Cartes graphiques : le comparatif performances mis à jour

A couple of weeks ago, I posted [this performance chart](https://www.reddit.com/r/nvidia/comments/1mx49ey/performance_chart_for_gamers_looking_to_buy_used/), based on **aggregated benchmark results**, to be able to better **compare the gaming performance of the various Nvidia GPUs**.

Based on the feedback I got from that project, I have now revised and expanded the ranking, to include not only Nvidia GPUs but also those from **AMD** and **Intel**. You can access this new ranking, together with all the data it is based on, via [this link](https://docs.google.com/spreadsheets/d/e/2PACX-1vRgu7g3uNNa8RpIC-SQr-PC9IcduMzVWaZNDG3qjEab1AkdoU6fuKPufPdto2_FS7ICPFiFhF1ixO8i/pubhtml).

The list is not complete, but includes most of the graphics cards released from 2015 and onwards, even including some professional cards, mining cards et cetera.

The main purpose of this exercise is *not* to aid dick-swinging regarding who has the best GPU, but rather to **aid people who are in the market for used GPUs to better assess the relative price-to-performance between various offerings**. Ie, the important thing to take away from this aggregation is not that the 8GB 5060 Ti is ranked higher than the 8GB 9060 XT, for example, but rather that they are very, very close to each other in performance.

Furthermore, the linked spreadsheet contains specific rankings for **1080p**, **1440p** and **4K**, though these (especially the 1080p one) are based on fewer benchmarks and are thus not as reliable as the overall chart.

You can read more about [the methodology](https://www.reddit.com/r/nvidia/comments/1nki8qo/comment/ney1hg9/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button) in my comments to this post, but the most important thing is that the raw performance score is **pure raster performance** (no upscaling, no ray tracing, etc) based on data from eight different 3DMark benchmarks (two are 1080p, two are 1440p and four are 4K) as well as the techpowerup performance ranking.

This raw performance score is then adjusted for 1) punishing cards with less than 16GB of VRAM and 2) features and functionalities (such as upscaling tech, I/O support and raytracing). How much weight to assign each of these factors will always be more or less arbitrary and heavily dependent on use case, but I’ve tried to be as methodical and factually grounded as I can.

**Note:** GPUs listed in parentheses are ones where the benchmark data was scarce (based on a small number of benchmark runs) and/or had to be inferred from other scores. The ratings for these GPUs (such as the non-XT 9060) are thus to be taken with a reasonable pinch of salt.

**EDIT:** Several people have commented that the aggregated benchmark results would be more reliable if I only based them on benchmark runs conducted at core GPU clock and memory clock settings. While true in theory, it is not so in practice. See [this comment](https://www.reddit.com/r/nvidia/comments/1nki8qo/comment/nf2w6lc/?context=3&utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button) for more information (and a bonus comparison spreadsheet!).

11 replies

  1. Samantha Thompson · 5 days ago

    J’ai une RTX 4070 depuis 3 ans et mon intention est de passer à une 5080 ou une 5070 TI, ou autre chose.

    Peut-être que j’attendrai la sortie de la RTX 6000.

    AMD est aussi intéressant, mais je suis un peu un adepte d’Intel/Nvidia.

  2. Adèle Gerard · 5 days ago

    Il existe une BIOS XOC publiquement disponible pour le 5090D, mais pas pour le 5090.

  3. Ryan Ryan · 3 days ago

    Y a-t-il quelque chose de faux avec la liste de techpowerup puisqu’elle affiche à peu près la même chose ?

  4. Ida Heß · 3 days ago

    On dirait un groupe d’yeux malveillants qui fixent. Je te nomme Le Regardant.

  5. Brooklyn Hopkins · 3 days ago

    enlevez les D et le 6000… ce n’est pas pertinent pour les personnes normales.

  6. Nicole Cruz · 2 days ago

    s’il vous plaît, connectez-le maintenant aux prix du marché en direct… 🙂

  7. Miyuki Nishiyama · 2 days ago

    donc je suppose que vous n’avez pas effectué les tests vous-même. Les sources seraient utiles, en particulier pour voir quand les tests ont été effectués. Les performances évoluent avec les mises à jour des pilotes et les tests effectués avec des pilotes de version finale ne reflètent peut-être plus les performances actuelles.

    Il semble également que vous utilisiez des tests dont il n’est pas clair si les cartes étaient surclockées. D’un côté, cela est intéressant car peu de comparaisons prennent en compte le surclocking.

    D’un autre côté, les résultats obtenus avec des modifs de shunt et un refroidissement au gaz ne seraient pas très représentatifs.

    Donc… merci pour votre travail et tout, mais je recommanderais la liste TPU (qui contient également des points de données autres que les scores 3DMark).

Laisser un commentaire

Votre adresse e-mail ne sera pas publiée. Les champs obligatoires sont indiqués avec *