GPU Benchmarks and Hierarchy 2023: Graphics Cards Ranked
On all of the current Nvidia and AMD graphics cards, we ran hundreds of GPU benchmarks before classifying them in our thorough hierarchy.
(new tab opens)
(new tab opens)
(new tab opens)
(new tab opens)
(new tab opens)
2023 GPU BENCHMARKS FOR HIERARCHY
All of the best graphics cards are ranked at the top of our GPU benchmarks hierarchy for both current and previous-generation graphics cards. Even the best CPUs for Gaming play a supporting role when it comes to performance, whether you’re playing games or working on demanding creative projects like 4K video editing.
Welcome to 2023 as well! Retesting all of the GPUs is now being done utilising a Core i9-13900K processor in place of the present Core i9-12900K. However, it is far from finished, therefore for the time being, we are using our 2022 test suite. We have updated our list to include the RTX 4070 Ti and RX 7900 XTX/XT, bringing us mostly up to date with testing.
The whole GPU hierarchy for conventional rendering, also known as rasterization, is presented first, followed by the hierarchy for the ray tracing GPU benchmarks. These, of course, need a GPU that can support ray tracing, so the only options are AMD’s RX 7000/6000-series, Intel’s Arc, and Nvidia’s RTX cards. The findings are all without the individual cards’ DLSS, FSR, or XeSS features enabled, mind you.
The RTX 40-series’ newest capabilities, notably DLSS 3 Frame Generation, are powered by Nvidia’s new Ada Lovelace architecture. The RX 7000-series is powered on AMD’s RDNA 3 architecture, albeit just two cards have been made available thus far. While adding a third dedicated GPU participant to the fray, Intel’s Arc Alchemist architecture is more of a rival for midrange products from the previous generation.
On page two, you may get our 2020-2021 benchmark suite, which uses a Core i9-9900K testbed to run our earlier test suite on all GPUs from the previous generation. For reference, we also offer the previous GPU hierarchy (ordered by theoretical performance without benchmarks).
According to our performance-based GPU gaming benchmarks at 1080p “extreme” for the main suite and 1080p “medium” for the DXR suite, the following tables order everything exclusively by performance. The rankings presented here do not take into account features, pricing, or the amount of power used by the graphics card. We’ve updated our test suite, swapped to a new Alder Lake Core i9-12900K testbed, and retested all of the previous GPU generations. Let’s move on to the tables and benchmarks now.
2023 GPU Benchmarks Ranking
We test (almost) all GPUs at 1080p medium and 1080p extreme for our most recent benchmarks, and the table is sorted according to the 1080p ultra results.
We also test at 1440p ultra and 4K ultra when it makes sense. All of the ratings are scaled in relation to the RTX 4090, which is the highest-scoring 1080p ultra card in our updated suite (at least at 4K and 1440p).
Swipe through the above gallery to view the 1080p medium, 1440p, and 4K extreme pictures. You can also view the summary table above, which compares the relative performance of the GPUs we tested throughout the previous three hardware generations. The GT 1030, RX 550, and a few Titan cards are among the possibilities that are absent, but otherwise it is mostly full. For some of the additional (older) GPUs, we do have information in the table below.
Borderlands 3 (DX12), Far Cry 6 (DX12), Flight Simulator (DX11/DX12), Forza Horizon 5 (DX12), Horizon Zero Dawn (DX12), Red Dead Redemption 2 (Vulkan), Total War Warhammer 3 (DX11), and Watch Dogs Legion are the eight titles we’re utilising for our normal GPU benchmarks hierarchy (DX12). The geometric mean (equal weighting) of the eight games determines the frame rate.
Hierarchy of Tom’s Hardware Rasterization GPU Benchmarks
Specifications for a graphics card in 1080p Ultra, 1080p Medium, 1440p Ultra, and 4K Ultra (Links to Review)
GeForce RTX 4090 (opens in new tab): 100.0% (147.8fps), 100.0% (186.9fps), 100.0% (143.1fps), and 100.0% (116.0fps) at 2520MHz with 24GB of GDDR6X memory operating at 21Gbps and 450W. (opens in new tab)
