Nvidia limited the RTX 2080 Super’s memory performance
The GPU’s 15.5Gbps memory speeds are an artificial limitation.
Published: 23rd July 2019 | Source: Anandtech | Author: Mark Campbell
Nvidia limited the RTX 2080 Super’s memory performance, here’s why
Nvidia’s RTX 2080 Super has been released today (Read our RTX 2080 Super Review here), but with this release comes a revelation. This revelation comes in the form of the RTX 2080 Super’s memory performance, which is lower than it could be, at least according to Samsung’s GDDR6 memory model numbers.
Yes, Nvidia’s RTX 2080 Super may have its memory running at 15.5 Gbps, but the Samsung modules within the GPU are designed to run at higher speeds. To be precise, these GDDR6 memory modules are designed to run at 16 Gbps, which means that Nvidia is running these modules in an underclocked state.
If Nvidia ran their RTX 2080 Super at its GDDR6 memory’s rated clock speed, the graphics card would offer users over 512GB/s of memory performance, amounting to a memory bandwidth uplift of slightly more than 3%. This raises a question, why didn’t Nvidia run its RTX 2080 Super at Samsungs rated memory speed?
According to Nvidia, as stated by Anandtech, the RTX Super series of graphics cards were designed t be as easy to manufacture as possible, utilising the same basic Turing Silicon as existing RTX series graphics cards alongside similar cooler designs and identical PCB layouts. The RTX Super series is a Turing refresh. Simply put, Nvidia didn’t want to push memory speeds too high for manufacturing reasons.
Nvidia’s original RTX series graphics cards offered GDDR6 memory with maximum speeds of 14Gbps. Nvidia’s PCB designs are rated for this level of performance, and full 16Gbps memory speeds would require a PCB redesign. Nvidia limited the RTX 2080 Super to 15.5 Gbps memory speeds so that they, and their AIB partners, didn’t need to design new PCBs for their RTX 2080 Super. Why design a new PCB for a single graphics SKU as part of a GPU refresh?
The upside to this is that most RTX 2080 Super graphics cards will be able to overclock to 16Gbps memory speeds with ease, as the memory is rated to run at those speeds.
While Nvidia’s PCB designs are not designed with such high levels of memory performance in mind, most end-users should be able to find stability if they crank their RTX 2080 Super’s GDDR6 memory to 16Gbps.
| Nvidia Founders Edition Graphics | Geforce RTX 2080 Ti FE | Geforce RTX 2080 Super |
Geforce RTX 2080 FE | Geforce RTX 2070 Super |
Geforce RTX 2070 FE |
Geforce RTX 2060 Super |
Geforce RTX 2060 FE |
| Architecture | Turing | Turing | Turing | Turing | Turing | Turing | Turing |
| CUDA Cores | 4,352 | 3072 | 2944 | 2560 | 2304 | 2176 | 1920 |
| Base Clock |
1350MHz | 1650MHz | 1510MHz | 1605MHz | 1410MHz | 1470MHz | 1345MHz |
| Boost Clock | 1635MHz | 1815MHz | 1800MHz | 1770MHz | 1710MHz | 1650MHz | 1680MHz |
| Memory Type | GDDR6 | GDDR6 | GDDR6 | GDDR6 | GDDR6 | GDDR6 | GDDR6 |
| Memory Capacity | 11GB | 8GB | 8GB | 8GB | 8GB | 8GB | 6GB |
| Memory Speed | 14Gbps | 15. 5Gbps |
14Gbps | 14Gbps | 14Gbps | 14Gbps | 14Gbps |
| Memory Bandwidth | 616GB/s | 496GB/s | 448GB/s | 448GB/s | 448GB/s | 448GB/s | 336GB/s |
| Memory Bus Size | 352-bit | 256-bit | 256-bit | 256-bit | 256-bit | 256-bit | 192-bit |
| SLI | Via NVLink |
Via NVLink |
Via NVLink |
Via NVLink | N/A | N/A | N/A |
| US Pricing | $1199 | $699 | $799 | $499 | $599 | $429 | $349 |
| UK pricing | £1099 | £669 | £749 | £475 | £569 | £379 | £329 |
On a positive note, this launch confirms that Samsung is producing 16Gbps GDDR6 memory modules in large enough quantities for a major graphics cards launch.
