The Finfet Wall

tl;dr

File: 900x900px-LL-e423364a_1060Results.png (41KB, 769x344px) Image search: [Google]

41KB, 769x344px

There is a thread on overclockers found here
http://www.overclock.net/t/1605348/bios-hardware-voltage-lock-is-preventing-gtx-1070-from-reaching-1080-performance
that states "Hypothesis - ALL GTX 1070's are either BIOS or Hardware voltage locked to not go beyond 1.0930v; as a result CORE clocks are limited regardless of whether you're on water or air cooling within a certain region (between 2000MHz and 2150MHz in most cases), preventing those with any GTX 1070 to achieve results similar to a non-overclocked GTX 1080."

Seems pretty plausible, lots of discussion in that thread. Seems no one there knows.
Here is the thread with the hard moded 1060
http://www.overclock.net/t/1616678/gtx1060-extreme-mod-log-come-watch-me-bench-the-blue-smoke-out-of-a-gtx1060

So in theory the colder the circuit the more conductivity it has since the material has a higher heat capacitance. If you can cool the circuits to -10 or lower you should be able to get more power through them resulting in higher clocks.
This is an interesting thread though since obviously high MHz like this is not impossible with the materials used so it's either a power draw issue, NM issue or designed to have this limit.

File: Galax GTX 1060 HOF GOC.jpg (56KB, 590x332px) Image search: [Google]

56KB, 590x332px

>>58460215
Yes subzero cooling does change a lot and i'm not educated enough on the topic to fully understand all of it. It bothers me personally because like I said my 1060 under water hits 50c absolute max at 2150mhz rock solid but 1mhz more and it's unstable as fuck. The hard mod pictured in OP eliminates power from the equation because that thing is an external VRM which is insane and enough to power anything. That same guy also used some sub-zero method and it barely got him anything.
So he removed power and temperature from the problem. That leaves us artificial bios based limitation or architecture. The galax HOF 1060 set a world record at 3ghz so maybe they were using a special bios to be able to do this. With volta on the horizon however what does nvidia have to gain from artificially limiting their card's potential?

File: image.jpg (297KB, 780x591px) Image search: [Google]

297KB, 780x591px

>1070 at 2167MHz

>>58460071
>>58460159
It would be nice if nVidia stopped doing shit like this...
The main reason I buy AMD is because of the open nature of their platforms.
No artificial limits on hardware access or performance.
Good, open source driver contributions.
Innovation in hardware, not just optimizing existing architectures.

Hopefully, Volta will bring something new and exciting to the table. Pascal was a bit of a let-down for me as they didn't really add much to the Maxwell architecture, and it's still basically the same as Fermi at its core.

>>58460333
You're still with in the frequency range I'm talking about. Dismount the air cooler and hold an AIO on the GPU and see if you can go to 2200. I bet you that it won't work even with the core at 45-50c. Which model are you using?
>>58460342
The rx470 is interesting because it scales well with clock speed as it has less cores per ROP and so more bandwidth per core. I'd like to see what one's potential would be with a good power delivery system and cooling. However, polaris seems to be near it's limit where as a card like my 1060 seems like it has so much more frequency to give but something is getting in the way. Regardless of temperature and power delivery.

>>58460071
Intel has been locking their CPUs for a long time and no one complained. Come on guys grow up.

>>58460373
I can make my 1070 immediately crash at 20° if I wanted, normal operating temps isn't what makes cards crash
Also mine hits 75° under benchmarks and it's the cheapest 1070 on newegg

File: a498e19ea036d0243de1b40697ca62b9.png (393KB, 448x567px) Image search: [Google]

393KB, 448x567px

>>58460379
But at least they tell you they're locked. With this limitation, that has been proven true, it doesn't matter what pascal card you buy you're only going to hit 2150. The EVGA gaming which uses air cooler in the picture hits this clock at 75c with like 3 or 4 power phases. If you spent more on a card with 10 power phases and a cooler 5 times the size you would get the same overclocking potential. It comes off as misleading and crooked.

>>58460373
The 1060 doesn't have more frequency to give, they already boost very high
This has been a thing since pascal came out, they're not great overclockers
As to why, that's just Internet memes at this point

>>58460071
>Is nvidia purposely holding back their cards with the intent to make more profit through a pascal rebrand, or possibly new "ti" versions, or is it an inherent flaw in the current 16nm TSMC/14nm samsung processes being used?

