Nvidia RTX 40-series allegedly getting down-binned GPU updates — certain 4060 and 4070 class cards to use larger, harvested chips

Nvidia GeForce RTX 4070
Nvidia GeForce RTX 4070 (Image credit: Amazon)

It's that time of the year again when Nvidia releases existing products with recycled or lower-binned silicon. According to hardware leaker MEGAsizeGPU, at least three of the chipmaker's GeForce RTX 40-series products, among the best graphics cards, will receive the treatment.

The GeForce RTX 4070 currently utilizes Nvidia's AD104 silicon, which also powers the other variants, including the GeForce RTX 4070 Super and GeForce RTX 4070 Ti. Nvidia will also start using the AD103 silicon for the GeForce RTX 4070. By contrast, AD103 is significantly larger than AD104 and houses more Streaming Multiprocessors (SMs), so it's a convenient way for Nvidia to maximize its silicon inventory.

The AD103 silicon is at the heart of the GeForce RTX 4070 Ti Super, GeForce RTX 4080, and GeForce RTX 4080 Super. Silicon that doesn't make the cut for the GeForce RTX 4080 is passed down to the GeForce RTX 4070 Ti Super. And now that Nvidia has added the GeForce RTX 4070 to the list, what doesn't qualify for the GeForce RTX 4070 Ti Super will find its way into the GeForce RTX 4070.

Swipe to scroll horizontally
Graphics CardPCBCurrent GPUNew GPURelease
GeForce RTX 4070PG141 SKU370AD104-251AD103-175-KXMarch 2H
GeForce RTX 4060 TiPG141 SKU359AD106-351AD104-150-KXApril
GeForce RTX 4060PG141 SKU368AD107-400AD106-255April

The present choice for the GeForce RTX 4060 Ti is the AD106 silicon. However, AD104 dies, which don't meet the requirements of the GeForce RTX 4070, will start finding their way into the GeForce RTX 4060 Ti. MEGAsizeGPU didn't specify whether the GeForce RTX 4060 Ti 16GB, which employs the same AD106 die, will face the same transition.

Meanwhile, the GeForce RTX 4060 is currently using the AD107 silicon but won't last long. Nvidia will substitute it with the AD106 silicon. It's the typical domino effect: Defective silicon from the GeForce RTX 4070 goes into the GeForce RTX 4060 Ti, and what isn't fit for the GeForce RTX 4060 Ti will end up in the GeForce RTX 4060.

The GeForce RTX 4070 (AD103) is slated for the second half of this month, so we should start seeing these graphics cards on the market. On the other hand, the GeForce RTX 4060 Ti (AD104) and GeForce RTX 4060 (AD106) won't show up until April. The latest GeForce 551.86 graphics driver, which came out on March 19, already supports the trio of reworked graphics cards, paving the way for a painless shift.

The only difference is the silicon inside the new graphics cards. Their specifications and performance remain unchanged, so the pricing should remain the same. Custom GeForce RTX 4070 models start at $519, whereas custom GeForce RTX 4060 Ti and GeForce RTX 4060 graphics cards have starting prices of $380 and $294, respectively.

Zhiye Liu
RAM Reviewer and News Editor

Zhiye Liu is a Freelance News Writer at Tom’s Hardware US. Although he loves everything that’s hardware, he has a soft spot for CPUs, GPUs, and RAM.

  • Eximo
    "Silicon that doesn't make the cute for the GeForce RTX 4080 is passed down to the GeForce RTX 4070 Ti Super"
    Reply
  • Eximo
    Should be interesting. Maybe get a 4070 that accepts a 4080 GPU waterblock. Pointless cost wise, but useful for SFF builds perhaps.
    Reply
  • TechLurker
    I'm sure the answer is "yes" in both cases, but I wonder if they are actually fusing off perfectly usable sections simply because they don't clock well at a given efficiency target, or if they're fusing off the unusable sections, all to prevent old-school GPU lotteries, where defective portions were technically usable and could sometimes result in a card being faster than baseline thanks to an extra bit of memory or raster cores that aren't entirely defective.

    I remember a time when certain ATI and NVIDIA cards were valued due to both companies having not bothered to completely fuse off/kill the defective sections, so someone could end up with a non-standard number of cores and memory that usually performed better than baseline. In rare cases, having the same amount as a card one step up but just clocked lower due to power issues (and the resulting deliberate OC'ing of voltage to make it match at the expense of energy cost).

    A more recent example was with Ryzen; IIRC, AMD had an unintended CPU lottery during the Ryzen 2000 era where some partially defective CCUs weren't fused off, so people got Ryzens with a non-standard number of cores or memory.
    Reply
  • Metal Messiah.
    The present choice for the GeForce RTX 4060 Ti is the AD106 silicon. However, AD104 dies, which don't meet the requirements of the GeForce RTX 4070, will start finding their way into the GeForce RTX 4060 Ti. MEGAsizeGPU didn't specify whether the GeForce RTX 4060 Ti 16GB, which employs the same AD106 die, will face the same transition.