This is great news for the future generation of both Radeon and Nvidia graphics. Samsung has already announced 18Gbps GDDR6 memory modules, but until now, only 14Gbps chips have been used within the GPU market. Perhaps we will see Nvidia release their next-generation of graphics cards with all 16Gbps GDDR6, or faster, memory solutions.
You can join the discussion on Nvidia limiting the RTX 2080 Super’s memory speeds to ease manufacturing on the OC3D Forums.
1 — The GPU’s 15.5Gbps memory speeds are an artificial limitation.«Prev 1 Next»
Most Recent Comments
Reply
Nvidia GeForce RTX 2080 and RTX 2080 Ti Overclocking Guide
Today we’re going to be diving a little deeper into overclocking the new Nvidia GeForce RTX 2080 Ti and RTX 2080, covering a whole range of things from how to overclock these cards, to typical achievable clock speeds, to performance and power consumption.
To start off this guide we’ll be going through our standard methodology for overclocking Nvidia graphics cards and specifically the new RTX series.
We’ll also be showing you how Nvidia’s new Scanner API works and just how good their one-click overclocking can be, comparing to a manual overclock, while also showing you how to do said manual overclock.
The cards we’re using today are the Founders Edition RTX 2080 Ti and Founders Edition RTX 2080. Both come ‘factory overclocked’ in some sense compared to base-level board partner models, but that OC is quite small at just 90 MHz on the core, and nothing on the memory. But the card we’re using doesn’t really matter, the steps you see here will apply to all RTX cards.
The first thing we’ll be doing is going through what’s potentially the easiest method of overclocking your RTX card: using Nvidia’s Scanner. One-click overclocking has been around for a while, but this time Nvidia has built a framework for doing it rather than letting board partners create their own algorithms, which the chip maker says is better tuned for their GPUs, more accurate, and more reliable.
To use Nvidia Scanner, you’ll need to grab the latest versions of any overclocking utility.
We tried EVGA Precision X1 and MSI Afterburner, both beta versions, and for the Scanner functionality we found Afterburner to be more reliable.
Had a few crash issues with the early version of Precision X1 which we assume will get fixed, whereas Afterburner seemed to work every time. It’s a bit disappointing as Precision has a much nicer interface than Afterburner, but we’ll get back to using Precision a bit later.
After you’ve installed and opened Afterburner you’ll see a bunch of dials and sliders, bit of a cluttered interface but that’s fine. What we’re interested in is the small bar graph icon to the left of the core clock slider. At this stage it’s not necessary to touch anything else in the app. Simply click on the bar graph icon, then click OC Scanner, and then in that window hit ‘Scan’. Now you can sit back and wait a while because the one-click overclock process takes around 15 to 20 minutes.
What the Scanner is doing is basically running a bunch of clock speed tests at a range of voltages to find the exact voltage-frequency curve for your graphics card.
It has its own test algorithm built-in to stress the GPU with the ability to recover from any hangs or crashes if the GPU is being pushed a bit too hard. It’s basically simulating what we’d do with a manual overclock, but doing it faster and potentially more accurately.
Don’t worry if the application hangs or your screen goes black temporarily during the process, that’s normal. At the end, you’ll be given an average overclock, but crucially you’ll get a full frequency curve, which potentially is a bit more efficient than a simple core frequency offset. You can see at the lower end of the voltage curve, we’re getting slightly higher overclocks than at the top.
For the RTX 2080 Ti on average we achieved +181 MHz, but at the top end our OC is just +150 MHz. With the RTX 2080, we achieved an average of +113 MHz and a top-end OC of around +90 MHz
At this point, we’d recommend heading back into the Afterburner main application and cranking up the power and temperature limits to the maximum.
We’ve never benchmarked a card, even a last-generation Pascal card, that didn’t like having these limits raised all the way, so with these RTX cards it just makes sense to crank it up. This will allow Nvidia’s GPU boost algorithm to push as high as possible on top of the frequency curve we’ve already set, going beyond the old limit.