Radeon RX 7900 XTX (opens in new tab) Navi 31, 12288 shaders, 2500MHz, 24GB [email protected], 960GB/s, 355W, 96.0% (141.9fps), 93.6% (174.9fps), 91.9% (131.4fps), and 81.8% (94.8fps) (opens in new tab)
Radeon RX 6950 XT (opens in new tab): Navi 21, 5120 shaders, 2310MHz, 16GB [email protected], 576GB/s, 335W, 95.7% (141.5fps), 102.0% (190.6fps), 82.7% (118.3fps), and 62.0% (71.9fps) (opens in new tab)
GeForce RTX 4080 (opens in new tab): 78.7% (91.2fps), 90.9% (130.1fps), 97.8% (182.7fps), AD103, 9728 shaders, 2505MHz, 16GB [email protected], 717GB/s, 320W (opens in new tab)
Radeon RX 6900 XT (opens in new tab) Navi 21, 5120 shaders, 2250MHz, 16GB [email protected], 512GB/s, 300W, 92.0% (136.0fps), 101.2% (189.1fps), 76.5% (109.5fps), and 56.4% (65.4fps) (opens in new tab)
Radeon RX 7900 XT (opens in new tab): Navi 31, 10752 shaders, 2400MHz, 20GB [email protected], 800GB/s, 315W, 91.2% (134.9fps), 93.2% (174.2fps), 83.2% (119.1fps), and 68.6% (79.6fps) (opens in new tab)
GeForce RTX 3090 Ti (opens in new tab): 67.1% (77.9fps), 81.0% (115.8fps), 95.4% (178.3fps), 89.0% (131.6fps), and 1860MHz, 24GB [email protected] (opens in new tab)
GeForce RTX 4070 Ti (opens in new tab): 88.0% (130.1fps), 90.9% (169.9fps), 79.7% (114.1fps), and 62.7% (72.8fps) at 2610MHz with 12GB of GDDR6X memory operating at 21Gbps and 285W. (opens in new tab)
Radeon RX 6800 XT (opens in new tab): Navi 21, 4608 shaders, 2250 MHz, 16 GB [email protected] Gbps, 512 GB/s, 300W, 87.6% (129.4 fps), 98.1% (183.3 fps), 72.2% (103.3 fps), 52.6% (60.9 fps) (opens in new tab)
GeForce RTX 3090 (opens in new tab): 61.1% (70.8fps), 76.0% (108.8fps), 96.4% (180.2fps), 1695MHz, 24GB [email protected], and 350W (opens in new tab)
GeForce RTX 3080 Ti (opens in new tab): 74.2% (106.2 frames per second), 85.4% (126.2 frames per second), 94.8% (177.2 frames per second), 59.4% (68.9 frames per second), GA102, 12GB [email protected], 350W (opens in new tab)
GeForce RTX 3080 12GB (opens in new tab): GA102, 8960 shaders, 1845 MHz, 12GB [email protected], 912GB/s, 400W, 85.1% (125.8fps), 96.5% (180.3fps), 74.0% (105.9fps), and 58.7% (68.1fps) (opens in new tab)
GeForce RTX 3080 (opens in new tab): GA102, 8704 shaders, 1710MHz, 10GB [email protected], 760GB/s, 320W, 81.1% (119.9fps), 94.1% (175.8fps), 69.3% (99.1fps), and 54.1% (62.8fps) (opens in new tab)
Radeon RX 6800 (opens in new tab) Navi 21, 3840 shaders, 2105MHz, 16GB [email protected], 512GB/s, 250W, 79.6% (117.7fps), 94.8% (177.1fps), 64.1% (91.7fps), and 45.7% (53.0fps) (opens in new tab)
GeForce RTX 3070 Ti (opens in new tab): GA104, 6144 shaders, 1770MHz, 8GB [email protected], 608GB/s, 290W, 71.3% (105.4fps), 87.0% (162.6fps), 58.2% (83.2fps), and 40.1% (46.5fps) (opens in new tab)
Radeon RX 6750 XT (opens in new tab): Navi 22, 2560 shaders, 2600MHz, 12GB [email protected], 432GB/s, 250W, 71.2% (105.3fps), 91.1% (170.2fps), 54.7% (78.2fps), and 37.1% (43.1fps) (opens in new tab)
Titan RTX (opens in new tab): TU102, 4608 shaders, 1770MHz, 24GB [email protected], 672GB/s, 280W, 68.3% (101.0fps), 84.7% (158.2fps), 56.2% (80.5fps), and 41.6% (48.3fps) (opens in new tab)
Radeon RX 6700 XT(opens in new tab) 67.5% (99.8fps) 86.7% (162.1fps) 51.3% (73.4fps) 34.9% (40.5fps) Navi 22, 2560 shaders, 2581MHz, 12GB [email protected], 384GB/s, 230W(opens in new tab) (opens in new tab)
GeForce RTX 3070(opens in new tab) 67.5% (99.8fps) 84.4% (157.7fps) 54.2% (77.5fps) 37.2% (43.2fps) GA104, 5888 shaders, 1725MHz, 8GB [email protected], 448GB/s, 220W(opens in new tab) (opens in new tab)
GeForce RTX 2080 Ti(opens in new tab) 64.9% (96.0fps) 81.1% (151.6fps) 52.6% (75.3fps) 38.5% (44.6fps) TU102, 4352 shaders, 1545MHz, 11GB [email protected], 616GB/s, 250W(opens in new tab) (opens in new tab)
GeForce RTX 3060 Ti: 61.9% (91.5 fps), 80.1% (149.7 fps), and 48.7% (69.7fps)
8GB [email protected], 448GB/s, 1665MHz, 4864 shaders, GA104, 200W (opens in new tab)
Radeon RX 6700 10GB (opens in new tab) Navi 22, 2304 shaders, 2450MHz, 10GB [email protected], 320GB/s, 175W, 59.3% (87.7fps), 78.6% (146.8fps), 44.4% (63.5fps), and 28.6% (33.1fps)
GeForce RTX 2080 Super (opens in new tab) TU104, 3072 shaders, 1815MHz, 8GB [email protected], 496GB/s, 250W, 57.4% (84.9fps), 73.8% (137.8fps), 45.4% (64.9fps), and 29.8% (34.5fps) (opens in new tab)
GeForce RTX 2080: 55.6% (82.2 fps), 71.3% (133.1 fps), and 43.6% (62.4fps)
