GamersNexus discovers Nvidia's secret highly binned 2070 chip | Not all 2070s are made equal

[williamcll]

Maybe they did this because they watched linus' video about the 1070 Ti

[Nicnac]

1 hour ago, Master Disaster said:

Pretty much yeah, I mean binning has been happening since forever. All Nvidia are doing is making the process fairer by offering AIBs the ability to purchase binned GPUs and use them on factory OCed cards instead of the lottery system that was in place previously.

 

It's actually better for consumers because as you said, it guarantees that if they buy an OCed card they'll get the best performance possible or they can pay less and enter the lottery for a card which might OC well or might not OC at all (which is how it's always been anyway).

it's fair to AIBs, yes but not to the consumer as it basically sells two different chips under the same name and only hardware enthusiasts will know there's a difference. NVIDIA is offering better out-of-the-box performance and experience from which someone less familiar with the topic might profit. Nevertheless it creates a different product where there shouldn't be one (whats so hard about naming it 2075 or something like that?). And, again, while people buying better cards might profit from it those buying lower end cards will have a worse chip (not necessarily but pretty much guaranteed). So I guess it zeroes out in the end. The problem I see is that traditionally AIB cards were different only in the potency of their cooling solution and maybe looks. Now they also differ in performance (to a point they might actually considered different models) and that is great for AIBs primarily but not for the consumer. 

OCing was always a niche market but I get why they wanna close that niche. Why sell card people can overclock when they will just buy more expensive ones? 

[mr moose]

19 minutes ago, Nicnac said:

it's fair to AIBs, yes but not to the consumer as it basically sells two different chips under the same name and only hardware enthusiasts will know there's a difference. NVIDIA is offering better out-of-the-box performance and experience from which someone less familiar with the topic might profit. Nevertheless it creates a different product where there shouldn't be one (whats so hard about naming it 2075 or something like that?). And, again, while people buying better cards might profit from it those buying lower end cards will have a worse chip (not necessarily but pretty much guaranteed). So I guess it zeroes out in the end. The problem I see is that traditionally AIB cards were different only in the potency of their cooling solution and maybe looks. Now they also differ in performance (to a point they might actually considered different models) and that is great for AIBs primarily but not for the consumer. 

OCing was always a niche market but I get why they wanna close that niche. Why sell card people can overclock when they will just buy more expensive ones? 

But it's not a different product, it is exactly the same bar for the ability to perform better with an OC.  The consumer actually has a better chance of getting what they pay for now because nvidia have effectively upped the quality control and are now distinguishing between a chip that can't OC and one that can.

 

People are going to have to get used to this,   OC headroom will either plateau or diminish with every new processor (GPU or CPU). In fact since about 2000-2005 we have seen a plateau in clocks with most improvements coming from IPC.  So when performance reserves in new tech diminish (as they have been of late) and the only performance tweak left is increasing clocks that are already at their maximum, then we are all going to have to accept that some chips will be more expensive because they perform better whilst being exactly the same chip as another cheaper one that barely makes base clock under load. 

 

I link the below graph for interests sake,  Whilst it is of CPU's the issue is the same for all silicon not just processors.

 

 

[Carclis]

2 hours ago, mr moose said:

So basically nvidia put in extra quality controls to ensure customers  get OC'd products that can actually OC and some people cry extortion?

I think the issue is more down to people disliking turning pricing than binning being bad practice.

[mr moose]

1 minute ago, Carclis said:

I think the issue is more down to people disliking turning pricing than binning being bad practice.

It's ok to hate the pricing of a product, I hate the price of whiskey in Australia.   But at the end of the day it's just the market operating outside of my preferred expenditure.  If that's what some people deem extortion then they are going to be in for a rather rude shock if they ever decide to run their own business. 

[Carclis]

7 minutes ago, mr moose said:

It's ok to hate the pricing of a product, I hate the price of whiskey in Australia.   But at the end of the day it's just the market operating outside of my preferred expenditure.  If that's what some people deem extortion then they are going to be in for a rather rude shock if they ever decide to run their own business. 

Well AFAIK binning has pretty much always happened whether it be i5 8400 vs 8600k or a Gigabyte Windforce 1080ti vs their Aorus Xtreme. Some people are probably seeing the $1200 pricing here compared to $700 last generation and just blaming it on binning.

Edit: I mean it's not exactly like we're seeing those promised $1000 2080ti cards which are presumably the lower binned ones.

[Nicnac]

1 hour ago, mr moose said:

But it's not a different product, it is exactly the same bar for the ability to perform better with an OC.  The consumer actually has a better chance of getting what they pay for now because nvidia have effectively upped the quality control and are now distinguishing between a chip that can't OC and one that can.

 

People are going to have to get used to this,   OC headroom will either plateau or diminish with every new processor (GPU or CPU). In fact since about 2000-2005 we have seen a plateau in clocks with most improvements coming from IPC.  So when performance reserves in new tech diminish (as they have been of late) and the only performance tweak left is increasing clocks that are already at their maximum, then we are all going to have to accept that some chips will be more expensive because they perform better whilst being exactly the same chip as another cheaper one that barely makes base clock under load. 

 

I link the below graph for interests sake,  Whilst it is of CPU's the issue is the same for all silicon not just processors.

 

 

I understand that overall there's less oc headroom with new tech but is it really that much lower? I think they could set their power limits higher and do more for the oc crowd instead of cashing in on them. Or maybe I am just nostalgic and miss the time when oc meant actual free performance from a product...

[mrthuvi]

If manufacturers bin the cards themselves (which they did for quite a long long time), the binning process still costs some money and the cost is passed on to the customer anyway.

