furmark and 281 under Metro 2033

If it' supposed to detect Furmark and throttle, it's clearly not working. If nVidia was detecting Furmark, I don't see it's how it's feasible that Gigabytes 980 Ti could hit 359 watts... more than 100 watts (44%) over nvidia's stated TDP of 250 watts.

http://www.techpowerup.com/reviews/Gigabyte/GTX_980_Ti_G1_Gaming/28.html

Yes, there was a throttle put in place by NVIDIA for furmark. Since the beginning of DX11 actually, dating back to Fermi. Google is your friend.

Want to keep using it? Go right ahead. It's not a tool that anyone really uses any more, because there's no use or need for it. It's a heat machine...that's all.

And yes, there are other "throttles" other than thermal. The power limit will throttle the Maxwell GPU during normal load testing with anywhere from mild to good overclocks. It's pretty rare to see a Maxwell hit decent overclocks without needing a custom bios file to get around the power limits present in the stock bios.

= )

 

[JackNaylorPE]

Yes, there was a throttle put in place by NVIDIA for furmark. Since the beginning of DX11 actually, dating back to Fermi. Google is your friend.

Yes there **was** ... I linked to it on nvidia support site .... but emphasis on past tense... google and yahoo found zero official or authoritative mentions of it past the 5xx series other than peeps parroting the 5xx issue. The 5xx series reference cards, particularly the 570 and 580 were hard hit here. This was especially true for the EVGA SC series cards which had the reference 4 phase VRM while MSI, Giga and Asus were using a non reference PCB with 6 or 7 phases. The 6xx, 7xx and 9xx benefit from more robust VRM designs but efficiency has been consistently improving over the last few generations making this pretty much a non issue.

Also reference my previous post where with my old lappie... nVidia was throttling OCCT but Nvidia themselves told me to use Furmark which was not throttled .... and when I did, it wasn't.

And yes, Google is very useful, but so is common sense. If nVidia is detecting Furmark and throttling the GPU, it's clearly not working. Otherwise how do you explain being able to pull 359 watts out of what nVidia says is a 250 watt TDP GPU ? Not doing much throttling if it's allowing a load 44% beyond it's stated capacity. .... and that's before overclocking....add 20% when OC'ing a 980 Ti ...that 14% speed bump that ya get when ya slide the AB slider to 120% doesn't come for free.

No one uses furmark ? Clearly not correct as Techpowerup uses it in their testing as do other reputable sites. I don't know a wc enthusiast who doesn't.

1. I use it for the same reason as I use P95 .... nothing better for setting TIM. With either of these I can raise CPU / GPU temps up to the max in a matter of seconds.... and let the CPU / GPUs cool down quickly as the water loop has lots of thermal capacity and recovers quickly from short runs... with other tests takes much longer to get CPU / GPU to max temps. After 4 or 5 cycles, I am now ready to start OC'ing knowing that maximum thermal conductivity has been reached (well as long as your not using AS5 .... then ya gotta wait 7 weeks).

2. Once that's done, before dialing in OC's, I'll need to set up system fan and pump curves for the system (10 radiator fans and 6 case fans in system I'm typing from). For this I'll need a constant max heat load that is repeatable from test to test so that data on pump speeds, ambient and CPU / GPU temps have relevance.

3. As for stability .....

-I have had OCs Heaven / Valley / Firestrike / Ice Storm / Cloud Gate / 3D Mark Vantage stable that failed under Furmark
-I have had Furmark stable OCs that failed under Heaven / Valley / Firestrike / Ice Storm / Cloud Gate / 3D Mark Vantage
-Every build since the 6xx series has had OCs fail in gaming that were stable under all of the above.

The Open GL nature of the benchmark is very useful as, like the box I am typing from, it serves as a A workstation during the day and gaming box "off hours". If you are using hardware acceleration in AutoCAD, Adobe Suites or other workstation apps, this is a worthy test.

So no I don't rely on it for stability but then again, nothing has proven 100% reliable. On the CPU end, I find RoG Real Bench to be my most reliable indicator over any synthetic application as it presents a suite of test programs in a multitasking environment. I can zero in my CPU OC's pretty close with a 2 minute run, tho don't get comfy till I pass 2 hours.

Nothing like that quite exists in the GFX world, at least in my experience. Using Heaven / Valley / Firestrike / Ice Storm / Cloud Gate / Vantage, I find that getting a higher score in one oft comes with a lower score in another. For example, at 1254 boost clock, I got a 10% higher score than I did at 1267 in Firestrike.

And while I use Furmark more for thermally setting up the TIM and cooling system than I do for stability testing, it does still occasionally find an OC that is not stable when Heaven / Valley / Firestrike / Ice Storm / Cloud Gate / Vantage says it is.....and also provides a sense of security for the workstation apps using Open GL for which the Gaming apps serve no purpose.

 

[Vellinious]

Do you type anything normal, or is it all this long drawn out story about how you use something?

TLDR. I have no time for you or your crap today. I'm done here. He wanted help getting his graphics card overclocked, and I helped him. If he were listening to you, he'd still be 2 paragraphs into your rant on something that has nothing to do with getting an overclock on a card. lol

Enjoy your furmark. rofl I'll spend my time gaming, cause Firestrike Ultra helped me find stable clocks. = )

Buh bye now

 

[JackNaylorPE]

- If ya got something that shows throttling on current cards then show us a link.
- If ya go something that explains how a 250 TDP card is being throttled, when it's drawing 359 watts "at stock settings" then show us a link.
- If you are going to state that no one uses a utility for anything, then don't complain when various uses are presented.

If one benchmark was the be all and end all of OC tweaking, the BF4 forum (for example) wouldn't be stocked with questions asking what's wrong w/ BF4.

 

[Vellinious]

Power limit throttling. Scroll up. Green in the perf cap reason line. Card throttles the voltage and the core clock to get the card back under the power limit. That would be the first picture I posted, explaining the "PWR perf cap reason", and what you can do to avoid it. /wink

I'm not complaining about anything. I merely stated that nobody really uses it anymore. Which, for all intents and purposes, I guess I should have said, "nobody that knows anything", because none of the guys that overclock use it for stability. As I stated previously, it's just a heat machine. That's all it is. And no...I don't count review sites like Tom's, TechPowerUp or Guru3D as guys that actually know anything. Most of them think a 1500mhz overclock is good...... Poor saps.

And as for stability.....that's why I said this. Had you taken the time to read the posts, you'd have seen it. Again, this is a case where scrolling up is useful.

"That clock likely won't be fully stable, depending on the game you're playing...some may like less clock, some won't mind the overclock at all. It's just a matter of figuring out which games you want to see a little more frames for, and finding settings that will work in each one.

There's always tweaks that can be done, but....sounds like you've got it dialed in pretty close. There ya go, congrats. = )"

Need anything else spoon fed to you there, shooter?