GeekRadar / The comparison desk

COMPARE.

GPU

Nvidia

Nvidia GeForce RTX 5090 (32GB GDDR7)

Graphics processor

Video memory: 32 GB^*
Memory bus: 512 bit^*
Rated memory bandwidth: 1,792 GB/s^*

Benchmark results

Reported score · version unreported · passmark_g3d_mark51,500^*Platform unreported Compare results

Processing & platform

Architecture: Blackwell Consumer^*
Process node: 4 nm^*
GPU boost clock: 2,400 MHz^*

Memory & storage

Video memory: 32 GB^*
Video memory type: GDDR7^*
Memory bus: 512 bit^*
Rated memory bandwidth: 1,792 GB/s^*

Power specifications

Thermal design power: 575 W^*

System configuration

Detailed specifications

Specifications 1 · 19 fields

GPU architecture — GeForce RTX 5090 — 32 GB GDDR7

Blackwell^*

Theoretical peak: BF16 tensor dense per GPU — GeForce RTX 5090 — 32 GB GDDR7

209.5^* TFLOPS

Scope and conditions

Accumulation: FP32^*

Theoretical peak: FP16 tensor dense per GPU — GeForce RTX 5090 — 32 GB GDDR7

209.5^* TFLOPS

Scope and conditions

Accumulation: FP32^*

Theoretical peak: FP16 tensor with FP16 accumulation dense per GPU — GeForce RTX 5090 — 32 GB GDDR7

419^* TFLOPS

Theoretical peak: FP32 non-tensor dense per GPU — GeForce RTX 5090 — 32 GB GDDR7

104.8^* TFLOPS

Theoretical peak: FP4 tensor dense per GPU — GeForce RTX 5090 — 32 GB GDDR7

1,676^* TFLOPS

Scope and conditions

Accumulation: FP32^*

Theoretical peak: FP8 tensor dense per GPU — GeForce RTX 5090 — 32 GB GDDR7

419^* TFLOPS

Scope and conditions

Accumulation: FP32^*

Theoretical peak: FP8 tensor with FP16 accumulation dense per GPU — GeForce RTX 5090 — 32 GB GDDR7

838^* TFLOPS

Theoretical peak: INT8 tensor dense per GPU — GeForce RTX 5090 — 32 GB GDDR7

838^* TOPS

Theoretical peak: TF32 tensor dense per GPU — GeForce RTX 5090 — 32 GB GDDR7

104.8^* TFLOPS

Physical configuration — GeForce RTX 5090 — 32 GB GDDR7

Desktop graphics card (PCIe add-in card)^*

Native hardware support for GPU interconnect — GeForce RTX 5090 — 32 GB GDDR7

No^*

PCIe interface declaration — GeForce RTX 5090 — 32 GB GDDR7

PCIe Gen 5^*

Maximum board power — GeForce RTX 5090 — 32 GB GDDR7

575^* W

Peak memory bandwidth per GPU — GeForce RTX 5090 — 32 GB GDDR7

1,792^* GB/s

Memory interface width per GPU — GeForce RTX 5090 — 32 GB GDDR7

512^* bit

Memory capacity per GPU — GeForce RTX 5090 — 32 GB GDDR7

32^* GB

Memory technology — GeForce RTX 5090 — 32 GB GDDR7

GDDR7^*

GPU model — GeForce RTX 5090 — 32 GB GDDR7

GeForce RTX 5090^*

Specifications 2 · 29 fields

gpu name: Nvidia GeForce RTX 5090 (32GB GDDR7)^*
sm count: 170^*
tensor cores count: 680^*
tensor core gen: 5th Gen^*
rt cores count: 170^*
base clock mhz: 2,010^* MHz
boost clock mhz: 2,400^* MHz
memory size gb: 32^* GB
memory type: GDDR7^*
bus width bit: 512^*
bandwidth gbs: 1,792^* GB/s
series: GeForce RTX 50^*
fp32 tflops: 104.4^* TFLOPS
tensor tflops: 3,350^* TFLOPS
tdp watts: 575^* W
passmark g3d mark: 51,500^*
time spy graphics: 48,500^*
launch price usd: 1,999^*
transistor density m per mm2: 122.67^* mm²
fp32 gflops per dollar: 52.23^* GFLOPS
fp32 gflops per watt: 181.57^* GFLOPS
tensor tflops per watt: 5.826^* TFLOPS
architecture: Blackwell Consumer^*
chip code: GB202^*
launch year: 2025^*
process node nm: 4^* nm
die size mm2: 750^* mm²
transistor count millions: 92,000^*
cuda cores: 21,760^*

* Provisional data