One thing to note here is that we’ve set the power and temperature limit after running Nvidia Scanner. Nvidia says the Scanner only modifies the core clock, so if you change the power and temperature limits before hand, the Scanner might find different and potentially higher core clocks. However in our experience, we actually achieved lower clocks in the Scanner setting the power limit beforehand, so we’d recommend cranking it up after the Scanner is complete.
The other obvious limitation is the lack of memory clock gains, Nvidia Scanner only handles the GPU core and doesn’t touch memory. So if you’re after that simple, one-click overclock solution, you’ll be missing out on any gains you’d get from tweaking memory.
Core overclocking is much more important for getting performance gains, but a boost to memory can help out in some situations.
We’ve got the Scanner OC results now in the bag. Now let’s see how to perform a manual overclock, and after we’ll compare the manual overclock to the Scanner results.
For this we’re switching to EVGA Precision X1 because it has a much nicer and more intuitive interface.
For a manual overclock we want to start from a decent point, so again we’re going to crank the power limit and temperature limit to the maximum. We’re also going to scroll the bottom section across to the Temp Tuner, and adjust the curve so we’re getting the maximum clock speed at all possible temperatures. For our reference card this isn’t a big deal as we’re not reaching up to those highest temperatures, but if you are running above 84C, you’ll want to adjust this curve for maximum performance.
From here, it’s all about adjusting the two main sliders for memory and core frequency.
The basic steps are we want to increase each value by a reasonable amount, apply the overclock, then validate it in a program like 3DMark to ensure we’re not getting crashes at those settings.
The amount you increase the values is up to you, but note that you’re not going to damage your card by choosing a value that’s too high; instead you’ll just crash your system and have to reset it. No big deal.
So for the RTX 2080 Ti we started with around a +150 MHz offset on the core and +500 on the memory, those are fairly conservative figures for this card. That worked in 3DMark, so we pushed the core up to +200 and memory up to +650. If you want to play it safer we’d recommend only changing one of those values per test run, but the more you become familiar with the process, the more you can adjust at once.
With +200 on the core our 2080 Ti crashed in 3DMark, so now it’s a process of finding the exact limit. We stepped down in 10 MHz increments, eventually discovering that +180 was unstable, but +170 was perfectly fine.
From here we can try to push up the memory even further using a similar sort of process, and for our model we settled on +700 being the sweet spot.
The final stable overclocks for this card were +170 MHz on the core and +700 MHz on the memory, and that’s a fairly typical figure going on what we’ve seen from others. Note that if you have a non-FE or non factory-OC card, the core offset you’ll need will be higher as you’re coming from a lower starting point.
The vanilla RTX 2080 comes factory clocked a bit higher than the 2080 Ti, so we started with +100 MHz on the core and followed the same process. Eventually we settled on +110 MHz on the core and +700 MHz again on the memory for the RTX 2080.
Once you find what you think are stable overclocks it’s always a good idea to validate them further in a game running for several hours, something very GPU intensive. While 3DMark’s Time Spy gives a good indication of whether a card will work at a certain frequency, sometimes it will only crash under a longer test, so it’s good to test both.
A quick note on voltage, Nvidia’s Turing cards are voltage locked. In other words, Nvidia does not expose proper voltage controls to the user like you might get with overclocking a CPU on a motherboard. Instead we have a voltage slider, but it’s not really an offset, what it theoretically does is raise the voltage limit by a few steps, but still within what Nvidia deems safe. And that comes at the expense of lifespan: Nvidia claims you should get 5 years out of stock voltage and just 1 year with the voltage limit raised slightly, so keep that in mind.
In practice, raising the voltage slider did absolutely nothing for our overclock, so it’s not worth thinking about.
So on a pure clock offset basis, the Scanner OC coming in around +150 MHz at the top voltage range is lower than our +170 MHz manual offset for the 2080 Ti, with a similar gap for the 2080. Plus, of course, with manual overclocking we increased the memory frequency as well.
This indicates manual overclocking remains the way to go for many, and you can see that is the case when looking at our Shadow of the Tomb Raider results next.
..
