Manufacturers roll out firmware updates for GeForce GTX 1070 due to memory Issue

by Hilbert Hagedoorn on: 10/21/2016 07:31 AM | source: | 227 comment(s)

If you own a GeForce GTX 1070, you might want to pay attention to this news-item as you could be effected. There have been some problems reported with certain GeForce GTX 1070 cards.

After further investigation it became apparant that the issues like flickering and poor memory overclocking results on the memory sub-system are graphics memory related. GeForce GTX 1070 cards fitted with Samsung memory do not have any issues, however some manufacturers have or switched towards Micron chips, and these are the culprit of the reported issues. Especially during overclocking of the graphics memory problems will appear fast with bad results like checkerboard patterns. The problem is was found in the the speed of the voltage adjustment from the low power idle P-States to support the memory voltage requirements under load. If you can keep the idle voltage above 0.800V before you apply the overclock you never see the issue.

Some users also have reported that at default configuration they noticed the occasional odd issue like flickering or experiencing graphical artifacts. Multiple brands are and could be effected. Right now most of them are working on (or already issuing) a BIOS firmware update for the graphics cards, these BIOS updates likely increase the voltage level of the GDDR5 memory. Mind you that this only effects GeForce GTX 1070 graphics cards fitted with Micron based GDDR5 memory.

I've been doing some rounds at the manufacturers with the following info (we'll update as feedback gets in). A number of manufacturers is or have been working on VBIOS updates:

EVGA: updates download here
Gainward: updates download here (select card -> BIOS download)
Palit: updates download here (select card -> download)
MSI : Updates downloads available (check for proper card model!)
Gigabyte: F11_Beta Update downloads available (check for proper card model!)
ASUS downloads available (check for proper card model!)

Inno3D: did not reply
KFA2/Galaxy: did not reply
PNY: did not reply
Zotac: no info available on website - update should be out

The most easy way to check what brand of memory your GeForce GTX 1070 is using is to download GPU-Z and read out the memory type (click thumbnail below for an example). The reality is that Micron’s GDDR5 memory chips shouldn’t really be any weaker opposed to what the Samsung ICs are doing. All cards we tested here ate Guru3D.com got Samsung memory. It however is not uncommon for manufacturers to switch to another brand of memory, as they purchase them in SMT batches and yes, a selection in price is made but should not effect performance. Due to the fact that our samples have Samsung memory we have not been able to verify any instability.

Have you noticed any issues with your GTX 1070? If so, let us know in the comments of this news-itsem.

Related Stories

Intel to manufacture ARM SoCs - 08/17/2016 03:16 PM
Pretty big news really, Intel entered a licensing agreement with ARM, basically this allows parties like LG, Qualcomm, Apple, and Samsung, to fab ARM SoCs at Intel fabs. ...

AMD Sells Majority Back-end Manufacturing Fabs in a Joint venture - 10/16/2015 11:05 AM
AMD sold a majority stake in its test, assembly and packaging facilities in Asia as part of a joint venture with China&rsquo;s Nantong Fujitsu Microelectronics....

AMD will manufacture GPUs at Global Foundries - 12/05/2014 11:24 AM
In our previous post about 16nm TCSM we've been wondering as to why AMD was not on the 16nm fab list, whereas other players like Nvidia have been named. The answer just came to us though, AMD will m...

Intel to Manufacture 14 nm Process for Panasonic System-on-Chips - 07/08/2014 08:14 AM
Intel Corporation today announced that it has entered into a manufacturing agreement with Panasonic Corporation's System LSI Business Division. Intel's custom foundry business will manufacture futur...

CES: NVIDIA GSync available in Q2 from 6 manufacturers - 01/06/2014 08:50 AM
We've been talking a lot about NVIDIA Gsync (read our preview here). Over at CES Nvidia announced that in Q2 these monitors will actually see availability, from six manufacturers. During the initial...

16 pages 1 2 3 4 > »

SirDremor
Senior Member

Posts: 586
Joined: 2008-06-20

#5348313 Posted on: 10/20/2016 12:02 PM

Here we go again. I hate those dirty discoveries months after a product's release. Sure this is not the 970 memory fiasco but still it's a dirty discovery hidden under the rug. Can NVIDIA stop this already with these expensive products?

It is not a "dirty discovery", it's not like someone was hiding this as a secret. So maybe you stop bashing Nvidia just because you don't like this news, ok?

In any way - the choice to use Micron memory was of respective manufacturer, not Nvidia. You can go and unload all the trash you have on those manufacturers instead...

jadium
Member

Posts: 59
Joined: 2005-10-10

#5348315 Posted on: 10/20/2016 12:09 PM

I do not see the issue. It's not like they falsely advertised or something or the cards burst in flamea (*cough* samsung batteries *cough*) It is just a minor issue that can be sorted with a bsimple bios update.

Denial
Senior Member

Posts: 11983
Joined: 2004-05-16

#5348320 Posted on: 10/20/2016 12:17 PM

jadium
Member

Posts: 59
Joined: 2005-10-10

#5348323 Posted on: 10/20/2016 12:22 PM

It's not really a hardware defect. The incorrect voltage is being supplied to the memory. They can tune the voltages up and down in software, people do it all the time with BIOS mods.

The voltage is lined up with the Micron specifications, if for some reason you have to push more to stabilize it, it's a hint that you have a flaw in your design or your chips.

More voltage or overclocking have a cost, usually life expectancy of the said product.

Denial
Senior Member

Posts: 11983
Joined: 2004-05-16

#5348326 Posted on: 10/20/2016 12:29 PM

The voltage is lined up with the Micron specifications, if for some reason you have to push more to stabilize it, it's a hint that you have a flaw in you design or your chips.

