GTX 1080 "Founder's Edtion" Throttling Discussed

[-rascal-]

Many were surprised when the "Founder's Edition" GTX 1080's were in hand, and testing was underway for the full *official* reviews.

The 67*C operating temperature and 2100 MHz Boost Clock, seen at GTC 2016, was not exactly what was seen in reality.

 

In JayzTwoCents' most recent YouTube video, he discusses the (thermal) throttling issues the GTX 1080 "Founder's Edition" experiences while under a "gaming" load in Crysis 3.

 

Quote

...16 nm FinFET is a very, very small [transistor], and what that means is you are going to have a lot of focused heat, so the idea of 67*C under load was really kind of mind boggling...because with it being so small and focused, we expected the temperatures to actually be kind of hot.

 

At stock settings...

• 100% Power Target
• 83*C Temperature Limit
• No Voltage increase
• Stock Core and Boost Clock configurations

The card actually, both, Power Limit throttles and Thermal throttles. In fact, you can see it throttles all the way back down to *near* Base Clock frequency (advertised Base Clock is 1607 MHz, lowest Clock shown in video is 1620 MHz).

 

Once the game finished up loading, a few minutes of talking later, the Boost Clock has already dropped from the initial 1860 MHz down to 1810 MHz.

GPU usage was indicated at 95%, temperature of 78*C, usage of 99%, and with fan speed of 50%. A few seconds later, once the temperature creeped up to 80*C, the Boost Clock continues to decrease down to 1797 MHz.

 

He then follows up by increasing the Temperature Limit to 92*C, and Power Limit to 120%.

Quote

...and then I have another profile set-up -- no overclocking or anything; just Power Target, or Power Limit, and our Temperature Threshold...and what that will do is - I will increase the temperature from the factory setting of 83*C, which is where it will start to thermal throttle, and then we will raise that up to 92, and then I have the power limit set to 120 percent...so we can exceed the power enveloped designed by nVidia by 20% before it starts to pull back power due to power draw.

Once the Power and Temperature Limit profile was increased, the Boost clock immediately jumps back up to 1785 MHz.

The card jumps up to as high as 105% power, and 86*C temperature, while maintaining a relatively consistent 1785 MHz Boost Clock.

 

For the final bit of the testing process, he manually sets the fan speed to a fixed 80%.

The temperatures and Core Clock stays stable at 80*C and 1797 MHz respectively.

 

Quote

...people are making a big deal, again, because of that [GTC] demo 'cause the reality here is all of the reference cards -- in the history of...the reference cards -- have done this. This is nothing new. 
In fact people wouldn't  even be making a point it is thermal throttling if nVidia didn't make the point it is "only 67 C." So I do think there was a mistake made on that stage...whether if there was a bit of an expectation...*planted* in all of our minds, followed by a big...disappointment when it came to the reality of...it's still a reference cooler that cannot keep these temperatures in check, with the way they are shipped out-of-the-box. You got to go in there an "tweak" some settings.

 

Also keep in mind, Jayz's test system is on an open test bench. If you were to have the system enclosed in a full computer case / chassis, the temperatures are expected to be higher. He clearly outlines this in closing of the video.

 

Quote

Now remember, this was in an open-air test bench, which means the only cooling factory on the card, is the card itself; case flow was not a factor. So if you have a small-form factor case, or even a *regular* enclosure of some sort, you might even see thermal throttling happening sooner, and you might see [the Core Clock] come down even farther...depending on what your case flow is like.

 

To conclude:

1. The card operates at a much higher than *shown* temperature of 67*C.
2. Using the "Founder's Edition" cooler, the GTX 1080 is unable to achieve a 2100 MHz Boost Clock under a gaming load. 

It was clear nVidia was just attempting to stir-up some "chattering" by purposely showing 67*C and 2100 MHz Core Clock during GTC 2016. Of course, this is all part of "marketing" and you cannot blame them for it -- but it is very misleading, nonetheless.

 

And... additionally, more information regarding to this situation may be found on this thread post here: 

 

[Fulgrim]

Nvidia's GPU Boot is shit. ( ͡° ͜ʖ ͡°)

 

Pascal = Fermi 2.0

[TrigrH]

this video goes over it pretty decently:

 

 

[Daegun]

Not really sure what people were expecting, did they really believe that?

 

'Hey guys, you can get a 33%, 500mhz overclock on this stock blower cooler while being below 70C."

