RTX 5070 vs RX 9070 XT: 14% More Frames for 9% More Money
The AMD Radeon RX 9070 XT is the faster card: 130 fps at 1440p ultra against 114 for the NVIDIA GeForce RTX 5070 — 14% more performance for 9% more money ($599 MSRP vs $549 MSRP), with 16 GB of VRAM instead of 12 GB. Performance rising faster than price is rare, and it means the more expensive card here is also the better ratio.
Short version. AMD Radeon RX 9070 XT wins the matchup on frame rate, VRAM and performance per dollar. NVIDIA GeForce RTX 5070 wins on power draw, ray-tracing efficiency, DLSS coverage and CUDA.
The part the headline leaves out: AMD Radeon RX 9070 costs the same $549 MSRP as the NVIDIA GeForce RTX 5070 and is 7% faster with 16 GB and 30 W less power draw. At 22.2 fps per $100 it is the best ratio anywhere between $450 and $850.
| GPU | 1440p fps | VRAM | Price | fps / $100 | Value |
|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5070 | 114 | 12 GB | $549 MSRP | 20.8 | 69 |
| AMD Radeon RX 9070 | 122 | 16 GB | $549 MSRP | 22.2 | 74 |
| AMD Radeon RX 9070 XT | 130 | 16 GB | $599 MSRP | 21.7 | 72 |
| NVIDIA GeForce RTX 5070 Ti | 130 | 16 GB | $749 MSRP | 17.4 | 58 |
Performance figures: Tom's Hardware GPU hierarchy — geometric mean of 15 games at 1440p ultra, no upscaling. We do not test hardware ourselves; our methodology documents where every number on this site comes from.
Rasterization across all three resolutions
The 14% headline number is the 1440p figure, and it is the smallest of the three gaps. At 1080p the margin narrows to 11% because both cards start bumping into the CPU. At 4K it widens to 19% — 88 fps against 74.
The direction of that trend is the useful part. A gap that grows with resolution is a memory-system gap, not a shading gap: the 12 GB / 192-bit configuration on the NVIDIA GeForce RTX 5070 runs out of room before its shaders run out of work. A gap that shrinks with resolution would mean the opposite. So if you game at 1080p on a high-refresh monitor, the two cards are much closer than the headline suggests, and the rest of this article matters more than the frame rate does.
The card that beats both of them
AMD Radeon RX 9070 sits at $549 MSRP — the same price as the NVIDIA GeForce RTX 5070 — and posts 122 fps at 1440p. That is 7% more performance than the NVIDIA card at identical cost, with 4 GB more VRAM and 30 W less power draw.
Against its own XT sibling the arithmetic is just as awkward. The XT costs $50 more and returns 6% more frames for 38% more power. Same Navi 48 die, same 16 GB, 256-bit bus on both — the XT is the same chip pushed further up its voltage curve, and voltage curves are not linear. That is where 30% of the efficiency difference between them comes from.
The XT is still the right buy if you want the highest frame rate in this bracket without stepping up to NVIDIA GeForce RTX 5070 Ti at $749 MSRP. But "which of these two" is the wrong question if nobody has told you the third card exists.
Ray tracing: AMD posts the higher number and still loses the argument
In absolute terms the AMD Radeon RX 9070 XT is ahead here too: 75 ray-traced fps against 71. That is the opposite of what a decade of marketing on both sides would lead you to expect, and it is easy to over-read.
The number that actually describes ray-tracing hardware is how much of its own raster frame rate a card keeps once RT is switched on. NVIDIA GeForce RTX 5070 retains 62%. AMD Radeon RX 9070 XT retains 58%. AMD leads on RT frame rate only because it starts from a much higher raster baseline — the card is doing more work to arrive at a slightly better result.
Two things fall out of that chart that a head-to-head cannot show you. First, RDNA 4 is a real generational jump for AMD, not a marketing one: AMD Radeon RX 9070 XT at 58% is 19 points ahead of the RDNA 3 AMD Radeon RX 7900 XT, and it is ahead of NVIDIA's own previous-generation NVIDIA GeForce RTX 4070 SUPER. Second, NVIDIA's current lead is 4 points, not a category win. "NVIDIA for ray tracing" was true in 2023 and is now a tiebreaker rather than a reason.
Where those 4 points compound is heavy path tracing, where the RT penalty is several times larger than in a typical game and a small efficiency edge multiplies out. In the far more common case of a game with a couple of RT effects layered on, the raster lead decides it and AMD wins. Buy on the specific games you play, not on the category.
What the synthetic benchmark says, and why we do not lead with it
3DMark Steel Nomad scores these cards 7,249 to 5,256 — a 38% gap, nearly 2.7× the 14% we measure across actual games.
Neither number is wrong. Steel Nomad is a pure rasterization workload with no CPU dependency, no driver overhead worth speaking of and no engine-specific optimisation, and RDNA 4 is extremely strong in exactly that profile. Real games add all three of those things back, and the gap collapses to less than half. This is the single best argument for ranking hardware on game averages rather than synthetics, and it is why the tables on this site use a 15-game geometric mean. If you read a comparison quoting a 38% margin between these two cards, you are reading a synthetic score presented as a gaming result.
