Pascal Card Binning

[Nicholas Peyton]

Hi,

I keep getting "standard" replies to questions to EVGA when trying to determine the answer to this question.

I know this question has been answered before to an extent but I still don't believe the answer "fully" explains the position.

To see what I mean this is the latest email I've sent off:

If I receive a worthwhile reply I'll post it -- or if anyone can shed any further light on the nitty gritty/detail I believe it would still be worth a read and maybe of interest to some people....


===========================================================
===========================================================
To: supportEU@evga.com

Hi,

I understand what you mean about the Silicon Lottery - sorry I am trying to explain what I mean my question a little better...

... I mean are the Nvidia GPU's that you place on your classified board tested for higher stability than for example the EVGA Founders Edition?

I will try to explain what I mean better by offering two schools of thought and asking which is true:


First School of Thought/Theory
A Founders Edition has a boost clock of 1700+.... so these cards may only be tested to ensure they reach 1700+ and may not even overclock to 1860 (Classified is factory overclocked to 1860)... so there must be at least "some" test to ensure they can perform stable at this factory overclock as they "are" clocked higher than a Founders Edition.

This "test" although does not guarantee any "even higher" overclocks would still 'in effect' give you at least a little extra luck at getting a more overclockable GPU. This would still be worth knowing!!


Second School of Thought/Theory
The second school of thought or theory is that in fact while all cards will overclock differently (silicon lottery as you say) and some may never reach 1900+, it is still generally accepted by EVGA that ALL NVIDIA Pascal GPU's will "AT LEAST" overclock to your factory overclock of 1860 therefore "no extra" testing is needed of GPU's before they are picked for use on a Classified board.

Which school of thought/theory is actually true of EVGA? (This is what I am trying to find out -- if there is potentially but not necessarily better quality GPU's used in the Classified due to the GPU's having to pass stability test at higher clock. The Silicon Lottery aside do I have any "extra" chance of getting a better GPU due to buying a Classified instead of a lower spec card at all? Even in the smallest degree?



Do you see what I mean now? If there is even a "slight" chance that purchasing a Classified will give me an extra 1% chance of having a better binned GPU this is still worth knowing about!

Or are ALL GPU's equally placed. Are they only tested once then no other tests are carried out before GPU's are allocated to a board?

. Is there "exactly" the same chance "statistically" of getting a more overclockable GPU whether you have a Founders Edition (low end) or Classified (High End).?

Thanks...

Nick Peyton

===========================================================
===========================================================

 

[slozomby]

if I was in the marketing department of any of the manufacturers I wouldn't let anyone answer that question either. no one wants to hear their $600 video card uses "lower" quality chips than a $625 card. they may also be limited by contract with NVidia about what they are allowed to talk about in regards to binning.

 

[Nicholas Peyton]

True, but when an Asus 08G Strix/EVGA Classified is now currently on many online websites at nearly £900 while you can pick up a lower clocked Strix/EVGA FE at just over £600 this information aught to make its way out into the open somehow....

 

[Athlon2K15]

I think all the chips are the same and the classified and other high end chips just have better onboard power management so you can put more voltage through for overclocking.

 

[Kanan]

I concur with that. Chips are all almost the same and variances are very slim. The differences are the pcb design and components supporting the gpu and the bios.

This holds true for maxwell as well. Probably almost any gpu. I just know some have higher variances but we're talking only about 50-100 mhz here - not 200 or even 300 mhz.

Example:
I have a high asic 780ti at 92.9% (only 1% of Gpus are that high). But my card is still limited by bios and pcb design and even it's cooler to overclock more than 1220ish. Highest clocks of 780ti are over 1300 though.

 

[Nicholas Peyton]

Received reply:

========================================================
Hi Nick,

They are tested to confirm they are stable at the clock specifications for the card however as every GPU is slightly different so different cards of the same model can do different max overclocks and we do not have any statistics or numbers to provide for the produced cards. Unfortunately we cannot guarantee overclock speeds and only what the card ships with. While we cannot officially support overclocking we do have a very active member base on our forums that have written numerous guides and can be extremely helpful when trying to push the bounds of your hardware. They can be found here.

Regards,
EVGA
=========================================================

 

[the54thvoid]

As an owner of a Kingpin card, the EVGA overclocking forums are frequented by Vince Lucido (Kingpin) and others. I got my info on modding a 980 KP Bitspower water block (to fit the 'Ti' version) from there. Very good guys and full of good advice - refreshing actually. Vince was quite adamant that you shouldn't buy a KP card unless you were going sub zero (if your goal was overclocking) as the cards all had an air or water limit of about 1550Mhz.

As for the Classified cards - you pay for a guaranteed marketed boost clock and better PCB components. Unfortunately, Nvidia have destroyed overclocking by hard limiting voltage controls. If you buy any 1080 for OC purposes, save your money and buy the cheapest card that is water block compatible. If you're not using water cooling - buy the best air cooler version.

To be fair, it's more process restricted (the voltage) as with smaller nodes the voltage ripple affects the stability more and more as the chip space shrinks smaller and smaller. It's hard to control voltage when you drop from 28nm to 16nm and smaller.

 

[GreiverBlade]

soooo even cheaper pascal do well then ... my Armor OC 1070 is one of the cheapest i could find around (still based of the Gaming X/Z series PCB minus some component like power phases, caps and 6pin connector )
base clock 1556, base highest sustained monitored boost 1930 (with 372.70) GPU-Z reported boost 1747 (primary panel), OC experimentation: max base 1700'ish max sustained boost (VRel in effect) 2088 peak 2100...

nonetheless i can't help but laugh at EVGA when they mention "handpicked" chips for the K!ngp!n based on the ASICS (wait ... Pascal finally has a ASICS readout? ) quality, other than Marketing ... i can't see one of those card getting a better OC on air (LN2 score is not a valide score, unless @the54thvoid wrote: aiming to break OC records ... for a couple of seconds/minutes ) and paying premium for that is ... well ... imho not really worth it (unless you want to showoff "heyz lookey i has a K!ngp!n card", remark not concerning @the54thvoid , since he know what to expect from the card.)

also for me a FE is not even a viable option, even with the ease to find a standard water block for it... the PCB layout is too much basic for me and the price is too high to be justified by a "slightly good albeit looking really nice" blower cooler

not that OC is actually needed for now (be it 1080 or 1070 or Titan X P or upcoming 1080Ti, that's specially true for the 2 later )