Thursday, February 14th 2019

NVIDIA TU116 GPU Pictured Up Close: Noticeably Smaller than TU106

Here is the first picture of NVIDIA's 12 nm "TU116" silicon, which powers the upcoming GeForce GTX 1660 Ti graphics card. While the size of the package itself is identical to that of the "TU106" on which the RTX 2060 and RTX 2070 are based; the die of the TU116 is visibly smaller. This is because the chip physically lacks RT cores, and only has two-thirds the number of CUDA cores as the TU106, with 1,536 against the latter's 2,304. The die area, too, is about 2/3rds that of the TU106. The ASIC version of TU116 powering the GTX 1660 Ti is "TU116-400-A1."

VideoCardz scored not just pictures of the ASIC, but also the PCB of an MSI GTX 1660 Ti Ventus graphics card, which reveals something very interesting. The PCB has traces for eight memory chips, across a 256-bit wide memory bus, although only six of them are populated with memory chips, making up 6 GB over a 192-bit bus. The GPU's package substrate, too, is of the same size. It's likely that NVIDIA is using a common substrate, with an identical pin-map between the TU106 and TU116, so AIC partners could reduce PCB development costs.
Source: VideoCardz
Add your own comment

35 Comments on NVIDIA TU116 GPU Pictured Up Close: Noticeably Smaller than TU106

#26
sam_86314
londiste said:
You are right, we don't. 2048 is unlikely though - GP104 has 2560 and Turing CUDA cores are far larger than Pascal's.
When trying to do high-level theorycrafting on SM sizes it seems that Volta SM is ~47% larger than Pascal's. Turing's SM is ~9% bigger than Volta's and ~60% bigger than Pascal.
Looks like Turing SM’s are smaller than Pascal. Pascal has 128 shaders per SM while Turing has 64.
Posted on Reply
#27
londiste
sam_86314 said:
Looks like Turing SM’s are smaller than Pascal. Pascal has 128 shaders per SM while Turing has 64.
Oh, sorry. I meant at the same CUDA core count. Pascal SM has 64, Turing SM has 32.
Two Turing SMs with the same CUDA core count as one Pascal SM is ~60% bigger than Pascal.
Posted on Reply
#28
efikkan
I wonder when Nvidia will phase out the term "cores" for GPUs. It used to refer to the combined FPU/ALU units, but in Turing these are now separated, which is one of the reasons why Turing have more performance per "core".
Posted on Reply
#29
yakk
I can't imagine nvidia wanting to voluntarily compete on price with the 590, maybe same price but probably higher, cause they can.
Posted on Reply
#30
jabbadap
sam_86314 said:
Looks like Turing SM’s are smaller than Pascal. Pascal has 128 shaders per SM while Turing has 64.
Well there's one Pascal gpu that have 64 shader SM: GP100 is with same 64 fp32 units per SM as Turing. For what is worth, RT cores are one per SM, the more SMs the more gpu have RT cores, thus Turing till now have had that 64 shaders per SM. Does that mean TU116 have to had 64 fp32 core SM, I would say no it may go back to 128 shader SM structure. Afaik not even tu102, tu104 and tu106 have same structure: tu102 have six 12 SMs GPC, tu106 has three 12 SMs GPC and tu104 has six 8 SMs GPC. Just realized that tu116 will be half of tu104, thus 3*8*64 = 1536 cuda cores, or if it have 128 cc per SM 3*4*128.

londiste said:
Oh, sorry. I meant at the same CUDA core count. Pascal SM has 64, Turing SM has 32.
Two Turing SMs with the same CUDA core count as one Pascal SM is ~60% bigger than Pascal.
Uhm no that is not correct. From pascal family gp100 has 64 cuda cores per SM rest have 128 cudas per SM. All turings now have 64 cudas per SM.
Posted on Reply
#31
lexluthermiester
efikkan said:
I wonder when Nvidia will phase out the term "cores" for GPUs. It used to refer to the combined FPU/ALU units, but in Turing these are now separated, which is one of the reasons why Turing have more performance per "core".
And they might get sued like AMD for claiming something that special-snowflakes don't agree with..
Posted on Reply
#32
Chloe Price
Hm, I was sure that it's just a TU106 with different name and the RT shaders disabled.
Posted on Reply
#33
Hardcore Games
Given the TU1116 is so new, to be fair game engines will need time to be able to leverage it better.

My GTX 1060 is last year's technology but it is not dead yet.

Maybe later this year I will get a new card but I am in no rush.
Posted on Reply
#34
Darmok N Jalad
Chloe Price said:
Hm, I was sure that it's just a TU106 with different name and the RT shaders disabled.
Anandtech has TU116 at 284mm2, making it 40% smaller than TU106. It has the architectural improvements of Turing, but none of the RT hardware.
Posted on Reply
#35
Chloe Price
Darmok N Jalad said:
Anandtech has TU116 at 284mm2, making it 40% smaller than TU106. It has the architectural improvements of Turing, but none of the RT hardware.
I know, that was just my guess when we didn't know much about the card.
Posted on Reply
Add your own comment