GPU XON THE BENCH
GeForce RTX 3070 8GB
8GB GDDR6
Benchmark FPS at 1080p
212FPS
- Our rating /100
- 63.3/100
- 3070s we've tested
- 244
Historical total for this model.
Explore this GPUFOR THE LOVE OF THE UPGRADE
RTX 3070 8GB vs RTX 5090 32GB
RTX 5090 32GB leads this benchmark at 784 FPS, compared with 212 FPS for RTX 3070 8GB. That is a 572 FPS difference in GPUsed’s continuous 1080p FurMark workload.
Powered by GPUsed01 / THE BENCH
GPU X
GeForce RTX 3070 8GB
GPU Y
GeForce RTX 5090 32GB
QUICK PICKS257 models. Pick your next move.
GPU YBENCHMARK LEAD
GeForce RTX 5090 32GB
32GB GDDR7
Benchmark FPS at 1080p
784FPS
- Our rating /100
- 100/100
- 5090s we've tested
- 46
Historical total for this model.
Explore this GPURTX 3070 8GB records 212 FPS versus 784 FPS for RTX 5090 32GB. GPU Y is the percentage reference.
RECORDED BENCHMARKSGPUsed’s recorded 1080p FurMark benchmark averages, not predicted gaming FPS. Quantities are GPUsed’s supplied historical model totals. Exact-model and shared-family historical totals are identified on each card; they are not the retained sample size of the displayed average.
THE MATCH-UP
RTX 3070 8GB vs RTX 5090 32GB: the context
The RTX 5090 is far ahead of the RTX 3070 in the supplied benchmark, with four times the graphics-memory capacity. This compares a previous-generation 70-class card with a newer flagship. It shows the scale of the measured difference, rather than suggesting every RTX 3070 system needs such an expensive class of upgrade.
The exact cards, at a glance
| Detail | RTX 3070 8GB | RTX 5090 32GB |
|---|---|---|
| Memory | 8GB GDDR6 | 32GB GDDR7 |
| Documented release | first documented retail availability. documented market. | first documented retail availability. documented market. |
| GPUsed archetype | Workhorse | Flagship |
| GPUsed Legend Rating Editorial legacy rating | 6.5 /10 | 7.5 /10 |
| More about this GPU | Read the RTX 3070 8GB profile | Read the RTX 5090 32GB profile |
How to read this comparison
The benchmark is GPUsed’s continuous 1080p FurMark workload. The figures compare that workload, rather than predict FPS in a particular game. The percentage on the bench uses GPU Y as its reference: (GPU X FPS ÷ GPU Y FPS − 1) × 100. Swap the cards and the reference changes.
Our rating compresses the benchmark scale out of 100. It does not drive the percentage calculation. GPUsed’s historical quantities retain their model or shared-family labels; they are not the retained sample size of the displayed FPS average.
Read how the benchmark and ratings workAnother opponent?
Keep one card on the bench and see how the nearby choices compare.
Sources and checks
Figures and variant matching checked . GPUsed supplies the benchmark averages and historical quantities. The references below support the GPU identity and specification context.
- RTX 3070 8GB: nvidia.com
- RTX 3070 8GB: arstechnica.com
- RTX 5090 32GB: nvidia.com
- RTX 5090 32GB: arstechnica.com
The pair note comes from GPUsed’s reviewed comparison material, with current variant and numeric checks. Individual GPU profiles retain their own documented release qualifications and review dates.
02 / A DIFFERENT VIEW
Put the numbers in perspective.
XRTX 3070 8GB
212FPS
YRTX 5090 32GB
784FPS
03 / KEEP IT ON THE BENCH
Worth another look.
Your saved match-ups.
Saved on this device
Keep a match-up here. Pin the ones worth coming back to.
04 / UNDER THE HOOD
Geek out.
Get the context.
The right numbers are useful.
Knowing what they mean is better.
What are these FPS readings?
GPUsed’s recorded average benchmark results from continuous FurMark runs at 1080p. They compare graphics cards under this workload. A game’s settings, your CPU, ray tracing and upscaling can all change in-game performance.
All 257 selectable variants use GPUsed’s supplied results. The reported runs were at 1080p and typically lasted 30–60 minutes. We do not add test sessions or infer gaming frame rates from those results.
Why does swapping change the percentage?
GPU Y is the reference. We divide GPU X’s FPS by GPU Y’s FPS, subtract one and multiply by 100. A positive figure means X leads; a negative figure means Y leads. Swapping changes the reference, so it changes the percentage too.
What is our rating out of 100?
It is the existing compressed benchmark scale, with the fastest card in GPUsed’s dataset at 100. It gives older cards more room on the scale. The percentage difference is always calculated from FPS, never from the rating.
What do the model test counts tell me?
These are GPUsed’s supplied historical totals for each model or recorded model family. They are distinct from the retained sample size of a particular FPS average. They show GPUsed’s experience with each model; they do not tell you how many readings went into a particular FPS average.