NVIDIA GeForce RTX Ampere Chips Feature Three Binning Tiers, Mostly Good Dies are Present

[Vayra86]

It has been common knowledge among extreme OC crowds and bench score ninjas that the FE series are the best at release. Later in the lifecycle it doesn't matter as much. If that trend continues is speculation, but why do you care anyways? You clearly aren't in those circles.

And yet the same guy says right below you that most of the time you really don't know. Even if there is a difference, the gap is minute at best and extreme OC crowds are not what the majority of these people are posting as. They are just posting as consumers, and those I do care about, because this is once again a whole lot of ado about nothing people get confused about. Bench score ninjas are what, 1%?

'Is my RTX a worse one' omg must buy top end version. That is the sentiment that is being falsely created here.

You'd also have to wonder when that point arrives where 'later in the lifecycle' it suddenly won't matter. Lottery is lottery. End of story. And all those stories about 'omg this stepping and this period' are very old, well known, and we've arrived at a point in foundry refinement that these things are nowhere near what they used to be. If the total performance gap is 2-3% you're already being very generous. I'm a big promotor of such perspective when these claims are made in news. Today the vast majority of this hype is just that, lots of hobbyists riding the train together.

But the most telling of it all, is that so far.... no sources were provided.
I rest my case and this article identifies as clickbait.

 

[TheoneandonlyMrK]

And yet the same guy says right below you that most of the time you really don't know. Even if there is a difference, the gap is minute at best and extreme OC crowds are not what the majority of these people are posting as. They are just posting as consumers, and those I do care about, because this is once again a whole lot of ado about nothing people get confused about. Bench score ninjas are what, 1%?

'Is my RTX a worse one' omg must buy top end version. That is the sentiment that is being falsely created here.

You'd also have to wonder when that point arrives where 'later in the lifecycle' it suddenly won't matter. Lottery is lottery. End of story. And all those stories about 'omg this stepping and this period' are very old, well known, and we've arrived at a point in foundry refinement that these things are nowhere near what they used to be. If the total performance gap is 2-3% you're already being very generous. I'm a big promotor of such perspective when these claims are made in news. Today the vast majority of this hype is just that, lots of hobbyists riding the train together.

But the most telling of it all, is that so far.... no sources were provided.
I rest my case and this article identifies as clickbait.

While performance differential could be argued Nvidia definitely do bin chip's, that's a known fact?!
And since they actually label them differently, is it arguable ?, we had this with Turing so it's happening.
New news it's not.

 

[Dyatlov A]

How to know i am buying the best bin 2? I will need to use with 450W PSU, therefore very important to me good undervolting capabilities.

 

[Assimilator]

How to know i am buying the best bin 2? I will need to use with 450W PSU, therefore very important to me good undervolting capabilities.

Buy a better PSU or don't buy an RTX 3000 card.

 

[Vayra86]

While performance differential could be argued Nvidia definitely do bin chip's, that's a known fact?!
And since they actually label them differently, is it arguable ?, we had this with Turing so it's happening.
New news it's not.

They labelled a few chips with A- versions indeed.

Is that the case here? I didn't read it. In fact I didn't read anything much of substance here. Devil as always is in the details...

In the same vein, we 'had this with Turing' but all we really had was a set of chips that got blessed with higher TDP budgets alongside some vagueness about binning.
Hey look, if I tweak the power target with any Nvidia card with GPU Boost, I get different numbers too. We all know the boost algorithm finds space within the aforementioned TDP budget. A card sold with higher TDP budget simply has more headroom to boost, and that is what they managed to put on a box too. Wow. Such Binning. Much performance.

People oughta wake the f up...

 

[TheoneandonlyMrK]

They labelled a few chips with A- versions indeed.

Is that the case here? I didn't read it. In fact I didn't read anything much of substance here. Devil as always is in the details...

In the same vein, we 'had this with Turing' but all we really had was a set of chips that got blessed with higher TDP budgets alongside some vagueness about binning.
Hey look, if I tweak the power target with any Nvidia card with GPU Boost, I get different numbers too. We all know the boost algorithm finds space within the aforementioned TDP budget. A card sold with higher TDP budget simply has more headroom to boost, and that is what they managed to put on a box too. Wow. Such Binning. Much performance.

People oughta wake the f up...

So it's all bullshit then?

They don't find the less leaky chips and pass those to the quadro team/buyer's?.

They then just have big bins of working and not working and it's lucky dip?.

Riiight, we disagree but do note I wasn't asleep , give yourself a nudge though, just check.

 

[Assimilator]

So it's all bullshit then?

They don't find the less leaky chips and pass those to the quadro team/buyer's?.

They then just have big bins of working and not working and it's lucky dip?.

Riiight, we disagree but do note I wasn't asleep , give yourself a nudge though, just check.

What exactly does Quadro have to do with any of this?

And of course there's binning. It's just that it won't matter to most consumers. Unless they want to pay more for a better-binned card. Which NVIDIA is perfectly within its rights to charge them extra for.

What will happen is that, since NVIDIA won't be able to sell 10% of its best-binned chips at a higher price (maybe 1%), most of them will end up going into the ordinary consumer pool, and the silicon lottery will work its democratic nature as always. In other words, it'll be just like Turing, and Pascal, and every series before...

 

[TheoneandonlyMrK]

What exactly does Quadro have to do with any of this?

And of course there's binning. It's just that it won't matter to most consumers. Unless they want to pay more for a better-binned card. Which NVIDIA is perfectly within its rights to charge them extra for.

