Not long following the RTX 3080 and 3090 debuted, rumors surfaced of superior VRAM variants that would supposedly seem in-sector by December. The reasoning was apparent — AMD is envisioned to start GPUs with a lot more VRAM than Nvidia’s equivalents. This is a regarded technique for Team Crimson, and they’ve deployed it continuously with the Polaris GPU family members from the GTX 1050 and 1060.
One reason a superior VRAM version of an Ampere GPU is so attention-grabbing is that it opens up a realm of extra opportunities in AI operate. According to our sister site PCMag, this program has been at least temporarily scuttled. They state: “Two impartial sources have confirmed the new RTX cards are no for a longer period coming to sector.”
Nvidia, even so, isn’t the only enterprise to blame for this generation trouble. Micron’s yields on GDDR6X are also reportedly poor. The variance concerning GDDR6 and GDDR6X is demonstrated down below:
GDDR6X is faster-clocked GDDR6, with larger RAM bandwidth for every pin and for every placement, and a significantly more compact electric power usage profit than the bounce from GDDR5 to GDDR5X (Presumably pJp stands for picoJoules for every little bit). With GDDR5X, Micron slash what I’m assuming is for every-little bit electric power by 11.12 per cent. With GDDR6X, the enterprise only managed to minimize electric power by 3.5 per cent.
The major innovations of GDDR6X are its new encoding scheme and the long-expression likely to double bandwidth for every pin compared with current GDDR6, but these enhancements come with their possess difficulties as far as hitting efficient produce. Up till now, we have centered on the Nvidia aspect of the equation, but Micron may be actively playing a aspect here as well. If yields on the RAM chips are very low, and just about every RTX 3080 / 3090 superior-VRAM GPU requires two times as significantly, that would proficiently slash Nvidia’s currently-confined potential to get cards into sector.
I would go so far as to argue that it tends to make a lot more perception that this is a VRAM lack problem than an RTX GPU problem. We don’t have any reason to imagine the qualification problems are on Nvidia’s aspect of items. Dividing a confined selection of Ampere GPUs concerning SKUs will make it more durable to discover areas. But needing to divert two times as significantly VRAM to just about every card could choke generation even even more if need is larger for individuals cards. Whilst we have no insight into the particulars of the scenario, it could quickly be actively playing a aspect.
If rumors are accurate, AMD will dodge this lack by relying on regular GDDR6 at decreased full bandwidth compared with cards like the RTX 3090. This is not automatically a undesirable issue — larger memory bandwidth is only useful to a GPU if the card is bandwidth-sure in the very first place.
It would not surprise me if these cards at some point do develop into accessible, after the generation-line problems are solved. A superior VRAM Ampere would be a charming aspect to possess for specified workloads.