 

Edit: I have no doubt that aftermarket cards will be able to at least hit 2.1ghz, we've seen maxwell have some amazing overclocks, but to think it would do so AND do it at 67C on the stock blower cooler is buying into Nvidia's marketing too much. It should have been obvious that average cards wouldn't be able to do that precisely because that's what was shown, if the average card could, they would have shown something more impressive.

[Hunched]

This wouldn't be news if people would quit being retarded enough to actually listen to what Nvidia says about their products.

Anyone who is surprised by this I'm sorry to say is an idiot.

Since when do reference coolers with 1 fan provide good cooling again? Oh literally never?

If only there was a recent example of this happening. Hmm... *cough* Titan X *cough*

 

People keep closing their eyes and wiping their memory to have unrealistic optimistic expectations or what?

Reference coolers are shit. Welcome to GPU 101 guys.

[sterlingr]

Just now, Hunched said:

This wouldn't be news if people would quit being retarded enough to actually listen to what Nvidia says about their products.

Anyone who is surprised by this I'm sorry to say is an idiot.

Since when do reference coolers with 1 fan provide good cooling again? Oh literally never?

If only there was a recent example of this happening. Hmm... *cough* Titan X *cough*

 

People keep closing their eyes and wiping their memory to have unrealistic optimistic expectations or what?

 

 

20 minutes ago, -rascal- said:

*Clipped*

 

 I'll be interested to see how 3rd party cards deal with this... 

[Hunched]

2 minutes ago, sterlingr said:

 

 I'll be interested to see how 3rd party cards deal with this... 

It's simple.

They'll put a good cooler on it, one that isn't closed and has double or even triple the fans, and it will cool it well unlike the reference cooler.

Like how its been since the beginning of time.

[PlayStation 2]

Why are you surprised.

[stconquest]

10 minutes ago, Hunched said:

It's simple.

They'll put a good cooler on it, one that isn't closed and has double or even triple the fans, and it will cool it well unlike the reference cooler.

Like how its been since the beginning of time.

...along with a more robust VRM (instead of the 5 phase one they have now) and add a additional 6/8 pin connector.

 

2.5GHz here we come!

[-rascal-]

25 minutes ago, Daegun said:

*snip*

 

21 minutes ago, Hunched said:

*snip*

 

12 minutes ago, Dan Castellaneta said:

*snip*

 

I'm not so much surprised this was the situation, but the fact that people are thinking 2100 MHz is possible on reference cooling all over the god damn web is just ridiculous. Considering how the the reference GTX 1080 is able to reach close to 1900 MHz, I am not at all surprised a custom cooled card could reach 2100 MHz or beyond.

 

That said, EVEN with an aftermarket cooler, people need to be realistic.

IMO, I don't 110% think if something like an ASUS STRIX, MSi TwinFrozr, or even a EVGA ACX 2.0 AIR cooler can do the hyped "2500 MHz or more." 1607 MHz to 2500 MHz is freakin' ~900 MHz.

[PlayStation 2]

1 minute ago, -rascal- said:

 

 

 

I'm not so much surprised this was the situation, but the fact that people are thinking 2100+ MHz is possible all over the god damn web is just ridiculous.

Considering how the the reference GTX 1080 is reach close to 1900 MHz, I am not at all surprised a custom cooled card can could reach 2100 MHz or beyond.

 

That said, EVEN with an aftermarket cooler, people need to be realistic -- IMO, I don't 110% think if something like an ASUS STRIX, MSi TwinFrozr, or even a EVGA ACX 2.0 cooler can do the hyped "2500 MHz or more." 1607 MHz to 2500 MHz is freakin' ~900 MHz.

I'm not hyped for the 1080 at all. Honestly, the Founders Edition BS killed the card permanently for me.

[Hunched]

19 minutes ago, -rascal- said:

the fact that people are thinking 2100+ MHz is possible all over the god damn web is just ridiculous.

That's what happens when you have cunts who care nothing but to report what Nvidia says as if their claims are fact before anyone can test it to verify, just so they can get their clicks and their money for their "reporting" and "journalism" copy/paste.

And all the idiots on forums who also further spew the likely invalid information everywhere because they have nothing better to do than to reassure their alliance with a team.

 

It's funny how when consoles or whatever come out with claims of performance everyone puts their skepticism hats on, but when Nvidia says their shit can do X, everyone equips their blind following glasses.

Then they're in shock and awe when for the 100th time they didn't tell the truth.

 

The cycles are just... really annoying. 

It would be nice if people could learn.

"oh my god, a single fanned hotbox doesn't cool well. This defies my grade 1 level understanding of heat transfer!"