TU104, 1710MHz, 8GB [email protected], 448GB/s, 215W, 2944 shaders (opens in new tab)
Intel Arc A770 16GB (opens in new tab): ACM-G10, 4096 shaders, 2100MHz, 16GB [email protected], 560GB/s, 225W, 54.3% (80.3fps), 66.4% (124.0fps), 43.6% (62.3fps), and 31.7% (36.7fps) (opens in new tab)
Radeon RX 6650 XT: 54.0% (79.8 fps), 74.1% (138.4 fps), and 39.7% (opens in new tab) (56.7fps)
Navi 23, 8GB [email protected], 280GB/s, 2635MHz, 2048 shaders, and 180W (opens in new tab)
Radeon RX 6600 XT: 52.8% (78.0 fps), 73.1% (136.5 fps), and 38.4% (opens in new tab) (54.9fps)
Navi 23, 8GB [email protected], 256GB/s, 2589MHz, 2048 shaders, and 160W (opens in new tab)
GeForce RTX 2070 Super: 51.7% (76.4 fps), 66.4% (124.1 fps), and 40.2% (57.4fps)
8GB [email protected], 448GB/s, TU104, 2560 shaders, 1770MHz, and 215W (opens in new tab)
Radeon RX 5700 XT (opens in new tab) Navi 10, 2560 shaders, 1905MHz, 8GB [email protected], 448GB/s, 225W, 49.8% (73.7fps), 67.3% (125.8fps), 37.3% (53.3fps), and 25.2% (29.3fps) (opens in new tab)
48.3% (71.4fps), 62.3% (116.4fps), 38.6% (55.2fps), 27.4% (31.8fps), Intel Arc A750(opens in new tab), ACM-G10, 3584 shaders, 2050MHz, 8GB [email protected], 512GB/s, 225W (opens in new tab)
GeForce RTX 3060: 47.5% (70.2fps), 63.6% (118.8fps), and 36.8% (52.6fps)
[email protected], 12GB, 360GB/s, 170W, GA106, 3584 shaders (opens in new tab)
Radeon VII (opens in new tab) Vega 20 3840 shaders 1750MHz 16GB [email protected] 1024GB/s 300W 47.1% (69.7fps) 61.0% (114.0fps) 37.0% (53.0fps) 27.1% (31.4fps) (opens in new tab)
GeForce RTX 2070 (opens in new tab): 35.7%, 59.2%, and 45.9% (67.9 fps) (51.0fps)
8GB [email protected], 1620MHz, TU106, 2304 shaders, 448GB/s, and 175W (opens in new tab)
Radeon RX 6600 (opens in new tab): 32.2%, 63.0%, and 45.1% (66.7 frames per second) (46.1fps)
Navi 23, 2491MHz, 8GB [email protected], 224GB/s, 132W, 1792 shaders (opens in new tab)
GeForce GTX 1080 Ti (opens in new tab): GP102, 3584 shaders, 1582MHz, 11GB [email protected], 484GB/s, 250W, 45.0% (66.5fps), 59.2% (110.6fps), 35.1% (50.3fps), and 25.5% (29.5fps) (opens in new tab)
GeForce RTX 2060 Super: 44.0% (65.1 fps), 56.7% (105.9 fps), and 33.7% (48.2fps)
TU106, 1650MHz, 8GB [email protected], 448GB/s, 175W, 2176 shaders (opens in new tab)
Radeon RX 5700 (opens in new tab): 33.0%, 59.5%, and 43.9% (64.8 frames per second) (47.2fps)
Navi 10, 8GB [email protected], 448GB/s, 1725MHz, 2304 shaders, and 180W (opens in new tab)
Radeon RX 5600 XT: 39.3% (58.1fps), 53.8% (100.6fps), and 29.4% (42.0fps)
Navi 10, 8GB [email protected], 336GB/s, 2304 shaders, 1750MHz, and 160W (opens in new tab)
Radeon RX Vega 64 (opens in new tab) 4096 shaders, 1546 MHz, 8 GB [email protected], 484 GB/s, 295W, 38.4% (56.8fps), 50.5% (94.3fps), 29.1% (41.6fps), and 20.2% (23.5fps) (opens in new tab)
GeForce RTX 2060: 37.3% (55.2 fps), 51.8% (96.8 fps), and 27.1% (38.7fps)
1920 shaders, 1680MHz, 6GB [email protected], 336GB/s, and 160W for the TU106 (opens in new tab)
GeForce GTX 1080: 35.9% (53.1 fps), 48.2% (90.0 fps), and 27.6% (opens in new tab) (39.4fps)
8GB [email protected], 320GB/s, GP104, 2560 shaders, 1733MHz, and 180W (opens in new tab)
GeForce RTX 3050: 34.8% (51.4 fps), 47.8% (89.4 fps), and 26.3% (37.6fps)
224GB/s, 8GB [email protected], 1777MHz, GA106, 2560 shaders, 130W (opens in new tab)
GeForce GTX 1070 Ti: 34.6% (51.1 fps), 45.9% (85.8 fps), and 26.5% (37.9fps)
8GB [email protected], 256GB/s, GP104, 2432 shaders, 1683MHz, and 180W (opens in new tab)
Radeon RX Vega 56 (34.2%), 50.6% (84.4%), and 25.8% (opens in new tab) (37.0fps)
Vega 10, with 3584 shaders, 1471 MHz, 8 GB of HBM2 at 1.6 GB/s, and 210 watts (opens in new tab)
GeForce GTX 1660 Super: 30.7% (45.3 fps), 44.3% (82.8 fps), and 22.6% (32.4fps)
TU116, 1785MHz, 1408 shaders, 6GB [email protected], 336GB/s, and 125W (opens in new tab)
GeForce GTX 1660 Ti: 30.4% (45.0 fps), 44.1% (82.4 fps), and 22.5% (32.2fps)
1536 shaders, 1770 MHz, 6 GB of GDDR6 at 12 GB/second, 120 W, and TU116 (opens in new tab)
GeForce GTX 1070 (30.3% (44.8fps), 40.2% (75.1fps), and 23.1% (opens in new tab) (33.1fps)
1920 shaders, 1683MHz, 8GB [email protected], 256GB/s, and 150W on GP104 (opens in new tab)
GeForce GTX 1660: 27.2% (40.2fps), 40.2% (75.1fps), and 19.9% (28.5fps)