Gigabyte Geforce RTX 2080 Ti Windforce 11GB GDDR6 Graphics Card, Cuda Cores 4352, 1545 MHz core clock, GDDR6, 352-bit, 14000 MHz memory clock, 3 x DisplayPort, 1 x HDMI, 1 x USB
Gigabyte Geforce RTX 2080 Ti Windforce 11GB GDDR6 Graphics Card overview
Gigabyte Geforce RTX 2080 Ti Windforce OC 11GB GDDR6 Graphics Card
REAL-TIME RAY TRACING IN GAMES
Ray tracing is the definitive solution for like life lighting, reflections and shadows, offering a level of realism far beyond what’s possible using tradition rendering techniques. NVIDIA Turing is the first GPU capable of real-time ray tracing.
NVIDIA TURING ARCHITECTURE
This revolutionary architecture, combine3d with our all-new GeForce RTX platform, fuse together real-time ray tracing, artificial intelligence and programmable shading. You’ve never created and enjoyed games like this before.
| UP TO 6X FASTER PERFORMANCE |
REAL-TIME RAY TRACING IN GAMES 9002 3 | POWERFUL AI ENHANCED GRAPHICS |
WINDFORCE 3X COOLING SYSTEM
The WINDFORCE 3X cooling system features 3x 80mm unique blade fans, alternate spinning fan, 6 composite copper heat pipes, heat-pipe direct touch and 3D active fan functionality, together delivering an effective heat dissipation capacity for higher performance at lower temperatures.
ALTERNATE SPINNING
The middle fan spins in reverse to optimize airflow for heat dissipation , enabling more efficient performance at lower temperatures.
UNIQUE BLADE FAN
The airflow is spilt by the triangular fan edge, and guided smoothly through the 3D stripe curve on the fan surface, effectively enhancing the airflow.
3D ACTIVE FAN
The 3D Active Fan provides semi-passive cooling, and the fans will remain off when the GPU is in a low load or low power game. It allows gamers to enjoy gameplay in complete silence when the system is running light or idle.
HEAT PIPES DIRECT TOUCH
The shape of the pure copper heat pipe maximizes the direct contact area with the GPU, enhancing heat transfer. The heat pipe also covers the VRAM through a large metal plate contact to ensure proper cooling.
COMPOSITE HEAT-PIPES
The composite heat-pipes combines thermal conductivity and phase transition to efficiently manage the heat transfer between two solid interfaces which increases cooling capacity.
RGB FUSION
With 16.7M customizable color options and numerous lighting effects, you can choose the right scheme for your gaming rig with the AORUS engine.
IN TOTAL COMMAND
The interface and functionality of the software is subject to each model.
SPECIFICATIONS
-
Graphics Processing
GeForce® RTX 2080 Ti
-
MHz
-
RTX-OPS
76
-
Memory Clock
14000 MHz
-
Memory Size
6
11 GB
-
Memory Type
GDDR6
-
Memory Bus
352 bit 005 Memory Bandwidth (GB/sec)
616 GB/s
-
Card Bus
PCI- E 3.0 x 16
-
Digital max resolution
7680×4320@60Hz
-
Multi-view
4
-
Card size
280.35*116.45*40.24mm
-
PCB Form
ATX
-
DirectX
12
-
OpenGL
4.
5 -
SU
750W ( With Two 8-Pin EXTERNAL POWER CONNECTOR)
-
SLI support
2-way NVIDIA NVLINK TM
-
I/O
DisplayPort 1.4 x3, HDMI 2.0b x1,
USB Type-C ™ (support VirtualLink TM ) x1
NVIDIA GeForce RTX 3060 Ti Founders Edition video card review — Boiling Machine
The NVIDIA GeForce RTX 3060 Ti appears to be one of the two most balanced cards in the GeForce RTX 30 series. Judging by the technical specifications, it can face off against the GeForce RTX 2080 Super, one of NVIDIA’s past flagships.
Packaging and delivery
The NVIDIA GeForce RTX 3060 Ti Founders Edition graphics card comes in a formal black package. The box has a minimum design, it contains the video card itself, quick guides and an adapter for the power supply.
NVIDIA GeForce RTX 3060 Ti Founders Edition
Looks Once upon a time, NVIDIA paid little attention to the design of original reference graphics cards.
