Il RTX 5080 e il RTX 4080 Super occupy the exact same slot in NVIDIA’s lineup — the $999 enthusiast card one tier below the flagship. Both carry 16 GB of VRAM. So the AI buyer’s question is simple: does Blackwell bring enough to justify choosing the 5080, or is the 4080 Super still the smart pickup?
The short answer: the 5080 is the better card, but the upgrade gap is narrower than the generation number suggests.
Punti chiave
- Both cards have 16 GB VRAM — identical model-size ceiling.
- The RTX 5080’s GDDR7 memory delivers ~960 GB/s vs the 4080 Super’s ~736 GB/s — a real ~30% bandwidth jump.
- Expect ~15–20% faster LLM inference on the 5080, driven mostly by bandwidth.
- Blackwell adds native FP4 support — useful for next-gen quantized models, irrelevant today.
- If you already own a 4080 Super, do not upgrade. If you are buying fresh, the 5080 is the better long-term card.
At a glance
| Specifiche | RTX 5080 | RTX 4080 Super |
|---|---|---|
| Architettura | Blackwell GB203 | Ada Lovelace AD103 |
| Core CUDA | 10,752 | 10,240 |
| VRAM | 16 GB GDDR7 | 16 GB GDDR6X |
| Larghezza di banda della memoria | ~960 GB/s | ~736 GB/s |
| FP16 Tensor (dense) | ~450 TFLOPS | ~390 TFLOPS |
| Low-precision | FP8 + FP4 | FP8 |
| TDP | 360 W | 320 W |
| Price | $999 | $999 |
16 GB: the shared ceiling
Neither card is a big-model machine. 16 GB of VRAM comfortably handles:
- Llama 3 8B at 8-bit, or 13B-class models at 4-bit
- Stable Diffusion XL e Flux.1 image generation
- LoRA fine-tuning of 7B–8B models
Neither card runs a 70B model in VRAM. If that is your goal, you want a 24 GB or 32 GB card and should stop reading here. For everyone else — the large majority of local AI users — 16 GB is the practical sweet spot, and both cards deliver it.
Where Blackwell pulls ahead: bandwidth
The CUDA-core counts are nearly identical (10,752 vs 10,240), so raw shader power is close. The real generational change is memory bandwidth. LLM token generation is memory-bound — the GPU spends most of its time reading weights, not computing — so the 5080’s GDDR7 advantage shows up directly:
| Workload | RTX 5080 | RTX 4080 Super |
|---|---|---|
| Llama 3 8B Q4_K_M | ~125 tok/s | ~108 tok/s |
| Llama 3 13B-class Q4 | ~78 tok/s | ~66 tok/s |
| SDXL 1024×1024 (30 steps) | ~14 it/s | ~13 it/s |
| Flux.1 dev (1024px) | ~3.1 s/image | ~3.5 s/image |
Note the split: LLM inference sees the biggest gains (~15–20%) because it is bandwidth-bound, while Stable Diffusion — which is compute-bound — shows only a small lead since the core counts are so close.
FP4: a feature for tomorrow
Blackwell introduces native FP4 (4-bit floating point) tensor operations. On paper this doubles low-precision throughput versus FP8. In practice, as of 2026, almost no mainstream inference stack ships production FP4 kernels for consumer workloads. It is a genuine advantage, but a future-facing one — it will matter more in 2027 than it does today.
If you keep GPUs for four or five years, FP4 support is a real reason to favor the 5080. If you upgrade every cycle, it is close to irrelevant.
Power and efficiency
The 5080 draws 360 W versus the 4080 Super’s 320 W. Blackwell is more efficient per operation, but the 5080 spends that headroom on higher clocks, so absolute draw is up. Both are happy on an 850 W PSU. Neither is a thermal problem in a well-ventilated case.
Scegli la RTX 5080 se
- You are buying fresh and want the longer-lived card
- Your main workload is LLM inference (bandwidth-bound)
- You want FP4 readiness for future quantized models
Choose the RTX 4080 Super if
- You find one discounted below $850 as stock clears
- Your focus is Stable Diffusion, where the gap is tiny
- You already own one — there is no reason to upgrade
The 16 GB warning
Whichever you choose, understand the limitation you are buying into. 16 GB is increasingly tight for 2026 AI work. Larger image models, longer LLM context windows, and fine-tuning all push against that ceiling. If your budget can stretch to a 24 GB RTX 4090 or 32 GB RTX 5090, the capacity headroom outlasts the speed difference between these two 16 GB cards.