6GB [email protected], 192GB/s, 1408 shaders, 1785MHz, TU116, 120W (opens in new tab)
Radeon RX 5500 XT 8GB: 26.9% (39.8 fps), 38.8% (72.6 fps), and 19.9% (28.5fps)
Navi 14, 8GB [email protected], 224GB/s, 130W, 1408 shaders, 1845MHz (opens in new tab)
Radeon RX 590: 26.6% (39.4 fps), 36.7% (68.6 fps), and 20.3% (29.1fps)
8GB [email protected], 256GB/s, Polaris 30, 2304 shaders, 1545MHz, and 225W (opens in new tab)
GeForce GTX 980 Ti: 24.3% (35.9 fps), 33.5% (62.6 fps), and 18.6% (26.7fps)
2816 shaders, 1075MHz, 6GB [email protected], 336GB/s, and 250W for the GM200 (opens in new tab)
34.5% (35.4 fps) Radeon R9 Fury X (opens in new tab) 24.0% (64.4fps)
Fiji, 4GB [email protected], 512GB/s, 4096 shaders, 1050MHz, and 275W (opens in new tab)
Radeon RX 580 8GB, 23.9% (35.3 fps), 33.0% (61.7 fps), and 18.2% (opens in new tab) (26.0fps)
8GB [email protected], 256GB/s, Polaris 20, 2304 shaders, 1340MHz, and 185W (opens in new tab)
GeForce GTX 1650 Super: 22.9% (33.9 fps), 36.4% (68.0 fps), and 16.1% (23.0fps)
TU116, 4GB [email protected], 192GB/s, 1280 shaders, 1725MHz, and 100W (opens in new tab)
22.6% (33.5 fps) 35.8% Radeon RX 5500 XT 4GB (opens in new tab) (66.9fps)
Navi 14, 4GB [email protected], 224GB/s, 1845MHz, 1408 shaders, and 130W (opens in new tab)
GeForce GTX 1060 6GB: 21.8% (32.2 fps), 31.1% (58.0 fps), and 16.1% (23.0fps)
1280 shaders, 1708MHz, 6GB [email protected], 192GB/s, and 120W on the GP106 (opens in new tab)
Radeon RX 6500 XT: 20.8% (30.8fps), 35.2% (65.8fps), and 12.6% (18.0fps)
Navi 24, 4GB [email protected], 144GB/s, 107W, 2815MHz, 1024 shaders (opens in new tab)
Radeon R9 390 (20.2%, 29.8 frames per second), and 27.4% (51.2fps)
Grenada, 8GB [email protected], 384GB/s, 1000MHz, 2560 shaders, and 275W (opens in new tab)
GeForce GTX 980, 19.5% (28.9 fps), 28.7% (opens in new tab) (53.7fps)
2048 shaders, 1216 MHz, 4 GB [email protected] Gbps, 256 GB/s, 165 W, GM204 (opens in new tab)
GeForce GTX 1650 GDDR6: 19.5% (28.8 fps), 30.3% (56.7fps)
TU117, 4GB [email protected], 192GB/s, 896 shaders, 1590MHz, 75W (opens in new tab)
Intel Arc A380: 19.2% (28.3 fps), 29.3% (54.7 fps), and 13.6% (19.5fps)
1024 shaders, ACM-G11, 2450 MHz, 6 GB [email protected], 186 GB/s, 75W (opens in new tab)
Radeon RX 570 4GB: 19.1% (28.3 fps), 28.7% (53.6 fps), and 14.0% (20.0fps)
Polaris 20, 1244MHz, 4GB [email protected], 224GB/s, 150W, 2048 shaders (opens in new tab)
GeForce GTX 1060 3GB 18.8% (27.8 fps) 28.1% (opens in new tab) (52.6fps)
3GB [email protected], 192GB/s, 120W, GP106, 1152 shaders, 1708MHz (opens in new tab)
GeForce GTX 1650: 18.2% (26.9 frames per second) 27.4% (51.1fps)
TU117, 4GB [email protected], 128GB/s, 896 shaders, 1665MHz, 75W (opens in new tab)
GeForce GTX 970 (26.5 frames per second; opens in new tab) 17.9% (49.1fps)
GM204, 4GB [email protected], 256GB/s, 1664 shaders, 1178MHz, and 145W (opens in new tab)
Radeon RX 6400 (23.7 frames per second; opens in new tab) 16.0% (52.0fps)
Navi 24, 768 shaders, 2321 MHz, 4 GB [email protected] GB/s, and 53 watts (opens in new tab)
GeForce GTX 780: 14.9% (22.0 fps), 20.6% (opens in new tab) (38.5fps)
3GB [email protected], 288GB/s, 900MHz, GK110, 2304 shaders, 230W (opens in new tab)
GeForce GTX 1050 Ti 13.4% (19.8 fps) 20.3% (opens in new tab) (38.0fps)
768 shaders, 1392MHz, 4GB [email protected], 112GB/s, and 75W for the GP107 (opens in new tab)
GeForce GTX 1630 (11.4 percent, 16.9 frames per second), and 18.1% (33.9fps)
4GB [email protected], 96GB/s, 1785MHz, 512 shaders, TU117, 75W (opens in new tab)
GeForce GTX 1050 (10.0%) (14.8 frames per second) 15.9% (29.8fps)
GP107, 640 shaders, 1455 MHz, 2 GB GDDR5 @ 7 Gbps, 112 GB/s, and 75 W (opens in new tab)
Radeon RX 560 4GB (14.8 frames per second) 10.0% (31.8fps)
Baffin, 4GB [email protected], 112GB/s, 1024 shaders, 1275MHz, and 60-80W (opens in new tab)
4GB Radeon RX 550 (opens in new tab)
10.5% (19.6fps) (19.6fps)
Lexa, 640 Shaders, 4GB [email protected], 112GB/s, 50W, 1183MHz (opens in new tab)
GPU GT 1030 (opens in new tab)
7.8% (14.5fps) (14.5fps)
384 shaders, GP108, 1468 MHz, 2 GB [email protected] Gbps, 48 GB/s, and 30 W (opens in new tab)
*: Because the GPU was unable to complete all tests, the overall result at 1080p Ultra is a little off.
