Just a day after the announcement – and about three weeks before the first card hits stores – Nvidia is already receiving criticism for its new graphics cards. Which they strictly get quite often.
The critics are speculating this time whether Nvidia, by launching two such different 4080 cards, is not trying to deceive users.
The lowest-specified 4080 card should have been called 4070 instead, several people think.
Well-known tech YouTube channels such as Hardware Unboxed, Paul’s Hardware, Gamers Nexus and Linus Tech Tips all have something to say about the issue.
Generally, cards with the same model number will also have more or less the same processor, while the amount of memory may vary. For example, the RTX 3080 is available in two different variants, one with 12 GB of memory and another with 10 GB of memory. The GPU itself is not exactly the same, but almost, with only slight differences in the number of Tensor cores, RT cores and CUDA cores.
With the 4080, on the other hand, there will be two variants that have 16 and 12 GB of memory respectively. But as Youtuber JaysTwoCents points out, only the amount of memory is stated on the box. In this country, it is much less common than in the USA to buy graphics cards in physical stores, but it is still absolutely possible to go on a spree here.
Performance will most likely be quite different from the two 4080 cards, just as the price is.
The differences between the two cards are quite large. The 12 GB variant is made on the AD104 chip, while the 16 GB version is made on the AD103. This is a clear indication that these are two different chip designs, and not just a reduced version of the same chip.
As for the 4080 cards, there are still many details that are not known, but something is already nailed down. For example, the 12 GB card gets over 2,000 fewer CUDA cores than its big brother with 16 GB. The GA104 chip is probably physically smaller than the GA103, and thus also cheaper to produce – you get more chips per wafer. There will also be fewer circuits on it, allowing it to run slightly faster as well, which may compensate – if only slightly – for the dramatically lower number of CUDA cores.
Traditionally, the 104 GPUs have also been found in the 70 series of cards, the critics point out. The RTX 3070, for example, is built on the GA104, while the GTX 1070 had the GP104. No rule without exception, admittedly: The chip in the RTX 2070 was the TU106.
|Graphics card:||RTX 4080 16 GB||RTX 4080 12GB|
|Recommended price:||From NOK 15,400||From NOK 11,600|
|GHz, Base:||2.21 GHz||2.31 GHz|
|GHz, Boost:||2.51 GHz||2.61 GHz|
|Remind:||16 GB GDDR6X||12 GB GDDR6X|
|Power consumption:||340 watts||285 watts|
|“Power limit”:||516 watts||366 watts|
Not the first time
It also happens that faulty goods from the niche are used in graphics cards in lower niches, in order to limit losses. If a piece is not quite good enough, or you have too many of it, it may be useful in a less expensive card, even if the normal is actually a smaller and cheaper piece. This is called “cutting down” the chip, as certain circuits on it are cut with a laser, so that it corresponds to the chip it is to replace, and any circuits with errors are disconnected.
This is exactly what Nvidia did with the RTX 3070 in 2021. Then GA104 chips that were originally intended for the RTX 3070 and 3060 Ti began to appear in RTX 3060 cards.
It’s not the first time Nvidia has resorted to similar tricks. The classic example is the GTX 1060, which was launched in a version with 6 GB of memory and a variant with 3 GB of memory. The GTX 1060 became very popular, and is still the most used GPU on Steam, for example. On this card too, it was more than just the amount of memory that made the difference. For example, the 6GB variant had 1280 CUDA cores, while the little brother had to make do with 1152. And that’s even though both cards were built around the GP106 chip.