What will happen is that, since NVIDIA won't be able to sell 10% of its best-binned chips at a higher price (maybe 1%), most of them will end up going into the ordinary consumer pool, and the silicon lottery will work its democratic nature as always. In other words, it'll be just like Turing, and Pascal, and every series before...

Yes where the best chip's , with the lowest leakage and operating volts go to enterprise/Quadro see the link.

 

[Assimilator]

Yes where the best chip's , with the lowest leakage and operating volts go to enterprise/Quadro see the link.

What link? The OP says nothing about Quadro.

 

[TheoneandonlyMrK]

What link? The OP says nothing about Quadro.

There's Links between Quadro and binning chips not a link to the Op, I originally replied to Vayra not you.

 

[Vayra86]

So it's all bullshit then?

They don't find the less leaky chips and pass those to the quadro team/buyer's?.

They then just have big bins of working and not working and it's lucky dip?.

Riiight, we disagree but do note I wasn't asleep , give yourself a nudge though, just check.

Nobody is denying binning... I am denying the need for this whole article to exist. What's being reported here? What's the point? This is like saying 'Cars run on petrol... but some run on electricity or gas'. Well, thanks for the update

Water is wet but here we are being led to believe something special is going on. OMG they have three tiers 'what if I get a bad one', 'how do I find a good one'...

Fake demand & clickbait. Look at the headline, 2nd part. 'Mostly good dies are present'... are we insinuating you might strike on a bad one with your purchase?! Nope... But its still in the title. Does this then mean most wafers are nearly perfect with just 'somewhat less perfect' dies on them? Obviously not. So what is it then?

So yes its all bullshit and many fools and money are involved. Any chip that is binned for a specific product is binned for that product, its the most basic thing since forever. I don't need to click links for that. Any time Nvidia thinks a specific die may go into another product, they actually tell you. We've had our share of GP104 movements for example with Pascal. We had A-dies with Turing. In neither case though did those qualify for a product that performed differently from the similar named one with a different chip inside. Even the so called remarkable Turing FE's needed a higher power target to 'be better'. Nuff said.

 

[londiste]

Yes where the best chip's , with the lowest leakage and operating volts go to enterprise/Quadro see the link.

Lowest leakage and operating volts may not - and often do not - coincide with best overclockability.

 

[TheoneandonlyMrK]

Nobody is denying binning... I am denying the need for this whole article to exist. What's being reported here? What's the point? This is like saying 'Cars run on petrol... but some run on electricity or gas'. Well, thanks for the update

Water is wet but here we are being led to believe something special is going on. OMG they have three tiers 'what if I get a bad one', 'how do I find a good one'...

Fake demand & clickbait. Look at the headline, 2nd part. 'Mostly good dies are present'... are we insinuating you might strike on a bad one with your purchase?! Nope... But its still in the title. Does this then mean most wafers are nearly perfect with just 'somewhat less perfect' dies on them? Obviously not. So what is it then?

So yes its all bullshit and many fools and money are involved. Any chip that is binned for a specific product is binned for that product, its the most basic thing since forever. I don't need to click links for that. Any time Nvidia thinks a specific die may go into another product, they actually tell you. We've had our share of GP104 movements for example with Pascal. We had A-dies with Turing. In neither case though did those qualify for a product that performed differently from the similar named one with a different chip inside. Even the so called remarkable Turing FE's needed a higher power target to 'be better'. Nuff said.

Oh riiiiiight, I can agree on the article, I mis understood you sorry..

Yay @londiste I was talking bins for pro /enterprise not over clocking , it depends on the cooling too but none of that matters to enterprise, just performance ,TCO.. .

 

[Berfs1]

I just can't see the watts on Samsung 7nm being worse than Turing, which is about what it'd need to matter I think.

Performance per watt is what matters. Not just watts.

What exactly does Quadro have to do with any of this?

And of course there's binning. It's just that it won't matter to most consumers. Unless they want to pay more for a better-binned card. Which NVIDIA is perfectly within its rights to charge them extra for.

What will happen is that, since NVIDIA won't be able to sell 10% of its best-binned chips at a higher price (maybe 1%), most of them will end up going into the ordinary consumer pool, and the silicon lottery will work its democratic nature as always. In other words, it'll be just like Turing, and Pascal, and every series before...

Because Quadros are made for reliability (as well as teslas), because reliability is more important in the workstation environment than in consumer cards. Having a graphics card that lasts 1 year longer is better than having it perform 2% faster for the gaming market. That is why workstation/server parts get the higher binned chips, because not only they are the more expensive market, they will care if the server lasts another entire year rather than just 1% more performance.

Nobody is denying binning... I am denying the need for this whole article to exist. What's being reported here? What's the point? This is like saying 'Cars run on petrol... but some run on electricity or gas'. Well, thanks for the update

Well, TechPowerUp has articles for NASes, so it is clear that TPU is NOT just a news publisher just for regular gamers, but for various different segments in the computer industry as well, overclockers included.

 

[EarthDog]

Buy a better PSU or don't buy an RTX 3000 card.

This. Undervolting is shooting yourself in the foot...

Well, TechPowerUp has articles for NASes, so it is clear that TPU is NOT just a news publisher just for regular gamers, but for various different segments in the computer industry as well, overclockers included.

well, that went over your head.

 

[R-T-B]

Performance per watt is what matters. Not just watts.

That's what I mean.

 

[BoboOOZ]

Soooo, if yields are so damned good, where the hell are those graphic cards? Only 50 global retailers got them?