Although the RTX 4090 technically wins at 1080p ultra, it’s the 1440p and, in particular, the 4K statistics that stand out. At 1080p extreme, it is just 4% quicker than the next closest RX 7900 XTX, but at 1440p and 4K, the difference rises to 9% and 22%, respectively. It is also a significant improvement over the RTX 3090 Ti, being 10% faster at 1080p, 23% faster at 1440p, and 51% faster at 4K.
The above fps values combine both the average and minimum fps into a single score, with the average receiving more weight than the 99th percentile fps (just in case you check our evaluations and see a discrepancy in ratings).
As we plan to apply the same test suite with the same settings on all current and older generation graphics cards, please in mind that we are not include any ray tracing or DLSS results in the above table. The only cards that support DLSS are RTX cards (and RTX 40-series if you want DLSS 3), which severely restricts the GPUs we can directly compare.
Even if the RTX 4090 is more expensive than the RTX 3090 from the previous generation, this is to be expected. The 3090 had only a little performance upgrade over the 3080 at the time of debut, so we’d say it’s much better now. With the 4090, Nvidia appears to have gone above and beyond the competition by raising the core counts, clock rates, and power caps.
While AMD’s previous generation RX 6950 XT officially takes the lead at 1080p medium, where CPU constraints become the limiting factor, the RTX 4080 and RX 7900 XTX, which are a step down from the RTX 4090, exchange blows at higher resolutions. We’ve upgraded our test PC since we need to retest all of the most recent game patches and driver updates (you can see early results in our most recent GeForce RTX 4070 Ti and Radeon RX 7900 XTX and 7900 XT(opens in new tab) evaluations). However, it will take time.
In our rasterization evaluations, the RTX 4070 Ti ultimately behind the RX 7900 XT by up to 8% at 4K on average. If you don’t care about ray tracing, you can easily argue that the 7900 XT is a better deal because it will win in that area.
When it comes to older GPU generations, the RX 5000-series, GTX 16-series, and RTX 20-series chips are dispersed among the results. The RTX 2080 Super falls just short of the RTX 3060 Ti, while the RX 5700 XT trails the RX 6600 XT by a few percentage points, according to the general rule of thumb that applies to the newer architectures.
If you look back far enough, you can see how current games severely penalise GPUs with less than 4GB VRAM when played at their highest settings. Since a few years ago, we have argued that 4GB is barely adequate and 6GB or more is preferable. Even if they perform better at 1080p medium, the GTX 1060 3GB, GTX 1050, and GTX 780 really failed to execute several of our tests, which slightly skews their findings.
Let’s now shift our attention to the ray tracing hierarchy.
VIDEOS WE RECOMMEND FOR YOU…
Selection of a CPU
Your computer’s central processing unit, or CPU, is its mind. It has a significant impact on how well something performs overall. Think about these 7 factors while selecting a CPU for your subsequent PC:
Ranking 2023 for Ray Tracing GPU Benchmarks
Framerates can drastically decrease when ray tracing is enabled, especially in demanding games like those we employ in our DXR test suite. Testing is being done using “extreme” and “medium” ray tracing settings. Ray tracing effects are enabled when using medium graphics settings (set to “medium” if available; otherwise, “on”), while putting on extreme activates nearly all of the RT possibilities.
We are grouping these outcomes according to the 1080p medium scores because ray tracing is so much more difficult. This is also true since testing at resolutions higher than 1080p medium would be useless because the RX 6500 XT, RX 6400, and Arc A380 can’t really perform ray tracing even at these settings. There are now no more GPUs that can support ray tracing, but there will be more in the near future.
Bright Memory Infinite, Control Ultimate Edition, Cyberpunk 2077, Metro Exodus Enhanced, and Minecraft are the five ray tracing games we’re utilising; they’re all DX12/DX12 Ultimate titles. The fps score is the geometric mean (equal weighting) of the five games, and the percentage is scaled relative to the fastest GPU in the list, which is once again the GeForce RTX 4090. (Note that we had to remove Fortnite from our most recent reviews because the new version broke our benchmarks and changed the available settings. Thanks, Epic!)
