Used RTX A6000 vs RTX 5090: The 42.5GB Line

Forty-two and a half gigabytes. That single number settles most of this purchase, and almost nothing else about either card does.

The exact Llama 3.3 70B Q4_K_M file is 42.52GB of weights. It does not fit on a 32GB RTX 5090. It does not fit across two 16GB cards, because those still add up to 32GB. A 48GB RTX A6000 is the only card in this comparison that holds it — and the A6000 is a 2020 board.

Everything that fits 16GB belongs to the RTX 5080.

Some retailer links below are affiliate links; if you buy through them TechFuel HQ may earn a small commission at no extra cost to you. As an Amazon Associate I earn from qualifying purchases. Commissions never influence what gets recommended — see our disclosure.

The line, and which cards cross it

Memory and power figures come from the manufacturer spec tables on the NVIDIA RTX A6000, RTX 5090 and RTX 5080 pages; PNY publishes 768GB/s of bandwidth and a dual-slot active fansink for its A6000 board.

The fit is binary and the price is not. That is the whole shape of this decision: one card clears a hard capacity threshold, two do not, and the cheapest of the three is the one you probably want.

16GB is already fast, which is the awkward part

Our RTX 5080 throughput dataset recorded gpt-oss 20B at MXFP4 decoding 178 to 190 tokens per second across two dated captures, with the whole model sitting in 13,381 to 14,067 MiB of the card’s 16GB. Qwen 2.5 14B at Q4_K_M decoded at 94 to 97 tokens per second. Every cell is the median of three runs, and the published CSV carries the driver version, the Ollama build, the model digests and the resident memory readings.

That result is the reason the 48GB card is not an automatic upgrade. A 20B model and a 14B model are already resident and already quick on a $1,399.99 card. Paying nearly four times that for 48GB buys nothing at all unless the extra capacity changes which model can stay loaded. The VRAM-tier guide maps the full ladder, and the VRAM calculator works out the context-sensitive version of the same question.

Skip the RTX 5090 for 70B work

This is the finding that surprised us. If you are buying a card to reach a 70B model, the RTX 5090 is the worst of the three, and it is not close.

At $4,829.99 it costs $3,430 more than the RTX 5080 — and it stops 10.52GB short of the file that would justify the upgrade. It buys a bigger pool that lands on the wrong side of the threshold this page is about, then asks 575W to do it. Meanwhile the A6000 sits $660 above it and actually clears the line at 300W.

One band does belong to it, and the RTX 5080 cannot take it: a working set between 16GB and 32GB — a 32B model at Q4, say — is too big for 16GB and nowhere near needing 48GB. If that is your workload, the 5090 is the only card in this table that fits it. Nothing here recommends it above that band, and no matched throughput number for it was captured on this page.

Availability made the point again. All three RTX 5090 card rows in the August feed were checked on September 7 and every one had gone out of stock. If you are shopping for a 32GB pool you are paying a scarcity price for capacity that does not reach.

Buy the RTX 5080

Start at $1,399.99 and stop there unless your model genuinely will not fit 16GB.

That is the recommendation for nearly everyone reading this. A 20B model runs at 178 to 190 tokens per second on it by our own measurement, it draws 360W, it is a current-generation card with a current warranty, and it costs less than a third of either 48GB or 32GB alternative.

Thirteen RTX 5080 cards were in stock in the August feed between $1,399.99 and $1,999.99, so there is real room to shop the exact board. Prices move — check before you order.

The one case for going after a used A6000

Buy 48GB when your resident working set lands above 32GB and comfortably below 48GB. That is a narrow window and it is the only window. A dense 70B at Q4 fits it. A 20B does not, and a 120B does not either.

Watch the headroom while you are in there. The weights take 42.52GB and leave 5.48GB for the KV cache and runtime buffers. Ordinary context lengths live in that space; long ones do not, and the failure shows up as an out-of-memory error partway through a session rather than at load time. Work out your context budget before you commit to the card, not after.

Three things the A6000 gives you beyond raw capacity: ECC memory, a 300W ceiling that fits a normal power supply, and a two-slot board in a market where the fast consumer cards have grown to three and four. NVIDIA also supports it on the RTX Enterprise driver branch rather than the Game Ready branch, which is the right trade for a machine that works and the wrong one for a machine that games.

What to make the seller put in writing

A used A6000 is a private-party or reseller transaction, and the coverage rides entirely on that paperwork. PNY publishes a three-year limited warranty on a new card, and NVIDIA’s warranty policy sends most claims to the reseller or board partner first. Neither of those facts follows a card to its second owner on its own. So before you pay, get:

1. the exact part number and a photo of the serial;
2. whether the card is used, refurbished, open-box, or pulled from a workstation;
3. the return window, and who pays return shipping;
4. the warranty provider and the remaining duration;
5. enough of that return window left to run a full 48GB memory test and an hour of sustained load.

The fifth one is the one people skip. Run it the day the card arrives.

And know the ceiling. A new PNY board lists at $5,490.00 — $660 above the RTX 5090 and roughly four times the RTX 5080. A used card asking within a few hundred dollars of that is not a deal; the discount is the entire reason to accept a five-year-old GPU with someone else’s hours on it. If you would rather have the new-card warranty than the savings, this is the exact board:

How this was checked

The capacity argument uses one exact model artifact rather than a rounded parameter count: the Hugging Face repository reports 42,520,398,816 bytes for Llama-3.3-70B-Instruct-Q4_K_M.gguf, which is 42.52GB. Memory, power and cooling figures were read off the current NVIDIA and PNY specification tables on September 7. Throughput comes from our own RTX 5080 dataset, captured on an RTX 5080 under Ollama with the method and raw CSV published alongside it — no A6000 was benchmarked here, which is why this page recommends the 48GB card on capacity and never on speed.

Prices are the lowest in-stock USD offer per exact model from the August 17 Newegg feed, restricted to actual video cards. No used price appears anywhere on this page: every used-A6000 listing checked on September 7 was expired, dead, or unreachable. That is itself the finding. Used A6000 supply is thin and it moves, so treat any single listing you find as a sample of one and price it against the $5,490.00 new-card ceiling above.

This page is re-checked every six months, and sooner if a new consumer card moves the 32GB ceiling or a major 70B release changes file size.