The price reality: what you actually pay, and which to buy
Spec sheets only matter once you factor in price, and this is where the two cards split decisively. They are no longer competing on the same shelf: the RTX 5080 is the current product, while the RTX 4080 Super has been discontinued and now lives almost entirely on the secondhand market. That changes the question from “which is faster” to “which makes sense at the price you can actually find.”
The 5080 carries a $999 MSRP, but Blackwell supply has stayed tight because NVIDIA prioritised enterprise AI silicon, so real street prices have sat well above sticker for most of 2026 — frequently in the $1,150–$1,250 range. The 4080 Super, by contrast, has settled into a used-market groove around $850–$900, with new old-stock units commanding inflated, often nonsensical, scalper pricing. So in practice you are weighing a new ~$1,200 card against a used ~$870 one.
Here is the honest way to decide:
- Buy the RTX 5080 if you want a warranty, the newest software path (the 5th-gen Tensor cores and FP4 support are a forward-looking bet), and the genuine ~30% memory-bandwidth uplift that helps inference throughput. It is the right call for a fresh build where you would be buying a new GPU anyway.
- Buy a used RTX 4080 Super if value-per-dollar for AI is the priority. You give up bandwidth and FP4, but you keep the same 16 GB ceiling — which is the real limiter for model size — and pocket roughly $300. For running quantised 7B–14B models and Stable Diffusion, that gap rarely shows up in everyday use.
- Do not “upgrade” from a 4080 Super to a 5080. Selling one to buy the other nets a single-digit-to-low-double-digit performance change for a real cash outlay. Put that money toward a 24 GB card instead, where the extra VRAM unlocks models neither 16 GB card can touch.
One wrinkle worth flagging: a rumoured RTX 5080 Super with 24 GB of GDDR7 has circulated, but it has been delayed indefinitely amid GDDR7 supply constraints, so it is not a card you can plan a purchase around today. If 16 GB is genuinely too tight for your workload, the answer is a 24 GB-class GPU now — not waiting on an unconfirmed launch.
Domande frequenti
Is the RTX 5080 worth upgrading to from a 4080 Super?
No. Both have 16 GB, and the 5080 is only ~15–20% faster. That is not enough to justify the cost of a full GPU swap. Upgrade only if you are jumping two tiers, to a 24 GB or 32 GB card.
Can the RTX 5080 run Llama 3 70B?
No. 70B at 4-bit needs roughly 40 GB. The 5080’s 16 GB forces heavy CPU offload, which is slow. For 70B in VRAM you need an RTX 5090 (32 GB) or a multi-GPU build.
Does FP4 support matter in 2026?
Not yet for most users. FP4 is real and future-proof, but production inference stacks have not widely adopted it. Treat it as insurance for 2027, not a feature you will use this year.
Which is better for Stable Diffusion, the 5080 or 4080 Super?
They are nearly tied. Stable Diffusion is compute-bound and the two cards have almost identical CUDA-core counts. The 5080 leads by only ~5–8%.
Is a used RTX 4080 Super a smart buy for AI in 2026?
For many people, yes. It shares the 5080’s 16 GB VRAM ceiling — the factor that actually decides which models you can load — while typically costing a few hundred dollars less on the secondhand market. You sacrifice the 5080’s higher memory bandwidth and FP4 support, but for running quantised 7B–14B models and Stable Diffusion that trade-off is easy to live with. Buy from a seller with returns, and stress-test the card on day one.
Should I wait for the RTX 5080 Super with 24 GB before buying?
We would not plan around it. A 24 GB GDDR7 “5080 Super” has been rumoured, but reports point to an indefinite delay tied to GDDR7 memory supply, so there is no reliable date. If 16 GB is enough for your models, buy a 5080 or a used 4080 Super now. If you genuinely need more than 16 GB, get a 24 GB-class card today rather than betting on an unconfirmed release.
Why does the RTX 5080 cost more than its $999 MSRP?
Because supply has been constrained. NVIDIA shifted much of its manufacturing toward enterprise AI accelerators, leaving consumer Blackwell cards in short supply, so the 5080 has frequently sold above its $999 sticker — often around $1,150–$1,250 — through 2026. Always budget against the real street price you can find in stock, not the MSRP, when comparing it to a used 4080 Super.
Verdict
For a fresh purchase, the RTX 5080 is the right call: same price as the 4080 Super, meaningfully more memory bandwidth, and FP4 headroom for the future. But this is an evolutionary step, not a revolution — anyone already running a 4080 Super should keep it. And both buyers should weigh the same hard truth: 16 GB is the real constraint here, and no amount of Blackwell polish changes that ceiling.