i
Zdroj: PCTuning.cz
Nvidia Turing: nové technologie pro chytřejší rendering | Kapitola 7
Seznam kapitol
Dnešním dnem končí embargo na detaily k nové architektuře Turing a podrobnosti o nových funkcích, které karty nabízejí. Asi jen s málokterou premiérou za celou dobu historie grafických 3D akcelerátorů představili výrobci karet tolik nových technologií najednou.
GeForce RTX 2080 vs GTX 1080 v parametrech
GeForce RTX 2080 je postavená na menším čipu TU104. Ten má aktivních jenom 23 TPC a tomu odpovídají další parametry.
GPU Features | GeForce GTX 1080 | GeForce RTX 2080 | Quadro P5000 | Quadro RTX 5000 |
Architecture | Pascal | Turing | Pascal | Turing |
GPCs | 4 | 6 | 4 | 6 |
TPCs | 20 | 23 | 20 | 24 |
SMs | 20 | 46 | 20 | 48 |
CUDA Cores / SM | 128 | 64 | 128 | 64 |
CUDA Cores / GPU | 2560 | 2944 | 2560 | 3072 |
Tensor Cores / SM | NA | 8 | NA | 8 |
Tensor Cores / GPU | NA | 368 | NA | 384 |
RT Cores | NA | 46 | NA | 48 |
GPU Base Clock MHz (Reference / Founders Edition) | 1607 / 1607 | 1515 / 1515 | 1607 | 1620 |
GPU Boost Clock MHz (Reference / Founders Edition) | 1733 / 1733 | 1710 / 1800 | 1733 | 1815 |
RTX-OPS (Tera-OPS) (Reference / Founders Edition) | 8,9 / 8,9 | 57 / 60 | NA | 62 |
Rays Cast (Giga Rays/sec) (Reference / Founders Edition) | 0,89 | 8 / 8 | NA | 8 |
Peak FP32 TFLOPSâ (Reference / Founders Edition) | 8,9 | 10 / 10,6 | 8,9 | 11,2 |
Peak INT32 TIPSâ (Reference/Founders Edition) | NA | 10 / 10,6 | NA | 11,2 |
Peak FP16 TFLOPSâ (Reference / Founders Edition) | NA | 20,1 / 21,2 | NA | 22,3 |
Peak FP16 Tensor TFLOPS with FP16 Accumulateâ (Reference/Founders Edition) | NA | 80,5 / 84,8 | NA | 89,2 |
Peak FP16 Tensor TFLOPS with FP32 Accumulateâ (Reference/Founders Edition) | NA | 40,3 / 42,4 | NA | 89,2 |
Peak INT8 Tensor TOPSâ (Reference / Founders Edition) | NA | 161,1 / 169,6 | NA | 178,4 |
Peak INT4 Tensor TOPSâ (Reference / Founders Edition) | NA | 322,2 / 339,1 | NA | 356,8 |
Frame Buffer Memory Size and Type | 8192 MB GDDR5X | 8192 MB GDDR6 | 16384 GDDR5X | 16384 GDDR6 |
Memory Interface | 256 b. | 256 b. | 256 b. | 256 b. |
Memory Clock (Data Rate) | 10 Gbps | 14 Gbps | 9 Gbps | 14 Gbps |
Memory Bandwidth (GB/sec) | 320 | 448 | 288 | 448 |
ROPs | 64 | 64 | 64 | 64 |
Texture Units | 160 | 184 | 160 | 192 |
Texel Fill-rate (Gigatexels/sec) | 277,3 / 277,3 | 314,6 / 331,2 | 277 | 348 |
L2 Cache Size | 2048 KB | 4096 KB | 2048 KB | 4096 KB |
Register File Size/SM | 256 KB | 256 KB | 256 KB | 256 KB |
Register File Size/GPU | 5120 KB | 11776 KB | 5120 KB | 12288 KB |
TDPé (Reference / Founders Edition) | 180 / 180 W | 215 / 225 W | 180 W | 230 W |
Transistor Count | 7,2 mld. | 13,6 mld. | 7,2 mld. | 13,6 mld. |
Die Size | 314 mm² | 545 mm² | 314 mm² | 545 mm² |
Manufacturing Process | 16 nm | 12 nm FFN | 16 nm | 12 nm FFN |
Note: Peak TFLOPS and TOPS rates are based on GPU Boost Clock. Power figure represents Graphics Card TDP only. The use of the VirtualLink™/USB Type-C™ connector requires up to an additional 35 W of power that is not represented in this power figure. |