Hierarchy of Tom’s Hardware Ray Tracing GPU Benchmarks
Specifications for a graphics card in 1080p Medium, 1080p Ultra, 1440p Ultra, and 4K Ultra (Links to Review)
GeForce RTX 4090, 2520MHz, 24GB [email protected], 1008GB/s, 450W, 100% (161.9fps), 100% (132.8fps), 100% (97.6fps), 100% (53.6fps) (opens in new tab)
GeForce RTX 4080 (opens in new tab): AD103, 9728 shaders, 2505MHz, 16GB [email protected], 717GB/s, 320W, 85.8% (138.9fps), 80.8% (107.3fps), 74.3% (72.6fps), and 69.3% (37.2fps) (opens in new tab)
GeForce RTX 3090 Ti, 1860MHz, 24GB [email protected], 1008GB/s, 450W, 71.5% (115.7fps), 65.6% (87.1fps), 61.2% (59.8fps), and 57.8% (31.0fps) (opens in new tab)
GeForce RTX 4070 Ti, 12GB [email protected], 504GB/s, 285W, AD104, 7680 shaders, 58.9% (78.3fps), 54.9% (53.6fps), and 50.8% (27.2fps) (opens in new tab)
GeForce RTX 3090 (opens in new tab): 65.1% (105.3 fps), 57.6% (76.6 fps), 53.4% (52.1 fps), and 49.3% (26.4 fps); GA102, 10496 shaders, 1695 MHz, 24 GB [email protected] GB/s, 350W (opens in new tab)
GeForce RTX 3080 Ti (opens in new tab) 64.5% (104.3fps), 56.5% (75.0fps), 51.9% (50.7fps), and 47.3% (25.3fps) GA102, 10240 shaders, 1665MHz, 12GB [email protected], and 350W (opens in new tab)
GeForce RTX 3080 12GB (opens in new tab) 62.5% (101.1fps), 55.5% (73.7fps), 51.7% (50.5fps), and 48.3% (25.9fps) at 1845MHz with 12GB of GDDR6X memory operating at 19Gbps and 400W. (opens in new tab)
Radeon RX 7900 XTX (opens in new tab) Navi 31, 12288 shaders, 2500MHz, 24GB [email protected], 960GB/s, 355W, 59.4% (96.1fps), 51.7% (68.7fps), 47.4% (46.3fps), and 41.9% (22.4fps) (opens in new tab)
GeForce RTX 3080 (opens in new tab): GA102, 8704 shaders, 1710MHz, 10GB [email protected], 760GB/s, 320W, 56.3% (91.1fps), 48.4% (64.2fps), 44.1% (43.0fps), and 41.3% (22.1fps) (opens in new tab)
Radeon RX 7900 XT (opens in new tab) Navi 31, 10752 shaders, 2400MHz, 20GB [email protected], 800GB/s, 315W, 51.5% (83.4fps), 44.1% (58.6fps), 39.1% (38.2fps), and 34.2% (18.4fps) frame rates (opens in new tab)
Radeon RX 6950 XT: 48.9% (79.1 fps), 42.0% (55.8 fps), and 37.0% (36.1fps)
Navi 21, 2310MHz, 16GB [email protected], 576GB/s, 335W, 5120 shaders (opens in new tab)
GeForce RTX 3070 Ti (opens in new tab): GA104, 6144 shaders, 1770MHz, 8GB [email protected], 608GB/s, 290W, 46.0% (74.5fps), 39.0% (51.8fps), 34.1% (33.3fps), and 29.8% (16.0fps) (opens in new tab)
Radeon RX 6900 XT (opens in new tab): Navi 21, 5120 shaders, 2250MHz, 16GB [email protected], 512GB/s, 300W, 45.9% (74.3fps), 40.1% (53.3fps), 35.9% (35.0fps), and 32.4% (17.4fps) (opens in new tab)
Titan RTX (opens in new tab): 39.3% (52.2fps), 34.3%, and 45.8% (74.2fps) (33.5fps)
4608 shaders, TU102, 1770 MHz, 24 GB [email protected] Gbps, 672 GB/s, and 280 W (opens in new tab)
GeForce RTX 3070: 43.8% (70.9fps), 38.2% (50.7fps), and 33.6% (32.9fps)
8GB [email protected], 448GB/s, 5888 shaders, 1725MHz, GA104, and 220W (opens in new tab)
GeForce RTX 2080 Ti (opens in new tab) TU102, 4352 shaders, 1545MHz, 11GB [email protected], 616GB/s, 250W, 43.1% (69.7fps), 36.1% (48.0fps), 31.8% (31.1fps), and 27.9% (14.9fps) (opens in new tab)
Radeon RX 6800 XT: 40.8% (66.0 fps), 34.5% (45.9 fps), and 30.1% (opens in new tab) (29.4fps)
Navi 21, 2250MHz, 16GB [email protected], 512GB/s, 300W, 4608 shaders (opens in new tab)
GeForce RTX 3060 Ti: 38.8% (62.7 fps), 28.7% (38.1 fps), and 26.0% (25.4fps)
8GB [email protected], 448GB/s, 1665MHz, 4864 shaders, GA104, 200W (opens in new tab)
Radeon RX 6800: 36.7% (59.4 fps), 30.8% (40.8 fps), and 27.0% (opens in new tab) (26.3fps)
Navi 21, 3840 shaders, 2105 MHz, 16 GB of GDDR6 at 16 Gbps, 512 GB/s, and 250 W (opens in new tab)
GeForce RTX 2080 Super: 36.7% (59.4 fps), 31.6% (42.0 fps), and 27.8% (27.1fps)
3072 shaders, 1815 MHz, 8 GB, 15.5 GB/s, 250 watts, GDDR6 (opens in new tab)
GeForce RTX 2080: 35.2% (57.0 fps), 29.9% (39.7 fps), and 26.2% (opens in new tab) (25.5fps)
TU104, 1710MHz, 8GB [email protected], 448GB/s, 215W, 2944 shaders (opens in new tab)
GeForce RTX 2070 Super: 32.3% (52.4 fps), 27.5% (36.5 fps), and 23.7% (23.1fps)
8GB [email protected], 448GB/s, TU104, 2560 shaders, 1770MHz, and 215W (opens in new tab)
Intel Arc A770 16GB 30.8% (49.8 fps), 26.1% (34.6 fps), and 22.2% (opens in new tab) (21.7fps)
4096 shaders, ACM-G10, 2100 MHz, 16 GB [email protected] Gbps, 560 GB/s, 225 W (opens in new tab)
Radeon RX 6750 XT: 30.6% (49.5 fps), 25.7% (34.2 fps), and 22.0% (21.5fps)
Navi 22, 12GB [email protected], 432GB/s, 2600MHz, 2560 shaders, and 250W (opens in new tab)
GeForce RTX 3060: 30.1% (48.8 fps), 22.4% (29.8 fps), and 19.2% (18.7fps)
[email protected], 12GB, 360GB/s, 170W, GA106, 3584 shaders (opens in new tab)
Radeon RX 6700 XT: 28.8% (46.5 fps), 24.4% (32.4 fps), and 20.6% (20.1fps)