You couldn't speak over the entire process, only the specific Vth used, every given process has numerous variants that target different power levels and frequency ranges. It could also be that Nvidia limits max clocks to a certain level to keep logic synced to their cache, or at least to maintain a certain ratio that the cache can't scale past. Its also possible that they expected heat density in certain blocks to be too high to adequately cool beyond certain clocks modeling average volts per clock.

Theres not enough information attainable to make a definitive statement one way or the other unless you're inside the company.

>>58460379
>no one complained
You have an odd definition of "nobody"

>>58460379
This isn't the same thing, it's an artificial limit in the way of getting more performance for your dollar.
You don't see these limits on the 1080's, and the silicon is the same. Why shouldn't users be allowed to push their card as hard as they want? nVidia allows overclocking on all of their cards, so you should be able to push them as far as they go.
It's likely that the voltage limit is a safety measure imposed by the BIOS, but that doesn't mean that there shouldn't be an unlimited BIOS version available to those who want to use it.

>>58460420
Thank you. Can you link me something I can read to better understand what you're talking about in regards to syncing "logic" and cache, or expand on it yourself?

>>58460429
Nvidia doesn't advertise overclocking when they sell their GPU chips to manufacturers, so nobody actually owes you jack shit past the boost clock

>>58460439
Can't you go troll on /b/ or something

>>58460429
>silicon is the same
The 1060 uses a GP106 GPU and the 1070/1080 use a GP104 GPU
Also the first result of "1070 overclock" shows a guy getting 2.2GHz

>>58460452
But it's true, do you ask Intel for a refund when your CPU is only stable at 4.5 instead of 4.6? And Intel explicitly sells and prices the ability to overclock
The whole point of overclocking since the beginning was that it isn't guaranteed for any product ever

>>58460484
And yes, that is a thermally-limited 2.2GHz, the cooler doesn't get dangerously hot but there are some hotspots on the board that apparently were too big of an issue to go further

>>58460434
Whether you're talking about a CPU, or a GPU, everything doesn't necessarily have a single clock speed. In a CPU the core logic, and uncore can operate at different clock speeds. This means your 4ghz core i7 whatever can have its caches running at 2ghz. Delving even deeper down specific units inside of the core can be clocked at different rates, but thats unimportant.
Depending on architectural nuances the memory may only work at certain frequencies, whether it is thermally limited, or there are pipeline limitations. IE: performance may benefit the most from having uncore clocked at a specific frequency, or a ratio compared to logic's clock. Like 2:1. 1:1. 1:4 or etc.
Increasing uncore past certain levels may harm total performance in some cases unless you increase core logic, and increasing core logic clocks will yield limited uplift unless you also increase the uncore.

AMD's GCN arch for example has the core clock speed really tied to the speed of the L2.

>>58460071
Is that a reference board or a GPU that was overclocked as high as possible, and then had a card and cooler built around it only allowing the power needed for the SUPERCLOCKED 1060?

>>58460071
and?
it doesn't matter, they dont overclock that well anyways except for the vram

>tfw running 600mhz overclock on my vram


plus the is put into place so you don't fry the thing

>>58460631
You're completely missing the point. Feel free to read when you get the time.

>>58460617
Wow it's almost like overclocking is a variable process and limits can vary chip vs chip and also brand vs brand
Sorry you can't beat the Internet in clock speed, at least I did

>>58460639
I read it, and don't really care
Nvidia locked the voltage to a certain point. why? probably to protect their higher end cards and not to have another 970 where it can overclock to be as powerful as a 980

>>58460612
It's a low end GPU because he attached his own power delivery system.
>>58460695
You're jumping to the conclusion that this thread was posted to find. You're assuming they locked the voltage. We're trying to see if they actually did or not. There are modded cards with full control over voltage and it doesn't change much.
>>58460657
It doesn't though. ~2150 no matter the cooling or power delivery

Nvidia shills feverishly moving goal posts all over the fucking place.

>>58460743
And? You paid for 1600-1800MHz didn't you? Isn't that what evga (not nvidia) promised you?