This is logical, because the company’s partners (ASUS, MSI, Gigabyte, Zotac, Palit and others) release video cards with improved cooling systems, so why bother? You can install a simple CO with a turbine, decorate the video card casing a little and you’re done. However, the reference models, which are now called the Founders Edition, were, are and remain popular among users. Since 2018, since the 20th generation of GeForce RTX, NVIDIA has changed the approach to the design of its own video cards.
Starting with the GeForce RTX 20 series, NVIDIA has abandoned the use of a turbine and replaced it with two large fans. Then, in the 30th series of video adapters performed by Founders Edition, the approach to the location of the «turntables» has changed dramatically. The RTX 3080 and RTX 3090 have fans on the outside and inside of the graphics card, while the RTX 3070 and RTX 3060 Ti have fans on one side, the outside.
Reference RTX 30-series graphics cards have changed PCB design.
It has become significantly shorter than the RTX 20 series. The heat pack of the video card has increased greatly and one fan with a conventional heatsink will not be able to remove so much heat. Nevertheless, the RTX 3060 Ti Founders Edition video card turned out to be more compact than many analogues from NVIDIA partners, its length and width are 242×112 mm, while the length of partner video cards is most often about 300 mm or more. In height, the RTX 3060 Ti FE will take up two slots — for modern graphics, this is in the order of things. The video adapter in question can be installed even in compact cases.
The black RTX 3060 Ti FE heatsink sits in an open shroud that frames the sides of the heatsink and visually divides the graphics card into two parts. As a result, the left fan exhausts hot graphics air through holes in the bracket that secures the RTX 3060 Ti FE to the system unit. The second, right fan, blows through the cooling system, for this a separate hole is made in the backplate.
Considering that fans are often installed on top of gaming computers for blowing, thoughtful circulation of air flows is provided. However, some users complained about the increase in CPU temperature in games. This is most likely caused by improper installation of the processor CO fan, which took air from the bottom of it. When assembling a computer, it is very important to place the fans correctly, otherwise you can even get a negative result.
The radiator is painted black and has a copper plate attached to it, from which the heat pipes extend. The design is standard, and given the RTX 3060 Ti’s not the highest GPU heat dissipation, a polished copper base will be enough to transfer heat from the video core.
Many video card manufacturers add a mechanism to stop the fans during idle time in the BIOS to reduce noise in the computer case. NVIDIA did the same — the RTX 3060 Ti turntables stop if the GPU temperature does not exceed 50 degrees Celsius and its power consumption is less than 30 watts.
This means that while doing office work and browsing simple Internet sites that do not create a large load on the GPU, the graphics will work silently in passive mode.
Three DisplayPort version 1.4a and HDMI version 2.1 are available for connecting monitors, which allows you to display images in resolutions up to 8K. The graphics card is powered by a new 12-pin Molex MicroFit 3.0 connector. The video card comes with an adapter for the power connector, which means that you don’t have to think about replacing the PSU.
As we wrote above, for the 30th series NVIDIA has developed a new reference PCB design — they have become more compact, and at the same time, the density of elements per square centimeter has increased. However, this did not prevent developers from placing 10 power phases that work according to the 8 + 2 scheme, where 8 phases are used for the video core, and the remaining 2 phases are for video memory.
Alpha & Omega Semiconductor AOZ5311NGI mosfets are used in the GPU power subsystem, and Sinopower SM7342EKKP is used to power the video memory.
Similar transistors are used in other NVIDIA video cards. The uPI Semiconductor uP9512R controller is responsible for managing the power phases of the video core, and another controller, the UPI Semiconductor uP1666Q, manages the 2-phase memory power subsystem.
The 8 GB GDDR6 video memory is based on eight Samsung chips, their exact model is Samsung K4Z80325BC-HC14. The nominal memory clock speed is 14000 MHz.