Navi 22, 12GB [email protected], 384GB/s, 2560 shaders, and 230W (opens in new tab)
GeForce RTX 2070: 28.5% (46.2 fps), 24.1% (32.0 fps), and 20.9% (20.4fps)
8GB [email protected], 1620MHz, TU106, 2304 shaders, 448GB/s, and 175W (opens in new tab)
GeForce RTX 2060 Super: 27.4% (44.3 fps), 22.9% (30.4 fps), and 19.7% (19.2fps)
TU106, 1650MHz, 8GB [email protected], 448GB/s, 175W, 2176 shaders (opens in new tab)
Radeon RX 6700 10GB: 26.5% (42.8fps), 22.0% (29.3fps), and 17.9% (opens in new tab) (17.5fps)
Navi 22, 10GB [email protected], 320GB/s, 2450MHz, 2304 shaders, and 175W
Intel Arc A750 23.8% (38.4 fps) 19.1% (opens in new tab) (25.3fps)
ACM-G10, 2050MHz, 8GB [email protected], 512GB/s, 225W, 3584 shaders (opens in new tab)
GeForce RTX 2060 23.2% (37.6 frames per second) 19.3% (25.6fps)
1920 shaders, 1680MHz, 6GB [email protected], 336GB/s, and 160W for the TU106 (opens in new tab)
Radeon RX 6650 XT (22.6 percent, 36.6 frames per second), 18.5% (24.6fps)
Navi 23, 8GB [email protected], 280GB/s, 2635MHz, 2048 shaders, and 180W (opens in new tab)
Radeon RX 6600 XT (21.7 percent, 35.2 frames per second), 18.2% (24.1fps)
Navi 23, 8GB [email protected], 256GB/s, 2589MHz, 2048 shaders, and 160W (opens in new tab)
GeForce RTX 3050: 15.4% (15.7fps) 19.0% (30.7fps) (20.5fps)
224GB/s, 8GB [email protected], 1777MHz, GA106, 2560 shaders, 130W (opens in new tab)
10.2% for Radeon RX 6600 (opens in new tab) (16.5fps)
Navi 23, 2491MHz, 8GB [email protected], 224GB/s, 132W, 1792 shaders (opens in new tab)
6.1% for Intel Arc A380 (opens in new tab) (9.9fps)
1024 shaders, ACM-G11, 2450 MHz, 6 GB [email protected], 186 GB/s, 75W (opens in new tab)
6.1% for Radeon RX 6500 XT (opens in new tab) (9.9fps)
Navi 24, 4GB [email protected], 144GB/s, 107W, 2815MHz, 1024 shaders (opens in new tab)
Radeon RX 6400 5.1% (opens in new window) (8.3fps)
Navi 24, 768 shaders, 2321 MHz, 4 GB [email protected] GB/s, and 53 watts (opens in new tab)
Just look at the results with ray tracing if you thought the RTX 4090’s performance was remarkable at 4K in our usual test suite. Although there are still further potential performance improvements for ray tracing with SER, OMM, and DMM, not to mention DLSS3 (which honestly just seems to cause occasional artifacts and doesn’t really feel much smoother), those start to show up here. We knew Nvidia had added even more ray-tracing enhancements to the Ada Lovelace architecture.
The RTX 4090 dominates all rivals and outperforms the RTX 3090 Ti by 40% even at 1080p medium, a rather low level for DirectX Raytracing. The advantage increases to 56% at 1080p extreme, and to almost 70% at 1440p. Even without DLSS 3, the RTX 4090 is 80% quicker than the 3090 Ti, contrary to Nvidia’s claims made before the RTX 4090’s introduction that it was “2x to 4x faster than the RTX 3090 Ti.”
AMD continues to put DXR and ray tracing on the back burner in favor of rasterization performance enhancement and chipset-based manufacturing cost reduction on the upcoming RDNA 3 GPUs. As a result, AMD’s ray tracing performance is still not all that impressive. The new RX 7900 XTX is roughly equivalent to Nvidia’s RTX 3080 12GB from the previous generation, putting it slightly behind the RTX 3080 Ti and RTX 3090. The new 4070 Ti outperforms it by 10% at 4K, 13% at 1440p, and 19% at 1080p. While the stepping-down RX 7900 XT sits just above the RTX 3080 10GB card and above the RX 6950 XT.
In our evaluation, you can also see what the DLSS Quality setting accomplished for the RTX 4090’s DXR gaming performance. In a nutshell, it improved performance by 78% at 4K ultra. In our preview testing, DLSS 3 increased framerates by a further 30% to 100%, although we are still dubious about the use of Frame Generation outside of faking higher frame rates. In our ray tracing test suite, the 4090 is therefore roughly four times faster with DLSS2 than AMD’s RX 7900 XTX. Ouch.
If AMD can achieve mass adoption, its FSR 2.0 technology might be advantageous in this situation. Currently, Cyberpunk 2077 is the only game in our DXR suite that supports FSR2. Without FSR, the fastest AMD GPUs can only achieve 60 frames per second at 1080p extreme, with an average of 40 to 50 frames per second at 1440p. However, only the 3090 Ti, 4080, and 4090 break the 30 fps barrier on the composite score, and some games perform poorly on the 4080 and 3090 Ti. Native 4K DXR is still out of reach for almost every GPU.
The bottom tier of DXR-capable GPUs barely manage 1080p medium, and the RX 6500 XT can’t even do that, with single-digit framerates in most of our test suites and one game that wouldn’t even work at our chosen “medium” settings. Midrange GPUs like the RTX 3070 and RX 6700 XT essentially manage 1080p ultra and not much more. To enable ray tracing, Control needs at least 6GB VRAM.