More voltage or overclocking have a cost, usually life expectancy of the said product.

You have a source for this? Pretty much everything I've read is that the voltage on the memory is unchanged between the Samsung and the Micron chips, the bios between the two was identical up until this point. That's not to mention that Nvidia doesn't specify any lifespan for the card. If it ends up lasting 14 years instead of 15, I doubt it matters.

alanm
Senior Member

Posts: 8513
Joined: 2004-05-10

#5348327 Posted on: 10/20/2016 12:32 PM

One thing I know we will see is that a lot of issues - aside from the actual memory problem - will be blamed on the vram. Like the game issue earlier mentioned.

Krimson
Junior Member

Posts: 7
Joined: 2010-05-28

#5348331 Posted on: 10/20/2016 12:42 PM

stigbeater
Member

Posts: 54
Joined: 2012-05-25

#5348357 Posted on: 10/20/2016 01:24 PM

i have micron on my MSI gaming X i did find this in a redit thread
BullnBear 3 points 1 month ago
Found an explanation on an EVGA thread which might shed some light on whats going on. The comment reads as follows:
The "Micron" problem is not the Micron Ram itself but, I believe, a bug in the micron version of the bios that increases the clock speed of the Ram from a low voltage state (below .800v) faster than the VRMs can be cranked up to provide enough voltage for the memory to support the high clock speed. An overclock from low voltage states will start giving checkerboard artifacts somewhere in the +400 to +500 range.
If the voltage is already set high when you increase the slider and hit the apply button. running Kboost or locking rge voltage in afterburner, the memory will clock to +600 or better with no artifacts, of course the exact levels will depend on the individual card.
The Samsung memory either copes with the few milliseconds of voltage starvation better or the different bios version increases the voltage faster to support the memory at the high clock speeds.

spine
Senior Member

Posts: 111
Joined: 2005-05-04

#5348360 Posted on: 10/20/2016 01:25 PM

Doesn't happen with Founders Edition cards.

Maybe their value is now becoming apparent...

Stormyandcold
Senior Member

Posts: 5133
Joined: 2003-09-15

#5348362 Posted on: 10/20/2016 01:39 PM

Doesn't happen with Founders Edition cards.

Maybe their value is now becoming apparent...

There is something to be said about 1st-batch products...

Undying
Senior Member

Posts: 10685
Joined: 2008-08-28

#5348365 Posted on: 10/20/2016 01:44 PM

Bios updates this days are easy as installing the drivers. Most cards even have dual-bios switch.

I dont see whats the big deal is.

Denial
Senior Member

Posts: 11983
Joined: 2004-05-16

#5348374 Posted on: 10/20/2016 02:03 PM

Doesn't happen with Founders Edition cards.

Maybe their value is now becoming apparent...

All the FE cards use Samsung.

fantaskarsef
Senior Member

Posts: 10065
Joined: 2014-07-21

#5348375 Posted on: 10/20/2016 02:06 PM

Doesn't happen with Founders Edition cards.

Maybe their value is now becoming apparent...

You do know that they're not the only cards using Samsung memory?
Or that a simple bios flash for micron cards might fix everything?
Or that when the fix does work, you paid more for the same card essentially?

Just thinking.

Weecka
Senior Member

Posts: 300
Joined: 2007-12-10

#5348377 Posted on: 10/20/2016 02:15 PM

Isn't any card you buy such as the "superclocked" SC model already overclocked at factory nearly to as fast as it can go reliably?

If you are then trying to add even more overclocking to it, then of course the GPU overclock and any benchmark run will fail. Doubt it is anything to do with the memory.

The sc/oc cards are not overclocked to max. They are overclocked to a certain degree so the superior cooling (aftermarket) would still maintain acceptable noise and cooling levels for a 24/7 use. Most of these cards can go alot higher as long as the heat is not going too high. As it is one of the fastest pc hardware killers. As you notice when Hilbert tests these cards he's pushing more or less 100/500 overclocks on core/memory respectively (on oc'd/sc'd models). For exaple, intel wouldn't recommend overclocking i7 2700k (that i still have - BOO) to 4.7-5 ghz for 24/7 use as it it is suppose to severely reduce the lifespan of the cpu. I've been running it at 4.7ghz for nearly 5 years, but wih a custom H2O loop that is keeping it very cool (60-65 C on 100% stress testing load)

Weecka
Senior Member

Posts: 300
Joined: 2007-12-10

#5348379 Posted on: 10/20/2016 02:20 PM

It is an issue, you just fail to see it. And that's ok.
Defective products (hardware level) must always! Always! be recalled, fixed and exchanged (but yeah it cost a lot of money to do so)...Soft patch a Software is ok, soft patch a Hardware defect is NOT...

The problem is, they get away with it way too easily lately...

If the card would have a chance of malfunctioning and maybe even damaging your pc then i have no doubt they would recall them. But if a bios fix can solve the problem without affecting performance/reliability of the card then it would make no sense to do so.

Here we go again. I hate those dirty discoveries months after a product's release. Sure this is not the 970 memory fiasco but still it's a dirty discovery hidden under the rug. Can NVIDIA stop this already with these expensive products?

You do understand that it is not nvidia that is putting the memory onto the card but the actual card manufacturers (such as evga, asus, gigabyte..etc.) nvidia pretty much only supply the gpu chip. So the problem lies (my guess) in lack of communication or information between micron and the aforementioned card manufacturers.

16 pages 1 2 3 4 > »

Post New Comment

Click here to post a comment for this news story on the message forum.