Ampere Architecture Specifications
One of the key benefits of the NVIDIA Ampere architecture is significantly increased performance. NVIDIA has been able to achieve this by increasing the number of CUDA cores and making significant architecture changes. As a result, the GeForce RTX 3060 Ti graphics card is close to the top-end GeForce RTX 2080 Super in terms of performance, although in terms of positioning it is a replacement for the RTX 2060 Super. Let’s take a closer look at the specifications of the RTX 3060 Ti and RTX 2080 Super.
| Model | NVIDIA GeForce RTX 3060 Ti | NVIDIA GeForce RTX 2080 Super |
| Video core | GA104, Ampere | TU104, Turing |
| Process | 8 nm Samsung | 12 nm (TSMC) |
| Number of transistors | 17. 4 billion |
13.6 billion |
| Crystal area | 392.5 mm 2 | 545 mm 2 |
| CUDA cores | 4864 | 3072 |
| TMU (texture blocks) | 152 | 192 |
| ROPs | 80 | 64 |
| Tensor cores | 152 | 384 |
| Ray Tracing Cores | 38 | 48 |
| Streaming Multiprocessors | 38 | 48 |
| GPU clock (base/boost) | 1410 / 1670 MHz | 1605 / 1815 MHz |
| Video memory | 8 GB GDDR6 | 8 GB GDDR6 |
| Video memory frequency | 14000 MHz | 15500 MHz |
| Video memory bus | 256 bit | 256 bit |
| PCI Express slot | 16x PCIe 4.0 | PCIe 3.0 |
| TDP | 200 W | 250 W |
| Power connectors | 12-pin | 6 poles + 8 poles |
| Recommended PSU Wattage (according to NVIDIA) | 600 W | 650 |
| Dimensions (length / width) | 242×112 mm (two PCIe slots) |
267x112mm (two PCIe slots) |
Switching to the new 8nm process technology allowed NVIDIA to increase the number of transistors in a chip by 27.
9% and increase the number of CUDA cores by 58.3%. However, the high performance of Ampere depends not only on the transistor budget, but also on the design features of the new streaming multiprocessors.
In the previous Turing architecture, the same Streaming Multiprocessor had FP32 (floating point) and INT32 (integer) blocks. Each block involved 16 cores and had its own data channel, which made it possible to carry out FP32 and INT32 operations simultaneously. In total, in one streaming multiprocessor, there were 4 clusters with FP32 and INT32 blocks, with 64 cores in each block, which performed 64 operations of each type per clock.
In the new Ampere architecture, NVIDIA not only added another FP32 block with 16 cores to each cluster, but also combined an INT32 block with one of the FP32 blocks in the first data link. The second channel still serves a separate FP32 unit. As a result of optimizations and an increase in the number of blocks in the streaming multiprocessor, 128 FP32 operations or 64 FP32 and 64 INT32 operations can be performed simultaneously in 1 cycle — as in Turing.
If the number of CUDA cores has increased, then the number of RT cores has remained unchanged — one core per streaming multiprocessor. At the same time, NVIDIA increased the performance of ray tracing cores from 34 RT-teraflops (in Turing) to 58 RT-teraflops (Ampere). The number of tensor cores used in NVIDIA DLSS smart anti-aliasing in the new Ampere architecture has been halved — one core per TMU instead of two. However, NVIDIA claims that due to the increase in the size of multiplied matrices from 4×4 to 4×8, the efficiency of tensor cores has doubled. It turns out that nothing has changed in terms of tensor cores — the increased performance compensates for the twofold decrease in their number.
The video memory type of the GeForce RTX 3060 Ti is the same as the RTX 3070 and RTX 2080 Super, it is GDDR6 with a high clock speed, which is 14000 MHz for the RTX 3060 Ti. The maximum clock speed of the video core depends on the conditions in which the video card will work.
For several years, GPU performance has been largely limited by its cooling system, the more efficient it works, the higher the performance increase can be.
The move to 8nm in the Ampere architecture allowed for more transistors and slightly lower GPU power consumption if we continue to compare the card to the RTX 2080 Super. So, for the GeForce RTX 3060 Ti, NVIDIA recommends using a 600W PSU, while for the RTX 2080 Super, a 650W PSU was recommended. We, in turn, recommend considering all system components when choosing a power supply. For example, the TDP of modern processors can reach 300 or more watts at the peak of their performance, although official websites may indicate a value of 125 watts. It is also worth paying attention to the 80 PLUS certificate for the power supply. It is best if it is a 80 PLUS Gold, 80 PLUS Platinum or 80 PLUS Titanium PSU.