>>58460743
they did lock the votlage, you can only increase it to a certain voltage and thats its limit

I can't remember its max voltage but you can't push it further than that

>>58460770
Completely irrelevant to the question in OP.

>>58460743
I just said I got 2167 on a 1070 and the first Google review hits 2.2GHz
Online almost everybody is claiming 2000-2200mhz, that seems like a pretty wide range of normal max clock speeds, especially considering the 10-- cards were never great overclockers
The more I read your posts the more I get the feeling that you're mad you can't hit 3GHz with your 1060. For the almost year before that Galax card hit 3Hz nobody could go much higher than the "normal" overclock, regardless of who made the card, what power was on it or the cooling
Galax has the anomaly here, it's not some conspiracy

my 1060 doesn't even hit 2050mhz
Thread debunked

>>58460071
>wasting all these precious capacitors on a fucking 1060
>wasting his EE talent on some overclocking bullshit

Why do people do this?

File: S7EYADx.gif (188KB, 550x400px) Image search: [Google]

188KB, 550x400px

I have the Asus Strix 1070 Gaming. The only problem I've running into is hitting the Voltage and Power limits. Card never gets hotter than 67* but I can only get about +95 on the core and +250 on the memory. Thats after the stock factory OC. Never been able to get more than 2110 on the core.

It seems the limiting factor on the Strix is the single 8-pin PCI power connector.

desu, 2150mhz overclock barely gives any performance gains and can be very unstable

>>58460906
He can reuse it, and because it's fun.

>>58460801
>is nvidia holding back their cards to release a refresh of [architecture] or to release [same card]ti?
Yes. They have for decades, AMD does too, Intel does with their processors
This is not related to your 1060 not performing like an overclocked 1080 like you want it to but simple marketing so retards like yourself would buy a 1060
The 1060 6GB is not a cut down 1070 or 1080 though, it's just a worse chip and a worse gpu
Sorry you bought it but you are NOT hitting 3GHz with it and I'd be surprised if you even hit 2.1GHz

>>58460925
It's going to be a bitch to desolder those SMD capacitors and not wreck the pins on them in the process.

>>58460910
>tfw dual 8 pin
though I doubt it helped at all but my ftw 1070 can do a stable 600mhz memory boost and 150mhz core boost

>>58460943
People like you are the bane of this board.

>>58460972
for you

>>58460071
>All of the pascal line hits a wall at ~2150mhz clock speed

>my personal card will hit 2150 at 45c but instantly crash at 2151

So which one is it? Because you clearly prove yourself wrong here >>58460159

Different arch tops out at different clocks. Hawaii tops out at 1200 on air, Fiji can't oc at all, maxwell does 1600 at best.

>>58460972
Bane?

File: image.jpg (56KB, 576x694px) Image search: [Google]

56KB, 576x694px

>>58460159
>preventing 1070 from reaching 1080 performance
You have a 1060 though

>>58460215
>>58460312
For what its worth colder transistors also have less leakage, though leakage of FinFET devices is already very low.

>>58461049
What are you talking about? The "wall" I'm talking about and all those people in those threads I linked are talking about is ~2150. My water cooled 1060 hits this wall with an AIO on it.
You see the clocks go up marginally in that log picture from the insane modding the guy did and then the sub-zero cooling. You also see this wall broken from all the world record attempts on the 1060.
The hard mod improved the power delivery system over the stock 4(?) power phases. Other cards with more than 4, or in the 1070/1080 case like 10, also hit the wall though.
This isn't just my idea, I knew there was some kind of limit on my 1060 but that's all. There are a bunch of much more knowledgeable people discussing this same thing, I just wanted to see if anyone here had anything useful to say.
>>58461093
Completely irrelevant. Like I told the other sniveling and brain dead faggots this is just a thread about why all of the pascal cards don't want to go past this limit.

What are the implications of this?

Are future cards going to hit some kind of wall which makes FinFET shitty?

>>58461279
I wanted to find out but apparently I just posted this thread because I own a 1060 and am thus a buttmad poorfag.

>>58461212
But many do? And they're outliers
Like someone said what are you expecting, everything with a clock has a clock limit and that's heavily based on die size and architecture, cooling really has little to do with it in comparison which is itself based on die size and architecture

This is fucking cool.

