The EVGA RTX 2080 Black Edition (08G-P4-2081-KR) is an elite gaming class display card. MSRP $829. NVIDIA raised prices for Turing vs Pascal and earlier.
- nVidia Turing
- TSMC 12nm
- 2944 CUDA cores
- 10.07 TFLOPS SP
- 182 TMUs
- 64 ROPs
- 46 SM
- L1 64KB per SM
- L2 4MB
- 364 tensor cores
- 46 RT cores
- 13.6 billion transistors
- 1515 MHz primary clock
- 1710 MHz maximum boost
- 256-bit memory
- 8192 MB GDDR6
- 256-bit memory
- 448 GB/s
- Dual HDB fans with metal backplate
- DisplayPort 1.4, HDMI 2.0b
- VirtualLink VR support
- 14 Power Phases
- 8-pin and 6-pin PCIe required
- 215W TDP, 250W boost power TDP
- EVGA ICX cooling
- Length: 10.62 in – 269.83 mm
- 3 year warranty
- UPC 843368059279
Best bet for running the RTX 2080 is with a 750W PSU or better. Extreme cards can use a lot of power when pushed hard. The EVGA ICX dual fan cooler handles the thermal load of the card easily. Having some slack on the PSU capacity allows it to be substantially more efficient.
The 64-bit float performance is horrible at 1:32 providing 314 GFLOPS. A Ryzen CPU can beat that suggesting NVIDIA intentionally crippled the ALU. The desire for 64-bit float is to provide for upcoming HDR displays with 10-16 bit color depth and digital photography.
The RTX line of cards are very popular as the RTX 2080 features hardware ray-tracing and has some additional logic for AI as well. The card is powerful enough to even outdo the old GTX 1080 Ti.
NVIDIA says the RTX 2080 is designed for overclocking, and the RTX 2070 will offer more performance (for ray-tracing) than the NVIDIA Titan XP card.
This card has RGB lighting which can add some flare to a rather droll setup. It is fully adjustable with the EVGA X1 software.
NVIDIA is continuing to sell Pascal cards beside the new Turing cards. Considering the cost of Turing, the Pascal cards are able to handle the lower price points without competing with the RTX cards.
The perfect TU104 die has 48 SM but the RTX 2080 has 46 SM so it has two SM disabled due to fabrication issues. Realistically so few cards with a perfect 48 SM would be found with the larger die sizes of the Turing architecture.
The more expensive RTX Titan has the TU102 with 72 SM is why it’s expensive. The RTX 2080 Ti is the binned card which is why it is also very expensive. The MSRP for the RTX Titan is a measly $2,499. Gaming enthusiasts are served well using barely affordable the RTX 2080.
The EVGA RTX 2080 Black Edition does run warm when loaded so its advisable to have proper ventilation in your machine. Loaded the card can reach 80-82°C but it cools quite quickly afterwards. The EVGA cooler is very effective.
While the Black Edition is setup for overclocking, its advisable to use the card at default to ensure a longer service life. The Turbo Boost can increase performance based on thermal conditions in real time eliminating worry.
The RTX performance will vary considerably from game to game. Generally the framerate will suffer considerably even with the RTX 2080. Battlefield V players noted that above 1920×1080 that rendering was very sluggish, but after some updates the gameplay was dramatically better.
Games that support the Turing RTX can move some shaders to the AI and ray tracing logic where it can be separated from the main SM work. The result is better rendering done faster.
DLSS (deep learning super sampling) allows hardcore gamers to use higher resolutions and settings while still maintaining solid framerates. The DLSS team first extracts many aliased frames from the target game, and then for each one NVIDIA generates a matching “perfect frame” using either super-sampling or accumulation rendering.
NVIDIA built DLSS to leverage the Turing architecture’s Tensor Cores and to provide the largest benefit when GPU load is high. To this end, they concentrated on high resolutions during development (where GPU load is highest) with 4K (3840×2160) being the most common training target.
NVIDIA Turning provides NVENC which provides better video encoding with support for HEVC (H.265) and VP9. Legacy standards are generally supported. Testing the RTX 2080 with some 4K video shows it can encode MPEG-4 video faster than 2x real time.
VirtualLink uses the USB-C port which has been developed as an open standard by an industry consortium of leading silicon, software, and headset manufacturers led by NVIDIA, Oculus, Valve, Microsoft, AMD, and HTC. This open industry standard for VR connectivity advances VR interoperability and provides economic benefits to head-mounted display (HMD) makers, original equipment manufacturers (OEMs), developers, and consumers. It will accelerate the development and implementation of new VR technologies.
Navi and Turing are classed as Single Instruction Multiple Data (SIMD) processors, although a better description would be Single Instruction Multiple Threads (SIMT).
Comparing RDNA vs Turing is much more interesting. The architecture is different looking but logically they are close to one another.
Maxwell and Pascal have 128 CUDA per SM while Turning has 64. The Turing SM is very similar to the GCN CU in AMD video cards. This allows the RTX 2080 to perform well with titles that are compiled for Radeon hardware. Turing has a group of 16 CUDA cores (same as a SIMD group from GCN of 16 ALUs), 4 of which make up a SM with its own warp scheduler and shared registers, identical to a Compute Unit in GCN.
By designing logic close to one another allows game developers some breathing space when using third party libraries. Generally both AMD and NVIDIA will offer PC gamers cards that are years beyond the wildest dream of a console.
The EVGA RTX 2080 Black Edition needs all of the rated 215W and then some. The HX1000i is plenty adequate for even RTX 2080 SLI. Its very advisable to use a high-end platinum PSU with the RTX 2080 for maximum service life.
Recall the RX 480 when it launched was using a single 6-pin cable which was not enough for the card’s needs. The Sapphire Nitro+ RX 480 8GB comes with an 8-pin cable which mitigates the direct power feed much better. AMD cards are more demanding of slot power which has driven the use of dual ATX12V cables.
NVIDIA’s reference regulators are very sophisticated which does add to the BOM. Still this solves the power problem as they are not adverse to asking for dual power cables.
The 8 phase LR22 regulators are 60A each. There are also several other controllers to handle the power more carefully to prevent the bus power from getting overloaded. Surrounding the regulators are Panasonic aluminum capacitors which can tolerate 105°C for up to 2000 hours.
The L47 regulators feed the GDDR6 memory. The GDDR6 typically needs about 20A of current.
All game benchmarks are at 3840×2160. Graphics are set for ultra. DLSS enabled where available.
- Middle Earth: Shadow of Mordor (78.53 average, min 45.86, max 106.37)
- MSI Kombustor
- Shadow of the Tomb Raider (59 fps)
- Thief (53.5 average, min 44.2, max 60.7)
- 3DMark Timespy