NVIDIA To Stop Differentiation of Better Binned A-dies for AIB Factory Overclocked Cards

[Raevenlord]

A report from Tom's Hardware.de claims that multiple industry sources have confirmed that NVIDIA will stop offering higher-binned, differentiated A-dies of their Turing silicon. If you'll remember, the company introduced specific A-binned chips for AIB partners to ship with factory overclocks to customers, due to their higher overclockability - and likely, better power consumption profile - when compared to non A-binned dies. This practice was reserved to the company's best, though, in the form of the TU104-400A-A1 die (compared to the TU104-400-A1 dies used in non-overclocked versions of AIB graphics cards). The company is now seemingly killing this practice by offering a one-off Turing die with no such limitations.

This move by NVIDIA - on which we reported firsthand here at TPU - was likely a solution to somewhat less than ideal yields for its TU-104 chips, ensuring partners could provide the best experience to users who were willing to pay the most. The fact of the matter is that AIB partners were locked out of overclocking non-A dies should they acquire them (which were going for less than their higher-binned A-cousins), though the end-user would not see such a limitation - besides the one imposed by the expectedly less capable dies present on those non factory-OC'd cards.



View at TechPowerUp Main Site

 

[Prima.Vera]

We need asap a 3rd competitor on the video cards business. Those callous practices needs to stop somehow...

 

[Caring1]

It's not because of consumer feedback, it makes sense due to binning and yields.
Now everyone has the same chance of buying a dud, or winning the silicon lottery.
I prefer the idea of buying a better chip if I want to overclock, and it shouldn't matter what brand.

 

[SteelCrusaders]

This practice should've never happened.... Navi needs to succeed to stop this kinda bs

 

[R-T-B]

You're bleeding extra a's all over your article, Raevenlord. One in the "overclocakabilty" and one in "dia" instead of die.

NVIDIAs fault surely. Too many A dies.

 

[OSdevr]

We need asap a 3rd competitor on the video cards business. Those callous practices needs to stop somehow...

We're getting one, but it may not be an improvement.

 

[mak1skav]

And what about prices? Are they still going to sell them as "premium" ?

 

[Deleted member 24505]

So now it really will be a silicon lottery instead of buying a guaranteed A die for good OC etc.

 

[Vayra86]

We need asap a 3rd competitor on the video cards business. Those callous practices needs to stop somehow...

Yeah, how dare they bin these chips and put them in different products!

You can really expect more of this regardless of competition as the nodes get smaller and the performance wins get smaller per release. Its only positive that yields are now good enough to get these large chips to a more consistent level. Because what this really means in practice, is that those lower priced AIB models now *might* have a few more clock bins to play with.

 

[nguyen]

There is no difference between non-A and A chip beside the power limits, thing is Turing is really hold back by power limits. So basically with A chip you pay for 5-10% more performance than non-A with 30% more power consumption. That is negligible performance difference for a very noticeable power increase (same deal as Vega 64 was). So if you are on wc or like benchmarking A chip is better, for gaming they are the same.

 

[jmcosta]

This practice should've never happened.... Navi needs to succeed to stop this kinda bs

This was a good and a bad thing, depending on your perspective. The power limitations and silicon quality difference were meaningless for the mainstream consumer and if you wanted a better overclock ability you would spend on a premium card either way.
as for the enthusiasts this move only simplified.

 

[Crackong]

No more A SKU and every 2080 selling at $699
vs
No more A SKU and every 2080 selling at $799

Which one?

 

[arbiter]

This was a good and a bad thing, depending on your perspective. The power limitations and silicon quality difference were meaningless for the mainstream consumer and if you wanted a better overclock ability you would spend on a premium card either way.
as for the enthusiasts this move only simplified.

Only difference between the 2 is which one will overclock and which one won't. Being an non-A or A is nvidia doing what all AIB did when they decide what chips go on their higher end cards that come per-overclocked. Now you just won't know.

 

[B-Real]

We need asap a 3rd competitor on the video cards business. Those callous practices needs to stop somehow...

https://www.guru3d.com/news-story/r...ven-navi-gpus-but-have-efficiency-issues.html

 

[jabbadap]

Only difference between the 2 is which one will overclock and which one won't. Being an non-A or A is nvidia doing what all AIB did when they decide what chips go on their higher end cards that come per-overclocked. Now you just won't know.

Is it really. Only non-A bin review I remember seeing is EVGA RTX 2070 black here on tpu and it OC as well as any RTX 2070:

 

[newtekie1]

This isn't really a big deal. Before, nVidia was pre-binning chips and the AIBs had to use the good chips in pre-overclocked cards. If you wanted one of the "better" A chips, then buy a pre-overclocked card. If you buy a stock clocked card, then you're playing the silicon lottery, just like you always did.

Now, they're going back to the traditional way, and everyone is playing the silicon lottery. How terrible?

 

[jmcosta]

Only difference between the 2 is which one will overclock and which one won't. Being an non-A or A is nvidia doing what all AIB did when they decide what chips go on their higher end cards that come per-overclocked. Now you just won't know.

All cards can be overclocked. I think Nvidia did this to get enough chips to fill the market demand because of the initial low yield, so they split into different standards, one for low cost cards and other for premium cards instead of moving to a tier lower (to disable cores/cut down the chip)

 

[EarthDog]

Since another thread beat the news to it.... here is a link to that thread for its conversation pieces...https://www.techpowerup.com/forums/...t-really-a-scandal-which-is-a-scandal.255158/

I'm forever thankful the word scandal didnt make it into the title.

 

[Deleted member 24505]

Since another thread beat the news to it.... here is a link to that thread for its conversation pieces...https://www.techpowerup.com/forums/...t-really-a-scandal-which-is-a-scandal.255158/

I'm forever thankful the word scandal didnt make it into the title.

My thread but turned into a shit storm

 

[John Naylor]

This isn't making a lot of sense .... hardware is always about one company trying to differentiate themselves from the pack. We had nVidia refence cards and we had AIB cards and everyone was fine with that. The AIB cards were always faster. Now vendows complain that they can't compete on proce... so nVidia says OH OK, we will give you access to the cheaper GPUs that meet specs but don't OC so well... Nvidia distiquishes itself with the FE ... and EVGA suffering from recent failures with their SC line can sell the black by folks thinging it's so,mething special when it's is a GPU not well suited for OCing.

But if they are going to stop binning, I don't see them tossing the ones that OC well .. and only keeping the class that doesn't ... if yields always improve over time so in the normal curse of events, it should be the slower ones that we stop seeing.

 

[nguyen]

There is absolutely no binning here between A chip or non-A chip people, Non-A chip overclock just as well as A chip but are locked down with a lower stock clock and more TDP restrictive bios. AIB can do whatever they want with A chip (OC version) regarding TDP, even remove power limit completely (XOC bios) while non-A chip is locked to lower power limit bios. TL;DR version as long a you don't touch the power limit, A chip and non-A chip overclock to pretty much the same clock.