INSIGHTS

MBX White Paper: Nvidia GeForce RTX 3000 Series

January 14, 2021 | Post

MBX Systems Analysis and Transition Recommendations

Nvidia has released its next line of GeForce GPUs with the GeForce RTX™ 3000 series. Unlike the 2000 series, which provided incremental improvements over its predecessor, the 3000 series provides significant performance increases, particularly for ray-traced applications. The RTX 3000 series is intended to push the boundaries of real-time rendering, and, as a result, MBX Systems expects to see much heavier adoption of these cards than the RTX 2000 (Turing series). 

But with these significant performance improvements, we’re also seeing significant changes to the form factor and other aspects of the cards, which means engineers will need to plan accordingly if they intend to include the RTX 3000 series cards in their designs.

Specifications 

The following charts provide a detailed breakdown of all aspects of the Founders Edition RTX 3000 Series.  

Card FE RTX 3090 FE RTX 3080 FE RTX 3070FE RTX 3060 Ti
NVIDIA CUDA® Cores 10496 8704 58884864
Boost Clock (GHz) 1.7 1.71 1.731.67
Standard Memory Config 24 GB GDDR6X 10 GB GDDR6X 8 GB GDDR68GB GDDR6
Memory Interface Width 384-bit 320-bit 256-bit256-bit
Ray-Tracing Cores 2nd Generation 2nd Generation 2nd Generation2nd Generation
Tensor Cores 3rd Generation 3rd Generation 3rd Generation3rd Generation
NVIDIA Architecture Ampere Ampere AmpereAmpere
Microsoft DirectX® 12 Ultimate Yes Yes YesYes
PCI Express Gen 4 Yes Yes YesYes
NVIDIA Studio Drivers Yes Yes YesYes
NVIDIA NVLink™ (SLI-Ready) Yes No NoNo

The RTX 3000 Series is built on the new Ampere architecture, which more than doubles the number of cores across the board for a significant performance improvement. Additionally, these cards use a new generation of both Tensor and ray-trace RT cores. The Tensor cores are primarily used with Deep Learning Super Sampling (DLSS) to scale up lower-res images to a higher resolution for high-res monitors, while ray-trace RT cores are only used in ray-trace enabled applications. When not in use, these cores are dormant.

Two of the cards in the news series also feature expanded memory over their predecessors. The 3090 now has 24 GB of RAM, which is equivalent to that of a Quadro RTS 6000 and more than double that of the 2090. The 3080 now has 10 GB of RAM, a 25% increase over the 2080. These two cards also use the newer GDDR6X memory, which provides higher memory bandwidth. The amount of memory in the 3070 remains the same as in the 2070.

Nvidia NVLink, which was formerly known as Scalable Link Interface (SLI), has been removed from the 3080 and 3090, as it’s been declining in customer use cases for some time now.

Display support and card dimensions

Display Support FE RTX 3090 FE RTX 3080 FE RTX 3070FE RTX 3060 Ti
HDMI 2.1 connectors 11
DisplayPort 1.4a connectors 33
Maximum Digital Resolution (1) 7680×4320 7680×4320 7680×43207680×4320
Multi Monitor 44
HDCP2.32.32.32.3

All four cards support a maximum digital resolution of 7680×4320. In fact, Nvidia has shown demos of the 3090 handling games at 8K resolution with good performance with ray tracing enabled.  Additional third-party testing has proven that the RTX 3090 cannot handle true 8K without the assistance of DLSS. Nvidia has added support for High-bandwidth Digital Content Protection (HDCP) 2.3, High-Definition Multimedia Interface (HDMI) 2.1 and DisplayPort 1.4a. They’ve dropped the USB C connectors, which was previously known as the VR port.