>>58460342
>Innovation in hardware, not just optimizing existing architectures.
I'm an AMD fan but really

It's been the same shit since the 7000 series

>>58461283
The fact that you strapped an aio to it expecting to gain .9GHz says more than your purchases, and the fact that you lump 3 completely different gpus into you 1060 problems for no reason says a lot too
2150 is good, what makes you think it's not a real limit anyways? Because it's not boiling water?

>>58461212
Some factory 1070/1080 boost well over 2150. Just what the hell are you talking about?

It sounds like youre whining because you think you're entitled to some insane oc

>>58461395
Once again, it's not just me. Threads linked near the top of the thread discussing it. I love how you morons can't resist the urge to make personal attacks simply because nvidia was mentioned.
>>58461401
No they don't.

You bought an expensive aftermarket cooler expecting to get big overclocked and found out it's not happening because of hardware power limits. Now you're upset that you spent all that money for nothing.
Return the cooler and move on. There is literally nothing more to know or understand.

>>58461431
>Other people are all hitting the same limit with the same GPUs like always
>Pascal card hits double the frequency of last gen card
>what's going on there's no such thing as a limit galax told me so

I'm going to make it a point to respond to all of the bait.
>>58461455
The AIO I used is 5 years old and has been sitting in my closet for the last 2. Once again there are a ton of people discussing this wall, I thought it would be interesting to try the same discussion here as once in a while there is a smart person here. Today he seems to be missing though.

>>58460071
>Is nvidia purposely holding back their cards with the intent to make more profit through a pascal rebrand, or possibly new "ti" versions, or is it an inherent flaw in the current 16nm TSMC/14nm samsung processes being used?- 59 posts and 7 image replies shown.
absolutely

they are a scummy company

File: Screenshot_20170112-214505.png (679KB, 1440x2560px) Image search: [Google]

679KB, 1440x2560px

>>58461431
>No they don't.

File: galax_geforce_gtx_1070-single-slot_678_678x452.png (364KB, 678x450px) Image search: [Google]

364KB, 678x450px

>>58460071
So single slot cards are back knowing thermals aren't the limit

File: image.jpg (155KB, 640x932px) Image search: [Google]

155KB, 640x932px

>>58461495
But why is it so hard to believe that it's a legitimate limit?
Its nearly double the GHz of the maxwell cards and has almost the same kind of overclocking headroom as previous cards, why are you so sure that pascal is supposed to be faster than the 2150mhz area?
>pic related, a 7GHz processor that Intel fucking limited! I can't even get 5GHz on my 6700k and neither can anyone else!

>>58460121
Kys

>>58461536
>hurr 30 mhz this thing that tons of people have confirmed is false because of my own personal brand loyalty.
End your life as soon as possible.
>>58461547
Yeah that thing can literally overclock to the same point that a 3x120mm rad AIO 1070 could.

File: image.jpg (49KB, 512x288px) Image search: [Google]

49KB, 512x288px

>>58461714
So do you think freezing your electronics should allow you to overclock them until you're limited by the speed of light?

File: image.png (248KB, 1132x1000px) Image search: [Google]

248KB, 1132x1000px

>>58461714
You bought a fucking watercooled 1060 didn't you
Is that what this whole thread is about?

File: Nvidia GTX 1060 world GPU frequency record.jpg (169KB, 590x332px) Image search: [Google]

169KB, 590x332px

>>58461769
I don't think one of those exists.
>>58461752
Kind of, yes. Pic related

>>58461782
Yeah I know galax told you there was no limit and 6700ks can't get above 5GHz, it's an outlier, an anomaly, professional overclocking hardware, power, mods and cooling designed for the dickwaving novelty, it doesn't mean that all 1060s must now reach 3GHz

>>58460414
>It comes off as misleading and crooked.

Shame on you for uttering blasphemous words against Nvidia.

They power efficient anon, don't lose you ways and go binge watch JayZ and read at least 50 articles on Tom's Hardware or Anandtech until you mend your ways.

>>58461927
I'm so sorry!
The rx 480 is bad bad bad bad you should buy a titan x pascal

File: made to goy.png (194KB, 603x802px) Image search: [Google]

194KB, 603x802px

>butthurt fanboys shit up a legitimately good technical thread
Shameful.

