NVIDIA GeForce RTX 2080 Ti TU102 Die-size Revealed

[btarunr]

Here are some of the first pictures of NVIDIA's GeForce RTX 2080 Ti ASIC, probably codenamed "TU102." GamersNexus took a ruler to this chip, and discovered that it's one of the biggest non-storage chips in existence. The rectangular die measures 31 mm x 25 mm, or 775 mm². The package has no IHS, but a metal brace along the periphery of the fiberglass substrate distributes mounting pressure from the cooler. NVIDIA is building the "Turing" family of GPUs on TSMC 12 nm FinFET node.



View at TechPowerUp Main Site

 

[Tsukiyomi91]

that is a big boi... =O

 

[Vya Domus]

Making the IHS redundant

 

[DeathtoGnomes]

HUUUGE, Youuuuge!

 

[Fluffmeister]

That ruler... rules.

 

[Gasaraki]

Yet people complain that it's "too expensive" like someone is point a gun at their head making them buy an expensive video card.

 

[Vayra86]

Yet people complain that it's "too expensive" like someone is point a gun at their head making them buy an expensive video card.

It still is tho

Whether you buy one or not is irrelevant to make that statement. Its price hike and its super obvious

 

[DeathtoGnomes]

Yet people complain that it's "too expensive" like someone is point a gun at their head making them buy an expensive video card.

i dont see people complain about the truth, whine is prolly more like it.

 

[Vya Domus]

Yet people complain that it's "too expensive"

So, because the die is big the price is alright ? Funny, no one thought that about Vega and that one didn't even come with a price premium. The color of the sticker really decides everything after all.

 

[Fluffmeister]

Sadly the price hike was always going to happen, if Vega actually troubled the $699 GTX 1080 Ti they might of felt the need to lower it's price. Now they can cash in and go higher.

It's not cool, but it makes sense, their shareholders will at least be happy which is more important than appeasing the odd forum warrior.

 

[londiste]

Revealed?
Nvidia themselves told publicly that TU102 is 754mm².

 

[DeathtoGnomes]

prices :

 

[londiste]

Now, give us a confirmed die size for TU104 and then we'd be talking news.

There are some pretty good pictures of the GTX 2080 PCB. We know from data sheet that Micron's GDDR6 size is 14.0mm x 12.0mm.
Doing some measuring on the images TU104(-400A-A1) should be 24.3mm x 22.6mm. That gives us the size of approximately 549mm².

 

[DRDNA]

That ruler... rules.

Right i spent at least 5 minutes of time with an enlarged image to study the beauty of that ruler...drooling over it too

 

[londiste]

So, because the die is big the price is alright ? Funny, no one thought that about Vega and that one didn't even come with a price premium. The color of the sticker really decides everything after all.

You know, Vega is actually a pretty good parallel. With Vega, AMD tried to introduce new features at least some of which would need some industry adoption (DSBR and Primitive Shaders respectively are the main ones). Unfortunately, they simply failed to get these implemented.
Nvidia is similarly trying to introduce new features (RT, DLSS) and whether Turing fails or succeeds is up to whether these new features get adopted.

Vega is the same size as GP102 (GTX 1080Ti and Titan XP), both in terms of transistor count and die size. Same power envelope. Same theoretical performance indicators (Single Precision GFLOPS, Memory bandwidth, Texel Fillrate), excels in some (Half/Double precision GFLOPS) and falls short on a few (Pixel Fillrate). For this exact reason it was a big surprise that it did not perform similarly and only competes to much smaller and weaker (and cheaper, at least in manufacturing cost) GP104 (GTX 1080).

 

[TheLostSwede]

Right i spent at least 5 minutes of time with an enlarged image to study the beauty of that ruler...drooling over it too

Then this is for you - https://blogs.nvidia.com/blog/2017/12/22/nvidia-ruler/

 

[phanbuey]

The guy in the picture clearly doesn't measure things often...

 

[Supercrit]

The guy in the picture clearly doesn't measure things often...

It's just for an idea for the scale of things, like putting next to a banana.

 

[DeOdView]

That > > RULE ...ER!
BTW what's that BIG shinny thing on the back ground?!

:Hick:hicks

 

[BadFrog]

Its price hike and its super obvious

Gotta recoup that R&D somehow.

Right i spent at least 5 minutes of time with an enlarged image to study the beauty of that ruler...drooling over it too

Buy one on Ebay! Search Nvidia ruler. I think alot of them are knock off's now but you might be able to find an original. They were selling for more than 100 when they were first spotted in the wild awhile back.

 

[Keullo-e]

Yet people complain that it's "too expensive" like someone is point a gun at their head making them buy an expensive video card.

If you can get a good gaming PC with it's price, IT IS expensive as crap.

 

[Fx]

You know, Vega is actually a pretty good parallel. With Vega, AMD tried to introduce new features at least some of which would need some industry adoption (DSBR and Primitive Shaders respectively are the main ones). Unfortunately, they simply failed to get these implemented.

I have a feeling that this initiative is too big to fail.

 

[Vayra86]

Gotta recoup that R&D somehow.



Buy one on Ebay! Search Nvidia ruler. I think alot of them are knock off's now but you might be able to find an original. They were selling for more than 100 when they were first spotted in the wild awhile back.

Right. No, because that premium was already earned a million times over with their Quadro and Titan line up. What happens here is that you overpay for vaporware that tequires dedicated hardware to run proper. It directly impacts the net performance of the products in their primary use case and steals TDP budget from it.

RT cores are detrimental to GPU performance for every game that does not use it. The net resilt is that you pay for a tier higher because of RT, yet you get performance of a tier below it.

The real question is: are a few RT effects that valuable versus the current effects in use? To me that is rhetorical, and the obvious answer is no.

 

[xorbe]

RT cores are detrimental to GPU performance for every game that does not use it. The net resilt is that you pay for a tier higher because of RT, yet you get performance of a tier below it.

There is one benefit, if the RT cores are inactive, the extra silicon makes it easier to cool the gpu otherwise.

 

[Vayra86]

There is one benefit, if the RT cores are inactive, the extra silicon makes it easier to cool the gpu otherwise.

Oh, silicon cools a GPU now?

That's just saying the same in a different way; if the RT cores weren't there, you could use the die space for real performance without a higher OC and its uncertainties