12 GB against 16 GB is the part you cannot fix later
12 GB on a $549 MSRP card in 2026 is the strongest argument against the NVIDIA GeForce RTX 5070, and it is a longevity argument rather than a benchmark one. Today it is enough at 1440p — that is what the frame rates above show. VRAM is not a performance dial that turns down gracefully; it is a cliff. Below the requirement a game stutters, swaps textures visibly, or drops to a lower texture tier automatically, and no amount of GPU horsepower behind the buffer changes that.
It is also the one specification you cannot add later, which is why it deserves more weight than a 14% frame-rate difference. Our VRAM guide works through what each tier actually buys you and where 12 GB starts to bite.
Power draw, and the efficiency ranking nobody prints
NVIDIA GeForce RTX 5070 draws 250 W and asks for a 600 W supply. AMD Radeon RX 9070 XT draws 304 W and asks for 700 W. On a five-year-old power supply or in a small case, 54 W is a real constraint and possibly an extra purchase, not a footnote.
Divide frame rate by watts and the order changes completely. AMD Radeon RX 9070 returns 0.55 fps per watt, NVIDIA GeForce RTX 5070 0.46, AMD Radeon RX 9070 XT 0.43, NVIDIA GeForce RTX 5070 Ti 0.43. The cheapest card of the four is also the most efficient by 22%, and both top-end parts sit at the bottom. That is what running silicon near the top of its voltage curve costs, and it shows up on your electricity bill and in your case temperatures rather than in any review headline.
Where NVIDIA still wins outright
- CUDA. If your work touches machine learning, 3D rendering or scientific tooling, much of the ecosystem is CUDA-first — not faster on NVIDIA, but supported on NVIDIA and sometimes only on NVIDIA. AMD's ROCm has narrowed that gap without closing it. If this applies to you it likely outranks everything else in this article.
- NVENC. NVIDIA's hardware encoder remains the reference for streaming and recording quality, and it is the reason a lot of streamers buy NVIDIA regardless of frame rate.
- DLSS coverage. Both vendors' upscalers are good now; DLSS is in more games and has a quality edge in most side-by-side testing. If you expect to run upscaled most of the time, that weighs more than a 14% native raster gap.
- Power and platform. 54 W less draw and a 100 W lower PSU recommendation, as above.
The rest of the bracket, plotted
Neither of these cards exists on its own shelf. There are 27 cards we track between $450 and $850, and plotting price against measured performance shows how much of that shelf is simply not worth buying.
The cluster along the bottom is previous-generation stock still listed at close to its original price. NVIDIA GeForce RTX 3080 at $699 MSRP returns 13.8 fps per $100 against the AMD Radeon RX 9070's 22.2 — a card from 2020 asking 2026 money. The four current-generation parts sit near the line; almost everything else sits well below it.
Which one should you actually buy
- Best value in the bracket, no conditions: AMD Radeon RX 9070 at $549 MSRP. Best fps per dollar, best fps per watt, 16 GB, and it costs the same as the NVIDIA GeForce RTX 5070.
- You want the fastest card here and $50 is not decisive: AMD Radeon RX 9070 XT. 6% more frames, and the better choice if you play at 4K where its lead is largest.
- Your work uses CUDA, or you stream: NVIDIA GeForce RTX 5070. Buy it for the ecosystem, the encoder and the power envelope. Do not buy it expecting to win the frame-rate comparison, because it does not.
- You want NVIDIA and the frame rate: NVIDIA GeForce RTX 5070 Ti at $749 MSRP. 130 fps with 16 GB — the card that genuinely matches the XT. It costs $150 more, which tells you where NVIDIA itself prices this performance tier.
- Heavy path tracing specifically: NVIDIA GeForce RTX 5070 Ti or above. This is the one workload where NVIDIA's remaining RT-efficiency lead compounds enough to flip the result, and the NVIDIA GeForce RTX 5070 does not have the raster headroom to absorb the penalty.
Common questions
Is the RX 9070 XT worth $50 more than the RTX 5070?
On frame rate, yes: 14% more performance for 9% more money is a better ratio, not a worse one. But the honest answer is that neither is the value pick — AMD Radeon RX 9070 at $549 MSRP beats both on fps per dollar.
Is 12 GB of VRAM enough for 1440p in 2026?
Yes, today. The frame rates in the table above are measured with 12 GB and they are fine. The question is 2029, and 12 GB is where the risk sits — not because games need more now, but because console memory budgets set the floor and that floor only moves up.
Does the RTX 5070 beat the RX 9070 XT in ray tracing?
No, not in frame rate — 71 fps against 75. It is more RT-efficient, keeping 62% of its raster performance versus 58%, which matters in path-traced titles and almost nowhere else.
Why does this comparison disagree with the 3DMark numbers I have seen?
Because Steel Nomad puts these cards 38% apart and a 15-game average puts them 14% apart. Synthetics remove CPU dependency, driver overhead and engine-specific optimisation, all of which reappear in real games.
Do these prices hold?
The table marks which figures are manufacturer MSRP and which are observed street prices; in this bracket most are still MSRP. Street pricing moves, and a $50 swing changes the recommendation here — the ratios are close enough that the ranking is a pricing outcome, not a permanent one. See MSRP vs street price for how far the two diverge.
Full side-by-side with every specification: NVIDIA GeForce RTX 5070 vs AMD Radeon RX 9070 XT.