Stop bumping op
start googling

>>58462095
OP sounds like a entitled cunt because his card can't go past a certain frequency

>>58460342
>Good, open source driver contributions.
>Good
nigger, I hate nvidia and their shenanigans as much as the next guy but you are delusional if you truly believe that they make good open source drivers, they are not innovating much as they are failing to deliver time and time again. I do hope this future release isn't all smoke and mirrors like before. I'm getting tired of people hyping shit time and time again.

Well all 1050 Ti cards are capped at 1911MHz. So OP is correct, Nvidia does do that.

>>58461532
and amd gimped the cache of the 845 so they could claim gains in their 7th gen carrizo apus

>>58460071
>Nvidia being maximum jewish
WELL COLOR ME SUPRISED.
Its's a cancer company for cancer gaymurs.

>>58461559
Are you trolling or really fucking retarded?

>>58460071
>or is it an inherent flaw in the current 16nm TSMC/14nm samsung processes being used
What the fuck are you talking about?
There are mobile CPUs built on this very process that clock at over 2.5GHz at a few watts.
There is Zen on the same 14nm Samsung/GloFo node that reaches 4 GHz easily.

Before you start to shitpost, at least do a tiny bit of research.

>>58460215
>>58460312
The whole point of liquid nitrogen overclocking is to make the chip more conductive not to cool it while it's running.

>>58461547
yay i cant wait for fans that go VVVVVVVVVRRRRRRRRRRRRRRRRRRRRRRRRRR again, used to own a gt8800, fuck that noise, give me proper quite cooling fans

>>58460342
You're posting like you're stuck in the 2005/2007 AMD golden age.

>>58460071
>Is nvidia purposely holding back
yes. both amd and nvidia have been doing so for almost ten years now.
they're trying to let shitty laptops catch up to desktops to remove normies from desktops
can easily see this with cpus that haven't progressed in years now.

>>58460502
just go to /b/ bud, this isn't the forum for you

>>58465745
AMD golden age is 2001-2005. In 2006 intel stomped them with core 2.

>>58460071
6870 for example has a BIOS hard lock in the firmware, refuses to boot and near instantly crashes any core clock which surpasses 1000mhz

haven't found anyone who managed to get a workaround for it, which is sad considering how there's an insane amount of headroom both voltage and cooling wise which would easily allow faster clocks.

there's been random shitposters claiming they can go over the limit, but it's just shitposters shitposting bullshit claims without any evidence or explanation of how they surpassed the firmware 'wall'.

>>58460312
>With volta on the horizon however what does nvidia have to gain from artificially limiting their card's potential?
the fact that people wont be able to easily overclock their $300 gpu to their overpriced underpowered $800 gpu.

>>58460071
There's really only a few options to explain this behavior. I'll list them in the order of most likely although they are all VALID reasons Nvidia would do this, and could be a combination of all these things.

>1) Founder's Edition was a cashcow for Nvidia.
If it turned out FE cards vastly underperformed AIB cards because of overclocking, they wouldn't be able to cash in on it as much.
They want early adopters to pay a premium and allowing AIB cards to blow past them in performance would shoot this money grab in the foot.
Realize I'm not saying FE performs the exact same as AIB, we all know it performs worse, but this "limit" keeps it closer and rationalizes paying the premium.

>>58465916

>2) Keeping the 1060 in the 60fps/1080p bracket, keeping the 1070 in the 60fps/1440p bracket.
The 1060 and 1070 are matched to perform at those fps at those resolutions.
If you pay attention to benchmarks overall it's clear Nvidia designed the cards to perform exactly in those brackets with a bit of wiggle room.
This easily incentivizes people to move up a tier for performance at certain resolutions, and a completely unlocked chip could damage those price points.
This is not some bizarre theory, Nvidia has been banking on higher tiered cards becoming a larger source of income in the past few quarters, which their CEO has repeatedly mentioned in earning calls, they have a clear financial motivation to do this.
If you look at the performance reported from the 1060 hitting 3Ghz, the overclocked 1060 exceeded pixel fill rates of the GTX 1080, meaning the realistic limit even for a fully unlocked AIB OC could have placed the card in the 60fps/1440p range thus invading on the higher priced 1070's market.
http://www.tweaktown.com/news/55521/galax-geforce-gtx-1060-hits-3ghz-gpu-sets-new-record/index.html
Extrapolating this performance gain to something like the 1070 which has a wider memory bus could have placed it well within the market the 1080 was meant to fill of ABOVE 60fps/1440p.
No matter how you look at it, Nvidia would lose money from allowing their cards to perform outside of the anticipated performance expectations since they are only competing with themselves at the high end.

