Nvidia's RTX 50 series “Blackwell” graphics cards arrive with the RTX 5090. At $2000/£1940, it's an extremely expensive proposition, but it makes for some enticing rewards: the GB202 GPU delivers the best gaming performance ever, 32GB of high-speed GDDR7 memory for gaming or content creation, DLSS 4 features multi-frame generation to maximize the maximum refresh rate of nearly any monitor in supported games. Does it deliver?
To find out, we tested the RTX 5090 Founders Edition in a new suite of challenging games, looking at RT and raster performance at a variety of resolutions and some of the most demanding scenarios we've identified from our PC tech reviews. games, many of which feature custom benchmark runs unique to Digital Foundry.
Before we discuss the benchmark results, let's cover the physical design of the card itself, performance expectations based on its specs, and how we'll be testing it.
It's clear that this Nvidia Founders Edition model is the most compact flagship card we've seen from Team Green so far, with a two-slot design that's actually suitable for small form factor PCs despite being rated at 575W. The magic is achieved via a central PCB and two circulation fans that direct air through two impressively dense fin sets and out the other side. This is a stunning piece of industrial design.
Third-party graphics cards from Nvidia partners are expected to stick to more familiar territory with a more common three-fan, three-fan design, so Founders Edition models may be harder to find than usual. (We've rounded up a list of UK and US RTX 5090 retailers to bookmark if you need to.)
Either way you can expect strong power delivery, the FE card requires a 600W 12V-2×6 power cable which can be connected directly from a compatible ATX 3.0+ PSU or to four 8-pin using the provided adapter PCIe cable. It's angled on the FE card to prevent excessive strain on the cable, and the revised connector should prevent the burnout we experienced on one RTX 4090 FE rig. In addition to the usual HDMI 2.1 ports, there are three DisplayPort 2.1 ports, allowing compatible 4K 240Hz HDR displays to be connected without the need for DSC (Display Stream Compression).
| NVIDIA | RTX 5090 | RTX 5080 | RTX 5070 Titanium | RTX 5070 |
|---|---|---|---|---|
| graphics processor | GB 202 | GB 203 | GB 203 | GB 205 |
| core | 21760 | 10752 | 8960 | 6144 |
| boost clock | 2.41GHz | 2.62GHz | 2.45GHz | 2.51GHz |
| tensor core TOPS | 3352 | 1801 | 1406 | 988 |
| RT core TFLOPS | 318 | 171 | 133 | 94 |
| memory | 32GBGDDR7 | 16GBGDDR7 | 16GBGDDR7 | 12GBGDDR7 |
| memory bus width | 512 bits | 256 bit | 256 bit | 192 bits |
| memory bandwidth | 1792GB/sec | 960GB/sec | 896GB/sec | 672GB/sec |
| Total graphics processing power | 575W | 360W | 300W | 250W |
| Power supply recommendation | 1000W | 850W | 750W | 650W |
| power connector | 600W PCIe 5.0 (4 x 8-pin) | 450W PCIe 5.0 (3 x 8-pin) | 300W PCIe 5.0 (2 x 8-pin) | 300W PCIe 5.0 (2 x 8-pin) |
| price | $1999/£1939 | $999/£979 | $749/£729 | $549/£539 |
| release date | January 30 | January 30 | February | February |
Specs-wise, we're looking at 21,760 CUDA cores, a rated boost clock of 2.41GHz and 1792GB/s of memory bandwidth, along with 32GB of GDDR7 connected via an ultra-wide 512-bit bus. That represents a sizable gain compared to the RTX 4090 FE, which “only” has 16,384 CUDA cores and a 384-bit memory interface good for 1018GB/s of bandwidth, although the card does top out at 2.52GHz. Press the clock to run. This means a 33% increase in the number of GPU cores, a 77% increase in memory bandwidth, and a 28% increase in power consumption.
Most of Nvidia's marketing material about the 5090 vs. 4090 has focused on multi-frame generation vs. single-frame generation, so it will be interesting to see how the two flagship cards perform in true apples-to-apples tests. Given the specification differences, we expect results in the 25% to 35% range should be possible, but there are a number of complicating factors that could alter the results in some way.
For all our results, we use a high-end system based on the fastest gaming CPU (AMD Ryzen 7 9800X3D) to shift the burden as much as possible to the graphics card. We also have 32GB of Trident Z5 Neo DDR5-6000 CL30 memory, a high-end Asus ROG Crosshair X870E Hero motherboard, and a 1000W Corsair PSU.
With all that said, let’s get into benchmarking.
Nvidia GeForce RTX 5090 Analysis
- introduce [This Page]
- RT Benchmark: Alan Wake 2, Avatar: Pandora's Frontier, Cyberpunk 2077
- RT Benchmarks: Dying Light 2, F1 24, Hitman: World of Assassination
- RT Benchmarks: Indiana Jones and the Circle, Metro Exodus Enhanced Edition, A Plague Tale: Requiem
- Game benchmark: “Alan Killer 2”, “Black Myth: Wukong”, “Cyberpunk 2077”
- Game benchmarks: F1 24, Forza Horizon 5, Senua's Saga: Hellblade 2
- Game benchmark: “Hitman: World of Assassination”, “A Plague Tale: Requiem”
- Conclusions, Values and Recommendations
iGamesNews
