Japanese repair shop sells GPU VRAM upgrades for $25 per GB during memory crisis — RTX 2080 Ti modded to 22GB of GDDR6 for just $282, double the VRAM creates a budget AI powerhouse

The GeForce RTX 2080 Ti is somewhat of a legend in the modding community when it comes to VRAM upgrades. The card’s strange 11GB memory config lends itself very well to repair jobs that double its memory pool to 22GB. We’ve seen it happen in China and the UAE before, and now, a Japanese vendor offering the same services has been spotted. The shop charges 45,000 Yen, or $282, for adding 11GB of memory to your existing RTX 2080 Ti, which comes out to about $25 per gigabyte.

The shop has a very uninspired name that roughly translates to “AI to GPU Expert,” since they take regular gaming cards and transform them into AI workhorses with a lot of VRAM. They operate out of Saitama Prefecture and have a store on Mercari where you can find a bunch of different modded GPUs listed for rather reasonable prices.

The cheapest pre-modded RTX 2080 Ti we could find was an MSI Venus 2X model for $428. Just a few days ago, we saw blower-style 22GB RTX 2080 Ti SKUs listed for $500 on eBay from a Hong Kong seller. It’s hard to get your hands on memory for a fair price these days, but it seems like this vendor has a very generous plug based on their prices.

They even posted about recently receiving a large shipment of 2GB GDDR6 modules made by Samsung, all intended for upgrading RTX 2080 Ti cards. The batch included 200 modules, so 400GB of extra VRAM that should be enough to upgrade 36 different RTX 2080 Ti cards to 22GB. Just an hour later, they had already upgraded the first card:

Replying to a comment, the shop clarified it charges 45,000 Yen for this service, which works out to $282.48 — that’s $25.68 per GB of extra VRAM, or $12.84 per GB for the entire 22GB pool. All things considered, not a bad price at all. You can very easily find stock RTX 2080 Ti cards for less than $250 on the secondhand market, so the total cost of a 22GB card would work out to a little over $500.

At that point, it’s better to buy one of the pre-modded units, but if you already have a 2080 Ti and wanted a boost in AI performance, you have the option. The extra VRAM won’t really help in games because the cut-down TU102 silicon inside the 2080 Ti is not powerful enough to benefit from 22GB of VRAM. Even in demanding, ray-traced titles, you won’t saturate the memory pool, and the GPU itself will become the bottleneck.

Keep in mind that your memory bandwidth is still locked to 616 GB/s after the upgrade, so anything that relies on speed will not see a noticeable improvement. On the other hand, more memory will greatly help AI models in both the training and inference stages. An RTX 2080 Ti is almost as powerful as an RTX A4000 in these workloads, but that has less VRAM and costs more. But if money is no object, perhaps a 44GB RTX 2080 Ti is more enticing to you.