Nvidia will stop producing non-A die variants of Turing for RTX 2080 and 2070

[onetheycallEric]

In brief: Nvidia has been dividing its Turing silicon into A and non-A dies for RTX 2080 Ti, 2080, and 2070 graphics cards. This allowed Nvidia to bin the higher quality Turing A-dies for board partners' flagship, factory overclocked cards. It also allowed use of the non-A die silicon for non-overclocked, lower price point cards. Now Nvidia will only produce the TU104-410 and TU106-410 dies, ceasing production on the lower quality non-A dies. This implies the manufacturing process has been optimized and we can expect higher quality Turing silicon in the future.

Igor Wallossek with Tom's Hardware Germany has divulged that Nvidia will stop binning the Turing A-dies and also cease production on non-A dies that are used on RTX 2080 and 2070 cards. What this means is that going forward, Nvidia will offer only one variant of the Turing silicon for the cards.

Theoretically, this is good news for gamers, as the non-A die silicon was the cheaper, less capable variant that limited overclocking headroom. This should mean that the TSMC 12nm "FFN" process is mature enough to maximize yields, meaning Nvidia no longer needs to segregate the Turing silicon. The "new" silicon, reportedly already in production, will be the TU104-410 (RTX 2080) and TU106-410 (RTX 2070). Nvidia will reportedly sell the new dies at the same price point as the previous non-A counterparts.

The separate Turing dies were spotted by TechPowerUp some time ago as it was discovered that there were two different device IDs per Turing graphics card. For example, the RTX 2080 uses both the TU104-400-A1 and TU104-400A-A1 variants. By so doing, this allowed Nvidia to deliver higher quality chips to board partners for flagship models, and the cheaper non-A die chips for lower price points. There's no word on whether Nvidia plans to do the same with the RTX 2080 Ti and the TU102 die.

Permalink to story.

https://www.techspot.com/news/79939-nvidia-stop-producing-non-die-variants-turing-rtx.html

 

[ThanosPAS]

Does this apply for laptops as well?

 

[texasrattler]

When buying these cards, how are you going to know which you are getting?

 

[RayJayS1986]

When buying these cards, how are you going to know which you are getting?

In all likelihood, you won't. Not unless the retailer specifies anyway, and if they end up doing that, there will possibly even be a price hike for the "better" silicon, though in actuality, it'll probably just be on par with the previous "A" chips. An article I read elsewhere said this is probably due to TSMC's 12nm process maturing into a better state, to the point where the majority of chips are in better shape than previous, allowing all chips that pass evaluation to be "overclocking" chips. How true that is I don't know, but it certainly makes sense. But back to your question, you just won't know unless they specifically state. I wouldn't hold my breath......

 

[Evernessince]

When buying these cards, how are you going to know which you are getting?

No way to know for sure but price is the best indicator. The lower end cards had lower binned silicon. You have to pay over MSRP if you want the performance at the price reviews state.

 

[EEatGDL]

And what about memory chips? I plan to buy a 2080 Ti in a few months and I keep reading horror stories about Micron's GDDR6.

 

[JerryDobson]

So when I bought my Gigabyte card that said it was good for overclocking, and the chip is specifically not good for overclocking, well, I guess I was lied to and the product I bought misrepresented to me. I would not have bought it if I knew the chip was second rate. As the article in Tom's Hardware says, that's just business. Yeah, crooked business.

 

[ZackL04]

So when I bought my Gigabyte card that said it was good for overclocking, and the chip is specifically not good for overclocking, well, I guess I was lied to and the product I bought misrepresented to me. I would not have bought it if I knew the chip was second rate. As the article in Tom's Hardware says, that's just business. Yeah, crooked business.

These chips havent overclocked well since the 980ti. Dont expect much over factory boost 3.0

 

[GamerNerves]

Don't count on getting an A-tier one if you buy now or after some months. They still have plenty of non A-cards left. This is good news yes, but shouldn't encourage to buy these cards thinking that you get something better, nor that you have even a higher chance of getting something better.

 

[Evernessince]

And what about memory chips? I plan to buy a 2080 Ti in a few months and I keep reading horror stories about Micron's GDDR6.

Most vendors will use whatever chips they have on hand. That said I do not that some of the high end overclocking chips use Samsung exclusively. For example the 2080 ti Kingpin but expect to pay a huge premium, the card is $1,899.

 

[EEatGDL]

For example the 2080 ti Kingpin but expect to pay a huge premium, the card is $1,899.

Wow, and I thought the EVGA Hybrids (liquid cooled) were expensive. It seems that if I'm lucky I'll stick with the first GPU I get, I'm already mentally prepared for trying up to 4 GPUs through RMA.

 

[krizby]

There is no differences between non-A and A chip beside the power limits. My 2080 Ti non-A chip can reach the same max clock as A chip variant, just that the 280W PL prevent the clock to ever reaching max clock in any demanding game. For A chip though you can flash a higher PL bios to get additional performance at the price of higher power consumption (generally <10% performance for 30% more power consumption). All non-A bios come with the same PL and cannot be flashed with A chip bios.

How to distinguish A chip and non-A chip ? any OC version are A chip and standard version is non-A, for example MSI RTX 2070 Ventus OC is A chip while MSI RTX 2070 Ventus is non-A.

 

[JerryDobson]

CPU-Z will tell you what chip you have. It tells me I have a: TU106-400 A1 instead of TU106-400A A1

 

[Ru James]

For example the 2080 ti Kingpin but expect to pay a huge premium, the card is $1,899.

Wow, and I thought the EVGA Hybrids (liquid cooled) were expensive. It seems that if I'm lucky I'll stick with the first GPU I get, I'm already mentally prepared for trying up to 4 GPUs through RMA.

I got an EVGA 2070 xc a couple of months ago. It has Samsung memory in it, and it ran really well. Buut, the gpu wouldn't apply any overclock. GPUZ would see the core clock increase (theoretically) but real time measurement still showed the card maxed at the base clock, no matter what software I used or what amount of overclock or underclock I tried. However, the memory overclocked just fine.

Needless to say, I RMAd it. I actually did the EVGA step up program RMA and have a 2080 xc on the way with the cost of the difference, obviously. But, we shall see how that one goes. lol