GTX 1080 volt issues past 1.25v

[Lays]

I was going to post this in graphics cards / tech news, but since the source isn't "concrete" I don't want to cause any issues.  If it needs to be moved somewhere else, that's alright.

 

 

So apparently the GTX 1080's that have been tested by extreme overclockers so far on Ln2 and other methods of cooling, have voltage "locks" at 1.25v, and it's not in the BIOS, or in the VRM design from the looks of it. Even an extreme overclocking BIOS with limitations removed still didn't solve the issue. I'm wondering (among others) that maybe there's some sort of limit to do with the architecture, or something else limiting the chip, whether it's the driver or something else.  As it stands right now, Ln2 on a GTX 1080 at 1.25v nets around ~2300-2400 mhz, which still isn't enough to stand up to a 980 TI on Ln2.    Considering 1.25v is still quite low voltage for even watercooling, it seems if we're lucky we'll only be seeing aftermarket cards & watercooled cards hitting ~2150-2250.

 

Essentially what happens when the GPU hits 1.25v is a driver crash & reset, which makes me wonder if it has something to do with the way the sensors / card communicate with the driver.

 

There's already a full fledged "uncork" guide on xdevs.com, and it seems some people that have even put "E-power" external VRM boards on it are still having this issue.  

 

Link to guide: https://xdevs.com/guide/pascal_oc/#intro

 

 

I'm very worried this may continue to be an issue, because some of the best hardware modding overclockers are stumped it seems in this thread:

 

http://forum.hwbot.org/showthread.php?t=157414

 

They've tested multiple cards already, and all seemed to have this issue. Hopefully we see a fix, otherwise the 980 TI will still reign supreme for extreme overclocking. 

 

 

Source: http://forum.hwbot.org/showthread.php?t=157414

 

For those who don't know, Der8auer is ranked #3 in the world on HWBOT.ORG, he's knows A LOT about the circuitry, and design & modification of PC hardware, he also makes Ln2 pots and sells them. 

[DXMember]

Overclockers dream: Part II

[Curufinwe_wins]

6 minutes ago, Lays said:

snip

Yea, I've heard this is an issue. I guess we will see if Nvidia addresses it. If not there won't even be much of a point for releasing extreme versions.

[Curufinwe_wins]

1 minute ago, DXMember said:

Overclockers dream: Part II

please tell me that is a Fiji reference....

 

Anyways, yea that's pretty disappointing. Additionally, it looks like core clock oc scaling is even worse than with Maxwell and the 1800 to 2200 oc level (along with 1k memory oc) is only like a 15% boost, well under what GM200 is capable of even on air from stock (this is pretty relevant in the 1070 v 980ti market, although obviously the 1070 is in a whole other league power wise).

[Lays]

5 minutes ago, Curufinwe_wins said:

please tell me that is a Fiji reference....

 

Anyways, yea that's pretty disappointing. Additionally, it looks like core clock oc scaling is even worse than with Maxwell and the 1800 to 2200 oc level (along with 1k memory oc) is only like a 15% boost, well under what GM200 is capable of even on air from stock (this is pretty relevant in the 1070 v 980ti market, although obviously the 1070 is in a whole other league power wise).

From what I've seen, almost every single review has used a titan x or 980 ti reference model to compare to the new cards, despite the reference models clocking to only like 1100 mhz..  Even though most aftermarket cards get close to 1400 out of the box, and 1500 with some tweaking, something we won't see from the 1080/1070 cards from the looks of it.

[Curufinwe_wins]

7 minutes ago, Lays said:

From what I've seen, almost every single review has used a titan x or 980 ti reference model to compare to the new cards, despite the reference models clocking to only like 1100 mhz..  Even though most aftermarket cards get close to 1400 out of the box, and 1500 with some tweaking, something we won't see from the 1080/1070 cards from the looks of it.

Not quite that bad (1300 or so most of the time stock boost), but the memory clock is also super low on the reference model (7000?).

 

Either way reference stock to aftermarket OC is like a 20-30% performance boost on the 980ti and it looks like the 1080/1070 is more like 15%.

 

Not to mention the insane levels done under  ln2 by gm200.

[Lays]

2 minutes ago, Curufinwe_wins said:

Not quite that bad (1300 or so most of the time stock boost), but the memory clock is also super low on the reference model.

 

Either way reference stock to aftermarket OC is like a 20-30% performance boost on the 980ti and it looks like the 1080/1070 is more like 15%.