    First off all, the original Tweet by the leaker doesn't seem to show up (this page doesn’t exist ?). Did he mention that only the 8GB variant would be affected by the change in silicon die, and not the 16GB card ?

    I can't find any info, and without any proof or backup, it is hard to trust this leak.

    Because it doesn't make any sense, since BOTH the 4060 Ti 8GB and 16GB cards use the same GPU silicon, albeit with a slightly different code, "AD106-351-A1" vs "AD106-350-A1".
    "351" die is being used in the 16GB SKU. So most likely he was referring the 16GB variant, since the 8GB RTX 4060 Ti card has been a hard sell, so Nvidia most likely won't give importance to this card.
    Reply
  • tamalero
    TechLurker said:
    I'm sure the answer is "yes" in both cases, but I wonder if they are actually fusing off perfectly usable sections simply because they don't clock well at a given efficiency target, or if they're fusing off the unusable sections, all to prevent old-school GPU lotteries, where defective portions were technically usable and could sometimes result in a card being faster than baseline thanks to an extra bit of memory or raster cores that aren't entirely defective.

    I remember a time when certain ATI and NVIDIA cards were valued due to both companies having not bothered to completely fuse off/kill the defective sections, so someone could end up with a non-standard number of cores and memory that usually performed better than baseline. In rare cases, having the same amount as a card one step up but just clocked lower due to power issues (and the resulting deliberate OC'ing of voltage to make it match at the expense of energy cost).

    A more recent example was with Ryzen; IIRC, AMD had an unintended CPU lottery during the Ryzen 2000 era where some partially defective CCUs weren't fused off, so people got Ryzens with a non-standard number of cores or memory.
    Wasnt it the legendary 9700 non pro that could be unlocked to full pro with a bios update?
    or AMD, where some of its cpus that used a pencil tip to enable the other core and overclocking?
    Reply
  • pixelpusher220
    TechLurker said:
    I'm sure the answer is "yes" in both cases, but I wonder if they are actually fusing off perfectly usable sections simply because they don't clock well at a given efficiency target, or if they're fusing off the unusable sections, all to prevent old-school GPU lotteries, where defective portions were technically usable and could sometimes result in a card being faster than baseline thanks to an extra bit of memory or raster cores that aren't entirely defective.
    Absolutely hate that the literally damage the cards to sell them cheaper. Almost should be illegal.

    Silicon Lottery could be a hella marketing and good will exercise. Wouldn't affect new sales much at all as it's just the lower binned stuff off a SKU
    Reply
  • russell_john
    Silicon that can't cut it for 4080 Super get cut down and used on 4070
    Silicon that can't cut it for 4070 Super get cut down and used on 4060

    It's a common thing to not waste silicon that's only slightly under spec by pushing it down the stack. I don't think it really has anything to do with when a new generation is coming but it might .... or it might be they have a lot of chips that just can't make the cut and don't want to waste them. Their A100 and H100 sales are still pretty strong and it could be they are doing this to juggle manufacturing time so they can concentrate on making H100's which are so profitable they can afford to take a slight loss on the consumer GPUs by down binning chips they already have stocked so they don't have to make more 4070 and 4060 chips and can concentrate on pumping out the H100
    Reply
  • Co BIY
    pixelpusher220 said:
    Absolutely hate that the literally damage the cards to sell them cheaper. Almost should be illegal.

    Silicon Lottery could be a hella marketing and good will exercise. Wouldn't affect new sales much at all as it's just the lower binned stuff off a SKU

    They literally do additional work to the damaged parts to make use of them in products for sale. If they allowed the crippled units to sell "as is" in lower priced products they would likely also have to create firmware to match each possible configuration.

    This is like a house builder salvaging something from a piece of damaged wood on a construction site. They cut off the damage and shorten it to the next needed length. They don't call the architect up and tell them to change the plans to accommodate a couple of extra "board feet".
    Reply
  • Co BIY
    russell_john said:
    It's a common thing to not waste silicon that's only slightly under spec by pushing it down the stack. I don't think it really has anything to do with when a new generation is coming but it might .... or it might be they have a lot of chips that just can't make the cut and don't want to waste them.

    Wouldn't surprise me if it has more to do with the workload of the engineering teams. At the mature end of a product cycle they can "glean the field" by reviewing the reject bins and make a plan to systematically integrate them into the product line.
    Reply
  • FoxTread3
    March 23, 2024 - It's nice to read an article that is positive about GPUs for once. It also shows how stupid I am rushing out to buy the newest latest thing. I'm glad that people that have waited to buy GPUs could benefit from down binning. I hope to hold out from buying a new GPU until the next generation of Nvidia GPUs is released. Then I'll buy a mid range GPU like the RTX 4070 Super Ti when I need it.
    Reply