Card Dimensions FE RTX 3090 FE RTX 3080 FE RTX 3070FE RTX 3060 Ti
Length 12.3″ (313 mm) 11.2″ (285 mm) 9.5″ (242 mm)9.5″ (242 mm)
Width 5.4″ (138 mm) 4.4″ (112 mm) 4.4″ (112 mm)4.4″ (112 mm)
Slot 3-Slot 2-Slot 2-Slot2-Slot

The 3000 Founders Edition series cards are a bit longer than their predecessors in the 2000 series, and the 3090, in particular, is significantly wider. The new longer length is necessary to accommodate the new cooling solution. Initial third-party card vendors (Asus, Gigabyte, PNY, and others) have designed some of their cards to be larger than the Founders Edition. Width has increased to 138mm (5.4 inches) on some cards, which, as noted above, will make that card incompatible with certain rackmount chassis configurations currently on the market. Slot size has also changed to a 2.7 slot design, which will limit its usage for rackmount chassis under 3U. Contact MBX for information on systems that will support the larger GPUs. 

Power and Thermal

Power and Thermal FE RTX 3090 FE RTX 3080 FE RTX 3070FE RTX 3060 Ti
Maximum GPU Temperature (in C) 93 93 9393
Graphics Card Power (W) 350 320 220200
Required System Power (W) (2) 750 750 650600
Supplementary Power Connectors 2x PCIe 8-pin 2x PCIe 8-pin 1x PCIe 8-pin1x PCIe 8-pin

Nvidia has increased the 3000 series’ maximum GPU temperature by 5 degrees to 93 C over the 2000 series, and graphics power has grown by 40% to 350 watts. This is a “firm” wattage. In previous generations, wattage factored in the use of the USB C connection for a virtual reality (VR) head-mounted display (HMD) or other devices.

This higher power consumption in the 3000 series comes with a 100-watt increase in the recommended power supply unit (PSU) wattage, so designers may need to upgrade to a larger PSU. The cards will ship with power adapters, and using third-party adaptors will void the warranty.

Transitions 

All of these changes have resulted in significant performance increases in benchmarking tests conducted by industry enthusiasts on the GeForce RTX 3080 Founders editions. Specifically, when compared to the 2080, the 3080 saw the following performance increases:

  • In standard rasterization-based applications, tests have shown a performance increase of up to 25%, with the average in the 15% to 20% range 
  • On applications using a combination of ray tracing and rasterization, performance increases have been as high as 30%
  •  With applications using only ray tracing, performance improvements climb to 45%
  • The DLSS technology can also significantly improve performance, but the amount of improvement will differ depending on the application it supports

Given these performance gains in the 3000 series, we recommend transitioning to the new cards, though be aware that there is supply constraint on certain 3000 series cards. Here are our engineers’ general recommendations based on the cards currently being used:

  • GeForce GTX 1080 Ti: We recommend moving to the GeForce RTX 3080, but if its MSRP of $700 – $800 is too high, then the GeForce RTX 3070 or Geforce RTX 3060Ti would be a good choice with price ranges between $400 to $500.
  • GeForce RTX 2070: Here, the GeForce RTX 3070  is in the same price range with performance meeting around the Geforce 2080Ti area albeit at a lower amount of VRAM.  The Geforce RTX 3060 Ti also works as a replacement for this card if you are only looking for a minor performance increase or a lower-cost card.
  • GeForce RTX 2080: If you wish to stay in the same price bracket but gain a significant performance jump, we recommend the GeForce RTX 3080.  If you do not need additional performance but need a drop-in replacement, the Geforce RTX 3060 Ti will meet the same performance at a far lower cost.
  • GeForce RTX 2080 Ti: If you require a high amount of VRAM for your application, then you will need the GeForce RTX 3090. Otherwise, we recommend transitioning to the GeForce RTX 3080 for its performance increase.  However, if you do not need a performance increase and your application does not require the amount of VRAM the 2080 Ti provided,  we would recommend transitioning to the Geforce RTX 3070.
MBX Systems

We're here to help

Chat bots are overrated. To talk to a real, knowledgable human, just tell us who you are and we’ll be in touch to answer your questions.