 

Not to mention the insane levels done under  ln2 by gm200.

Really coz a few of my friends with ref models only boost up to like ~1200ish, (stock boost is 1076, so add like ~100-150 to that for actual boost if it's not throttling)

 

Either way yeah, the % gain of OC'ing a 980 ti seems to be much larger than a 1080/1070 right now.  Hell, I saw one comparison where someone was comparing like a 1450 980 ti to a ~2000 mhz 1080, and the 1080 was only like 10% faster, versus the 20-30% other people are showing.  And 1450 isn't really even the "high end" thats more of a mediocre OC

[Curufinwe_wins]

2 minutes ago, Lays said:

Really coz a few of my friends with ref models only boost up to like ~1200ish, (stock boost is 1076, so add like ~100-150 to that for actual boost if it's not throttling)

just looked back at it, my bad, looks like stock clock boost is closer to 1240-1180. Don't know what I was thinking.

[DXMember]

12 minutes ago, Curufinwe_wins said:

please tell me that is a Fiji reference....

 

Anyways, yea that's pretty disappointing. Additionally, it looks like core clock oc scaling is even worse than with Maxwell and the 1800 to 2200 oc level (along with 1k memory oc) is only like a 15% boost, well under what GM200 is capable of even on air from stock (this is pretty relevant in the 1070 v 980ti market, although obviously the 1070 is in a whole other league power wise).

yes that is a Fiji reference indeed, Jen Hsun talked about crazy world records under LN2 in the Pascal unveiling just like Lisa talked about crazy overclocks with Fiji (:

 

Can someone confirm that on aftermarket coolers 980ti and the new Pascal can maintain the boost clock? I've never been head to head with a GPU Boost 2.0+ solutions

Looking at OC3D, PCPer and GamerNexus review of the 1080 made me really concerned whether the GPU Boost technology is just a marketing stunt to crush benchmarks...

[Curufinwe_wins]

3 minutes ago, DXMember said:

yes that is a Fiji reference indeed, Jen Hsun talked about crazy world records under LN2 in the Pascal unveiling just like Lisa talked about crazy overclocks with Fiji (:

 

Can someone confirm that on aftermarket coolers 980ti and the new Pascal can maintain the boost clock? I've never been head to head with a GPU Boost 2.0+ solutions

Looking at OC3D, PCPer and GamerNexus review of the 1080 made me really concerned whether the GPU Boost technology is just a marketing stunt to crush benchmarks...

I can confirm on my shitty sub 70% asic MSI stock cooling TF5 980ti SLI I was able to hold constant 100% load clocks of ~1494 with dips to 1480. That with a 8000 mem oc.

 

But GPU boost 2.0+ really shows its balls on watercooling. Like lord all mighty does GM200 love to be cool.

[BuckGup]

9 minutes ago, Lays said:

Really coz a few of my friends with ref models only boost up to like ~1200ish, (stock boost is 1076, so add like ~100-150 to that for actual boost if it's not throttling)

 

Either way yeah, the % gain of OC'ing a 980 ti seems to be much larger than a 1080/1070 right now.  Hell, I saw one comparison where someone was comparing like a 1450 980 ti to a ~2000 mhz 1080, and the 1080 was only like 10% faster, versus the 20-30% other people are showing.  And 1450 isn't really even the "high end" thats more of a mediocre OC

Yeah my GTX980ti which is a great OCer. 1567mhz on the core and 2035 on the memory scored almost 700 graphics points better then 1080 in Fire Strike Extreme. But the chart they used the card got almost 4,000 points less then me

[Lays]

1 minute ago, SamStrecker said:

Yeah my GTX980ti which is a great OCer. 1567mhz on the core and 2035 on the memory scored almost 700 points better then 1080 in Fire Strike Extreme. But the chart they used the card got almost 4,000 points less then me

I know right?   When I saw benches of the 1080 I was like uhh... What..  Because I can bench on water at 1550|2250 in FSE, and I get quite a bit higher GPU score.   I think all this hype and publicity marketing stuff is overwhelming people into thinking the 1080 is way better than it is.  (Don't get me wrong it is very good)

[iamdarkyoshi]

You gotta wonder if they put some sort of voltage monitoring inside the die itself that causes the chip to force itself to crash. That would be a dick move

[BuckGup]

2 minutes ago, Lays said:

I know right?   When I saw benches of the 1080 I was like uhh... What..  Because I can bench on water at 1550|2250 in FSE, and I get quite a bit higher GPU score.   I think all this hype and publicity marketing stuff is overwhelming people into thinking the 1080 is way better than it is.  (Don't get me wrong it is very good)

Yeah you get it. Also the node shrink should have yielded much higher performance. But Nvidia likes to push things of then draw them out as long as they can so they make more money

[iamdarkyoshi]

2 minutes ago, DEcobra11 said:

A collaboration of AMWeed and Ngreedia

What about Shintel Shintergrated Shgraphics?

