RTX 2080 Ti
11 GB → 22 GB
22 GB of CUDA memory has no new or used alternative in India below ₹74,999. This is the mod that actually changes what you can run — 32B-class models at Q4 stay resident.
Indicative ₹22,000 – ₹26,000
We de-solder the 1 GB memory chips from your graphics card and fit 2 GB chips in the same footprint, then move the board's memory straps so the card reports the new capacity. Same GPU core. Same clock. Double the VRAM.
Built for people running LLMs, Stable Diffusion, Blender and video work on one card — not for gaming. If a used card is the better buy for you, we will say so before you spend anything.
11 GB → 22 GB
22 GB of CUDA memory has no new or used alternative in India below ₹74,999. This is the mod that actually changes what you can run — 32B-class models at Q4 stay resident.
Indicative ₹22,000 – ₹26,000
8 GB → 16 GB
16 GB at 448 GB/s — about 1.5× the token throughput of a new RTX 4060 Ti 16GB for memory-bound generation, for less money.
Indicative ₹15,000 – ₹22,000
8 GB → 16 GB · 6 GB → 12 GB
Same technique, smaller boards. Whether your specific model is eligible depends on the PCB revision — send a photo of the memory area and we will tell you.
Quote per card
Indicative bands only. Memory-chip prices move weekly and every card is quoted individually before you send anything. A failed bench test is not charged.
The reason most people are nervous about a modded card is that sellers publish nothing. Every card that leaves this bench ships with:
We are building the first batch now. A deposit is the only number that means anything to us: an email list is free to join. It is refundable on request at any point before your card is opened, and in full if we do not deliver.
No. It is a modification. Your card will report the new capacity and the GPU core is untouched, but it is not a factory product, it has no manufacturer warranty, and it should be treated as a mod. We tell you that in writing, on the card, before you send it.
It happens — a small percentage of cards fail on the bench. You are not charged, and the card is returned to you at its original capacity with its report. This is why every card is baselined on arrival and why the deposit is refundable instead of non-refundable.
If you own a 3070 and can find an A4000 16GB at the stock-and-price that gets advertised, yes — it is the same GA104 die with ECC and pro drivers, and that is a better deal than the mod. We will tell you that when you ask. The mod makes sense for 22GB cards, where no CUDA alternative exists below ₹74,999, and for 3070 owners who want the higher memory bandwidth rather than the lower price.
More VRAM does not raise average frame rates in most games, and the money is better spent on a faster card. The workloads this fixes are the ones that simply refuse to load: LLM inference, fine-tuning, large scenes, high-res SD batches and video timelines.
Bengaluru first, walk-in, so there is no courier in the loop while we prove throughput. Insured mail-in across India after that. Courier insurance matters — standard uninsured liability is capped at ₹5,000, so it is quoted into the service.
Terms for the first batch are being finalised and will be published before anyone pays for the rework itself. Nobody pays for a mod until those terms are in writing. The ₹500 deposit is refundable regardless.
Where this stands: this is a demand test, not a shop with a queue. The rework process is documented and the bench is still being built out — the first batch has not been run yet. If you reserve and we cannot serve you, your money comes back in full.