Interestingly given that AMD has 16 Ray Accelerators compared to Intel’s 8 RTUs, the Arc A380 narrowly edges over the RX 6500 XT in terms of ray tracing performance. Intel published a detailed analysis of its ray tracing hardware, and Arc seems very impressive—except that the A380’s RTU count drastically restricts performance. In DXR testing, the top-end A770 still only has 32 RTUs, which is just enough for it to surpass the RTX 3060 (barely), but it can’t go much farther than that. Arc A770 also outperforms AMD’s RX 6800 in DXR performance, highlighting how inadequate AMD’s RDNA 2 technology is for ray tracing.
Examining the generational performance of Nvidia’s RTX GPUs is also fascinating. The RTX 2060, the slowest 20-series GPU, still performs somewhat better than the new RTX 3050, while the RTX 2080 Ti, the fastest, is behind the RTX 3070 by a little margin. The performance of the 3090 is about quadruple that of the 3050, whereas the 2080 Ti is effectively twice that of 2060. Hopefully, a future RTX 4050 will offer advances comparable to those of the 4090 at a far lower cost.
A Graphics Card to Buy
What kind of graphics card do you require? We developed this hierarchy of GPU benchmarks, which includes several GPUs from the last four hardware generations, to aid in your decision-making. Unsurprisingly, the fastest cards either feature AMD’s Big Navi or Nvidia’s Ampere architecture. Without ray tracing, AMD’s most recent graphics cards perform admirably, but once RT is activated, they tend to lag behind—even more so if you use DLSS, which you should. But now that GPU costs have finally stabilized, it’s a great time to update.
Of course, it involves more than just playing games. Numerous programs utilize the GPU for additional tasks, and we discussed some expert GPU benchmarks in our RTX 3090 Ti review. But a competent gaming graphics card will generally perform just as well in demanding GPU computing jobs. Purchase one of the best cards, and you’ll be able to run games at high resolutions and frame rates with all of the effects turned on, as well as do content production tasks just as effectively. As you move down the list, you’ll need to start lowering the settings in order to get respectable performance in both standard gameplay and GPU benchmarks.
Of course, it’s not just about expensive GPUs. We examined Intel’s Xe Graphics DG1, which is essentially an integrated graphics alternative. We didn’t even try running any of those at settings higher than 1080p medium because the results weren’t particularly pleasing. However, it is still clear where such GPUs fall on the list of 2020–2021 GPU benchmarks. Fortunately, Intel’s Arc Alchemist, aka DG2, seems to be made of a totally different material. Mostly, that is.
You cannot ignore the CPU if gaming is your primary objective. Investing in the greatest gaming GPU won’t do you much good if your CPU is outdated or inadequate. Make sure you have the appropriate CPU for the level of gameplay you want to reach by checking out the Best CPUs for Gaming page and our CPU Benchmarks Hierarchy.
(Photo courtesy of Tom’s Hardware)
How We Test for GPU Benchmarks and Our Test System
For our testing, we used two separate PCs. While our prior testbed used Coffee Lake and Z390, the most recent configuration for 2022 and after uses an Alder Lake CPU and platform. These are the two PCs’ specifications.
Tom’s Hardware GPU Testbed for 2023
Core i9-12900K by Intel (opens in new tab)
Pro Z690-A WiFi DDR4 from MSI (opens in new tab)
2x16GB of Corsair’s DDR4-3600 CL16 (opens in new tab)
2TB Crucial P5 Plus (opens in new tab)
Gold Cooler Master MWE 1250 (opens in new tab)
PL360 Cooler Master Flux (opens in new tab)
CM-HAF500 Cooler Master
64-bit Windows 11 Pro
Tom’s Hardware GPU Testbed for 2020–2021
Core i9-9900K by Intel (opens in new tab)
RGB Corsair H150i (opens in new tab)
MEG Z390 Ace by MSI (opens in new tab)
2x16GB DDR4-3200 from Corsair (opens in new tab)
2TB XPG SX8200 Pro (opens in new tab)
(Opens in new tab) Windows 10 Professional (21H1)
The testing process is the same for each graphics card. After starting the game, we perform one pass of each benchmark to “warm up” the GPU before running at least two passes at each setting and resolution combination. We use the fastest of the two runs if there is just a 0.5% difference or less between them. We repeat the exam at least twice more to assess what “normal” performance should be if there is a significant variation from the norm.
RTX 3070 Ti, RTX 3070, and RTX 3060 Ti, for instance, all typically going to perform within a limited range — 3070 Ti is around 5% faster than 3070, which is approximately 5% faster than 3060 Ti. We also look at all the data and check for abnormalities. We will retest any cards that are exhibiting the anomaly in order to determine the “proper” outcome if we come across games with obvious outliers (performance is more than 10% higher for the cards just indicated).
Updated drivers and game updates that might affect performance will unavoidably be released due to the time needed to evaluate each GPU. We periodically retest a few sample cards to make sure our findings are still accurate, and if not, we retest the GPU and the game(s) that were affected (s). If a new game that is well-liked and suitable for testing is released in the upcoming year, we could add it to our test library as well. To learn what constitutes a good game, check out our selection criteria.
Game charts for each GPU benchmark
The performance is summarised in the tables above, but we also have individual game charts for both the conventional and ray tracing test suites available for those who are interested. Testing was done using the most recent AMD and Nvidia drivers in most cases, although some of the cards were tested using somewhat older versions as of December 13, 2022.
There are 29 cards in the 1080p charts, so you could argue that we’re already well over that threshold. Take note that we’re just including the most recent and earlier generations of hardware in these charts since otherwise, things get too crowded. (Tip: If you’re using a PC, click the enlarge icon.)
| gpu benchmark |
| gpu comparison |