 

- We just don't know if Nvidia's binning is more expensive or efficient than previous binning process as we lack that data. Has there been any leaks about manufacturers' opinions this?

- Customer can now roughly know the crème de la crème gpus from all manufacturers are now quite equal silicon wise compared to different standards of binning previously (except Nvidia's FE might have an unfair advantage).

- We now suddenly realize that 1st few batches of 1080ti FE's suspiciously better overclockability against any other cards as Nvidia testing the water and perfecting their binning process.

[TVwazhere]

6 hours ago, Master Disaster said:

Also Steve didn't discovery this, Paul posted a video on October 18th about it

-snip-

20 hours ago, TVwazhere said:

Paul's Hardware did some testing on this a while ago

-snip-

¯\_(ツ)_/¯

I mean if people are surprised at this point that chips get binned, they're new to the GPU market. Nvidia just made them easier to spot with a naming scheme

[Mira Yurizaki]

4 hours ago, Nicnac said:

I understand that overall there's less oc headroom with new tech but is it really that much lower? I think they could set their power limits higher and do more for the oc crowd instead of cashing in on them. Or maybe I am just nostalgic and miss the time when oc meant actual free performance from a product...

Considering that world records have floated around 7-8GHz and the fastest clocked factory parts are within 2GHz of that range, yeah, I'd say we're approaching the limit of how fast we can clock things.

 

There's also other reasons why clock speed hasn't risen dramatically over the years as it once did: https://software.intel.com/en-us/blogs/2014/02/19/why-has-cpu-frequency-ceased-to-grow

 

EDIT: The other thing between now and then is that increasing clock frequency is a linear change in power consumption and back then, power consumption was much lower. A 100MHz Pentium overclocked to 133MHz would provide 33% more performance, but raise the power consumption by 3W, which is less than what a USB device will eat up at 1A.

 

Now to get even the same amount of performance bump on even a Ryzen, you would need at least 1.2GHz, bump, which would also add another 20W+

[corsairian]

This reminds me of the 3gb and 6gb 1060, also the 960 3.5gb. Or even AMD doing that thing with the 580.

 

This somewhat reminds me of the GTX 260 and GTX 260 Core 216 (2008.. wow I feel old)  that launched shortly after. It was a better performing card for the same price. It had more computer units and possibly a larger bus, but its been so long I forget. A similar thing with the 280 and 285. But they gave the cards different names.

 

At least give the card a different name lol. Its probably not fast enough to be a 2070ti tho. If i was going to buy a 2070 I'd surely want this new fancy 2070 chip. 

[mr moose]

8 hours ago, Nicnac said:

I understand that overall there's less oc headroom with new tech but is it really that much lower? I think they could set their power limits higher and do more for the oc crowd instead of cashing in on them. Or maybe I am just nostalgic and miss the time when oc meant actual free performance from a product...

I think maybe the concept of free performance was misplaced.  Nowadays every consumer gets that "free" performance out of the box because today's CPU/GPU's can essentially overclock themselves due to all the new throttling tech they have.  So it's not like you can't get it anymore, it's just that everyone gets it without having to do anything because manufacturers have worked out how to do it before it leaves the factory.

[Mira Yurizaki]

1 hour ago, corsairian said:

This reminds me of the 3gb and 6gb 1060, also the 960 3.5gb. Or even AMD doing that thing with the 580.

 

This somewhat reminds me of the GTX 260 and GTX 260 Core 216 (2008.. wow I feel old)  that launched shortly after. It was a better performing card for the same price. It had more computer units and possibly a larger bus, but its been so long I forget. A similar thing with the 280 and 285. But they gave the cards different names.

 

At least give the card a different name lol. Its probably not fast enough to be a 2070ti tho. If i was going to buy a 2070 I'd surely want this new fancy 2070 chip. 

Except for the GTX 280 and 285, all of the examples you listed were actually different GPUs. i.e., they had different core configurations. And even with the 280 and 285, the 285 was a die shrink, which for all intents and purposes tends to make it a different thing anyway.

 

This is the same exact GPU, just that due to manufacturing fairies, some came out more well within tolerance than others. And even then, the reports of higher performance aren't even enough to warrant an entirely new video card SKU.

 

If we're going to start making new GPU models over manufacturing differences, we may as well have 10 versions of the damn thing, all priced to match

 

EDIT: And now that I think about it, it would probably hurt the company to create even more GPU models over such things because let's say they do have a system where there are 10 versions of the same GPU config. Now we have a 2070, 2071, 2072, etc. to 2079. People are just going to want the 2079, even though the performance differences are likely not going to matter for most people. So now you have stores with stocks of perfectly find GPUs that apparently nobody wants and social media whining about how there's no more 2079s in stock and that NVIDIA needs to get off their asses and build more.

[Dezz]

Intel does the same with their K and non-K CPUs, don't they?

 

On 11/5/2018 at 4:43 PM, Origami Cactus said:

Now we just need an rtx 2070 5gb, rtx 2070 3gb, rtx 2070 ddr4 and rtx 2070 6gb.

Also a version with raytracing cores disabled called the gtx 2070. Thanks @PatXioPC for the idea.

no '3.5gb' version? nah.. I'll keep my 970

[Tech Enthusiast]

On 11/6/2018 at 8:47 AM, mr moose said:

So basically nvidia put in extra quality controls to ensure customers  get OC'd products that can actually OC and some people cry extortion?

Exactly that, yes.

It is just the cool thing to do really.

Hating on the top dog has never been more in fashion. Who cares if the reasons make no sense?! This is a principle thing!