>>58465927

>3) Priming the market for a Pascal refresh or Volta.
It had been unsure if Nvidia actually ever wanted to jump to 10nm because of the poor improvements of the process, and it would help them out to have wiggle room if AMD's Vega exceeds their expected performance.
This market strategy has worked out for Intel when they can't jump down to a smaller node and Nvidia has definitely considered doing the same thing since they are a business who pays attention to these things.
(This next part may be a little bit more personal opinion.)
Essentially, Pascal doled out the ever slightest acceptable margins of expected performance at the aforementioned resolutions, and the consumer market ate it up.
Seriously, the 1060 over an average of most recent demanding games only gets an EXACT average of 60fps in 1440p and drops below this in most of the titles.
https://www.youtube.com/watch?v=3Jml0uItdnE
It's blatantly clear Nvidia is planning the obselence of their own cards at this point and limiting how much extra fps consumers can pull out of their cards to set up buyers for next year.
Because remember guys, Nvidia is only competing with themselves in the high end of the market, and they need their customers from last year to be customers this year too.

>>58465771
not AMD tho

>>58465682
WRONG
the point is to keep it so cold it wont fry itself

>>58466713
>The carrier mobility describes the drift velocity of a particle in an applied electric field. MOSFET mobility has very complex temperature dependence, defined by the interplay of the following four scattering parameters: phonon scattering, surface roughness scattering, bulk charge Coulombic scattering, and interface charge Coulombic scattering mint. Each of these scattering parameters is related to the temperature of the material, and the effective transverse electric field in the channel.

File: galaxyoc2.jpg (237KB, 1000x563px) Image search: [Google]

237KB, 1000x563px

Good morning, OP here. Still an anally ravaged poor fag.
>>58465936
>>58465927
>>58465916
I like you. You are right with the tailored to resolution thing. They do preform almost exactly where they are "supposed" to, even though the marketing material doesn't directly state that as their goals. I don't think it does anyway. The FE theorey was a consideration for nvidia, I'm sure, but in reality it's mostly like a combination of multiple things that lead them to lock the cards somewhere in software. That's just my opinion of course. To be fair though the cooler design of the FE cards looks great and to normies it probably looks like some shit from the future or a spaceship or whatever.
Someone in another thread last night mentioned the 760ti MSI lightning, which didn't hit the market or was very delayed to market because nvidia didn't like that it clocked so high.
http://www.tweaktown.com/news/38314/msi-gtx-780-ti-lighting-the-gpu-you-can-t-buy/index.html
Maybe this is why you are unable to buy the 1060 HOF from galaxstore even though they just spent all that money on the world record publicity stunt.
>>58466858
>>58466713
>>58465682
It's probably a combination of both right?

>>58460071
they just don't want your computer to blow up you fucking idiot

File: 447.png (398KB, 833x696px) Image search: [Google]

398KB, 833x696px

>>58466903

>>58460159

What sucks most about the locked voltage is, that the shit tier Micron memory requires higher voltage to clock better.
Since the voltages are locked, people who didn't get Samsung memory are stuck with less room for OC.
I firmly believe that they do shit like this to prevent the 1070 to hitting stock 1080 levels, which it already nearly hits anyways.
Remove the limits and 1070 would go higher than the stock 1080, which would naturally kill a big portion of the cards sales.

File: amd4850-1.jpg (128KB, 800x600px) Image search: [Google]

128KB, 800x600px

>>58466985
Even with voltage completely unlocked the 1060 wouldn't hit the levels seen in the world record attempts with out sub-zero cooling but that said the world record OC had a higher pixel fill rate(the fuck does that even mean) than the 1070. Every 1060/1070/1080 can clock higher than they are being sold, and far higher if the limit wasn't there. I think nvidia could increased the performance of every single pascal card, thereby preserving their tier's of performance. Maybe it's just to give them some breathing room in case AMD dropped on a bomb on them out of fucking no where.