"oh my god, a company said their product was better than it actually was. That's impossible"

 

I mean, they're so stupid they're willing to pay $100 more for it too, and preorder them to the point they sell out of Amazon asap.

Literal sheep.

 

Maybe they should charge $200 more for the 1180 and ship it just PCB, no cooling of any kind.

Super Cool Kid Edition.

Cooler sold separately. 

People would buy it.

[Carclis]

I think we all knew that the card would not be hitting 2100 core clock at 67C on a reference card. The main concern here is that the card thermal throttles in an OPEN test bench, the most ideal situation, down to it's base clock. Of course reference cards suck (or blow) at cooling compared with custom solutions, but it would be nice to see them actually run as intended under reasonable circumstances. Having a boost clock which kicks in sometimes and reverts back to normal speeds because of thermals is not ideal or desirable and I hope nothing like this or the 290x reference ever happen again. It's bad publicity and results in lower benchmarks. I think the best move by Nvidia would be to ditch the boost clock for an achievable base clock so that people get a consistent experience.

[Misanthrope]

I think you cannot stress this enough: they expect you to pay 100 extra for the privilege of a crippled out of the box product.

[Bouzoo]

Well if this is to be believed, power throttling has been solved. Either my eyes deceive me or those are 2x8 pin connectors (lower one looks like 8+6?).

 

 

[DXMember]

built of premium components my ass

[RagnarokDel]

5 hours ago, Hunched said:

This wouldn't be news if people would quit being retarded enough to actually listen to what Nvidia says about their products.

Anyone who is surprised by this I'm sorry to say is an idiot.

Since when do reference coolers with 1 fan provide good cooling again? Oh literally never?

If only there was a recent example of this happening. Hmm... *cough* Titan X *cough*

 

People keep closing their eyes and wiping their memory to have unrealistic optimistic expectations or what?

Reference coolers are shit. Welcome to GPU 101 guys.

except that they intentionally mislead people into thinking it would be cool

[3DOSH]

Since when do reviewers set the fan speed manually when they benchmark a GPU without disclosing it ? specially at launch when hype is at all time high.
 

[Big-G]

6 hours ago, -rascal- said:

....

 

Yes, but what you also need to remember, they only ran a render sequence at 2.1GHz and 67c... They never gamed it at those settings.

 

we all jumped to conclusions... smart marketing 

 

6 hours ago, Hunched said:

It's simple.

They'll put a good cooler on it, one that isn't closed and has double or even triple the fans, and it will cool it well unlike the reference cooler.

Like how its been since the beginning of time.

And add modified BIOS's

 

1 hour ago, Bouzoo said:

Well if this is to be believed, power throttling has been solved. Either my eyes deceive me or those are 2x8 pin connectors (lower one looks like 8+6?).

8+6 pin, that's an SC edition... Only the FTW and Classified and Kingpin will most likely see 8+8 pin

 

56 minutes ago, DXMember said:

built of premium components my ass

Premium PR components yes...

 

20 minutes ago, RagnarokDel said:

except that they intentionally mislead people into thinking it would be cool

Actually no, they said nothing of gaming, they just showed that it could run 2.1GHz at 67c without showing the power, thermal targets and fan profile... It was also running a simple render... No gaming. We all made assumptions...

[Bouzoo]

12 minutes ago, GidonsClaw said:

8+6 pin, that's an SC edition... Only the FTW and Classified and Kingpin will most likely see 8+8 pin

Personally I hope there is no need for a 8+8, they've been talking about low power consumption and have been hyping it on that, but if it has to have a 2nd 8 pin to max it, then that is a disappointment. 

[jaggysnake57]

dont know why people think this is a big deal, the founders is overpriced and hot.  oh well ill just have to get an AIB card.......again

Edited May 27, 2016 by Godlygamer23
Clean up.

[Big-G]

11 minutes ago, Bouzoo said:

Personally I hope there is no need for a 8+8, they've been talking about low power consumption and have been hyping it on that, but if it has to have a 2nd 8 pin to max it, then that is a disappointment. 

Nope, it means those cards are focused on extracting maximum performance... Watts be damned!

 

 

[Bouzoo]

5 minutes ago, GidonsClaw said:

Nope, it means those cards are focused on extracting maximum performance... Watts be damned!

As I've said, If a card that is supposed to be very efficient, for a flagship, needs that much power to max it, then that's for me disappointing. But as always, that's NV marketing, can't say I'm surprised by that. It's actually my fault for believing for a second 1 8 pin is gonna be enough.

[Lethal Seraph]

Sounds like the batmobile shroud cooler lives up to the reputation of batman:AK. Lel