[Lays]

2 minutes ago, SamStrecker said:

Yeah you get it. Also the node shrink should have yielded much higher performance. But Nvidia likes to push things of then draw them out as long as they can so they make more money

I mean ya can't really blame em, they're a company after all and the goal is to gain as much capital as possible while maintaining good customer satisfaction (In theory lol)

 

From what I've seen, node shrink doesn't directly correlate to perf increase, rather just less power draw.  Usually when people are shrinking nodes, they're also putting a brand new arch on it which seems to be where the extra perf comes from

[DXMember]

3 minutes ago, Lays said:

I mean ya can't really blame em, they're a company after all and the goal is to gain as much capital as possible while maintaining good customer satisfaction (In theory lol)

 

From what I've seen, node shrink doesn't directly correlate to perf increase, rather just less power draw.  Usually when people are shrinking nodes, they're also putting a brand new arch on it which seems to be where the extra perf comes from

they can also get increased clock speeds where most of the Pascals performance comes from...

 

so what about the 1.25V lock? is it real or just in some peoples head? I think I just saw @Gofspar talking about 1.5V but now I don't see it... was it just in my head?

[Lays]

5 minutes ago, DXMember said:

they can also get increased clock speeds where most of the Pascals performance comes from...

 

so what about the 1.25V lock? is it real or just in some peoples head? I think I just saw @Gofspar talking about 1.5V but now I don't see it... was it just in my head?

Gofspar is trolling, the lock is real.

[App4that]

****, so we can expect this with the 1080ti too then I bet. Might have to pick up that second 980ti and skip a series after all.

[Bouzoo]

You know what I'm curoius about, seeing 1080 downclocked to the speeds on 980Ti, to see how they perform when they have the same clocks. I'll damned if 1080 doesn't do worse than a 980Ti. I'm not convinced Pascal is even as close as good as they praise it to be, so seeing it fight against Maxwell on same terms would be interesting for me personally. Has anyone done that? @Lays

[DXMember]

2 minutes ago, Bouzoo said:

You know what I'm curoius about, seeing 1080 downclocked to the speeds on 980Ti, to see how they perform when they have the same clocks. I'll damned if 1080 doesn't do worse than a 980Ti. I'm not convinced Pascal is even as close as good as they praise it to be, so seeing it fight against Maxwell on same terms would be interesting for me personally. Has anyone done that? @Lays

980ti has more stream processors

[boboman342]

1 hour ago, Bouzoo said:

You know what I'm curoius about, seeing 1080 downclocked to the speeds on 980Ti, to see how they perform when they have the same clocks. I'll damned if 1080 doesn't do worse than a 980Ti. I'm not convinced Pascal is even as close as good as they praise it to be, so seeing it fight against Maxwell on same terms would be interesting for me personally. Has anyone done that? @Lays

PC Per's interview with an Nvidia dude said clock for clock (shader per shader), pascal is the same performance as maxwell.

[DXMember]

2 minutes ago, boboman342 said:

PC Per's interview with an Nvidia dude said clock for clock (shader per shader), pascal is the same performance as maxwell.

pfail, who's rebranding now? Pascal = 10x Maxwell my ass

[Lays]

1 minute ago, boboman342 said:

PC Per's interview with an Nvidia dude said clock for clock, pascal is the same performance as maxwell.

That guy is a dam idiot then.

 

A 1450 mhz 980 ti nearly keeps up with a 1080 at 2100.  A 980 ti at 2100 on Ln2 would absolutely destroy a 1080.

Go find an overclocked result of a 1080 at ~2100 on firestrike, compare the graphics score to the one in my signature of my 980 ti at 1800 mhz, it's a big difference.

[Bouzoo]

4 minutes ago, boboman342 said:

PC Per's interview with an Nvidia dude said clock for clock (shader per shader), pascal is the same performance as maxwell.

I can't really believe that, 980Ti is breathing at 1080s neck with ~600MHz lower clocks. I'd like to see some actual results.