>>58466985
Core voltage doesn't control vram voltage idiot

>>58466893
The thermal resistance of the circuit itself is what makes the temperatures higher, however. The circuits become more efficient at lower temperatures meaning higher clockspeeds at lower temps. If it was ONLY about dissipating heat to below tjunction then a massive water cooler would get clocks similar to a liquid nitrogen run, and quite obviously, the water cooling doesn't.

File: 4bc1bbc9ffda57fd735aad8d31bde36fbeeeb8422b1747feb79ed8ca0ddb5e4d.jpg (59KB, 520x678px) Image search: [Google]

59KB, 520x678px

>>58460159
No way NVIDIA would pull another 3.5... Right?

File: NVIDIA-Pascal-GP104-400-A1-GPU-1[1].jpg (469KB, 1081x891px) Image search: [Google]

469KB, 1081x891px

http://www.overclock.net/t/1613644/the-definitive-gtx-1070-micron-memory-thread
Good read about the micron memory.
"The published datasheets for both Samsung and Micron, where they have comparable specifications, both show identical 1.35-1.5v voltage requirements and other physical specifications. The exception to that statement is that Samsung has not published ram timings where Micron has.". It also says in there that vram voltage is proportional to the GPU's voltage so maybe the limitation has to do with the fact that if the GPU is using a certain voltage the card may be hard wired to feed the memory an unsafe voltage.
>>58467063
Is shit like this covered in basic EE/CE classes or something? I need to change my major. I've been a computer hobbyist for like 10 years but the extend of my knowledge was cold=good basically

>>58460342
>I buy AMD is because of the open nature of their platforms.
>>58466689
>not AMD tho

lmao, see

>>58465800
>HD radeon 6870 for example has a BIOS hard lock in the firmware

they definitely do hold it back. otherwise people would be overclocking to reach performances which shit on their future and even current gpu line up

they have retarded fanboys so disillusioned they have them spending $800 on a fucking gpu that can't even run the latest games stable at 1080p

Meh, I haven't overclocked my 1070 yet.
But it does reach almost 2Ghz boost at stock.
It's a Gaming X.

I'm curious if it's the same with other people that have the same card.

>>58467105
Yes. More for EE than CE however. There's a bunch of overlap but you can choose the right classes in either to dive into the specific properties of silicon and such. At least that was my experience.

>>58467157
Yes, that's roughly what you should expect. These cards actually boost past their set boost clock. My 1060 runs at 1911 stock while it's stock boost clock is lower. Overclocking any of the current cards, 1050/ti/1060/1070/1080 gives almost exactly 10% performance over refrence regardless of which card you buy

>>58460071
>implying it matters if you can oc to 2100mhz core + 4400-4500mhz mem, and performance will be much better than rx 480

>>58467038
>pic of 4850
>which oc way worse for me than gtx 1060

???

>>58468803
It was the first dedicated GPU I ever bought so I just posted it. It's not relevant to anything.
>>58468791
In my situation since I have a spare AIO the 480 would have been the better buy if I was able to hit 1500mhz with it, which I probably could. Whatever though I only play 1080p 120hz ulmb

>Nvidia is artificially holding back 2+ghz overclocks that only a fraction of users even attempt

Nobody actually believes this right?

>>58470792
Top OC'ers have been doing bizarre hardware and BIOS mods to get over the voltage hard limits imposed on Pascal by changing the circuitry so it reports lower values. It's not conjecture at this point really.

>>58470792
You're extremely uninformed. Half of the pascal cards will boost to 2ghz out of the box. The problem here is that there is 0 reason to buy anything better than the cheapest card you can find outside of quieter operation.


After doing more research it seems that in the 1080 at least there is a hard limit in the silicon it's self of 1.25 volts. All the cards you can buy on the market are BIOS limited below that but if you use an evga epower to bypass the VRMs you still run into the 1.25 volt limitation on the 1080.

Did EVGA fix their Capacitor issue?

is it safe to buy their 1070s?

also with all these limits in mind, does it even matter which brand i get if they are all going to properly cool themselves and wont be overclock significantly enough to differentiate themselves from the group?