Another advantage of the GeForce 30-series and RTX 3060 Ti graphics is support for the Gen 4 PCI Express bus. All new NVIDIA graphics cards can be connected to PCIe 4.
0 x16 slots with twice the bandwidth of PCI Express 3.0 x16. Backward compatibility with previous versions of PCIe is also declared — you don’t have to assemble a new system to install the GeForce RTX 3060 Ti.
Other NVIDIA Ampere Architecture Changes
Updated DLSS 2.1 Smart Anti-Aliasing System
Smart Anti-Aliasing has become a hallmark for all GeForce RTX series graphics cards. Using DLSS, the image is rendered at a resolution lower than the resolution set in the game, and then, using tensor cores and artificial intelligence, the frame is extrapolated to the required resolution. The DLSS 2.1 version of Smart Anti-Aliasing does a better job and requires less input from the game to work.
In practice, its support depends on the game developer.
In order for it to become available, developers must negotiate with NVIDIA and implement DLSS support in their product. Unfortunately, because of this, it will not be possible to use DLSS on older games. Developers of new and popular games like Cyberpunk 2077 add DLSS at the time of game creation. And in Tom Clancy’s Rainbow Six Siege, support for smart anti-aliasing appeared in June 2021. And in Fortnite, DLSS support appeared in September 2020. Those. there is a kind of “finishing projects”. It is also worth noting that some users notice “soap” on the screen if they select the DLSS mode marked as “maximum performance” in the game. We recommend not using anything other than the «quality» mode. In extreme cases, use the «balance», everything else greatly negatively affects the quality of the picture.
Latency Reduction, Nvidia Reflex Technology
Today’s gamers want to not only increase FPS, but also reduce input lag in their games, especially during online battles. The term gaming lag (also known as Input Lag) is familiar to console users — not all TVs are designed for gaming, different TV models will have different reaction times to player actions.
For example, on systems with high latency, more time will elapse between pressing the fire button and firing on the screen. We are talking about milliseconds, and on the one hand, this is an extremely short time, and on the other hand, sometimes one moment is not enough for a long-awaited victory.
Computers also have a high latency problem. In order to reduce input lag and increase player responsiveness, NVIDIA developed Nvidia Reflex technology, which appeared along with the Ampere architecture. If the performance in the game is limited by the video card, then during the shot using Nvidia Reflex, the processor immediately sends a signal to start rendering a new frame to the video card. In this case, the rendering of a new frame begins before the GPU copes with the previous frame. As a result, the render queue is either significantly reduced or eliminated altogether, and the player reacts faster to events in the game.
Like DLSS, NVIDIA Reflex has its own quirks. For example, the technology must also be supported by the game developer.
On the other hand, Reflex is supported by NVIDIA graphics cards starting with the GeForce GTX 900.
Following NVIDIA Reflex, the Reflex Latency Analyzer technology was introduced, which allows you to measure latency. The technology itself recognizes mouse clicks and measures the time from the click to the action. But to activate it, you need a certified gaming mouse and the same gaming monitor.
NVIDIA Resizable Bar
The amount of internal memory for video cards is gradually increasing, but few gamers know that no matter how much memory a video card has, the processor has access to only 256 MB at a time. As a result, the loading time of textures increases and the loading time of the game itself increases. NVIDIA decided to eliminate the «bottleneck» and developed the Resizable Bar technology.
The Resizable Bar allows the system to access all of the graphics card’s memory and excludes the processor from loading textures, which can increase FPS. The new technology is supported by all GeForce RTX 30 series graphics cards.
To activate Resizable Bar, you need a motherboard that supports this technology, support for the game itself, and video drivers version 456.89or newer. Declared support for AMD 500 and 400 series chipsets (for boards compatible with Ryzen 5000 processors) and support for 11th generation Intel Core processors. The Intel Z590, Z490, B460 and h570 chipsets are compatible with the new technology. Check your motherboard manufacturer’s website for details on NVIDIA Resizable Bar support, as a BIOS update may be required to activate the new technology.
Nvidia Broadcast and help for a streamer
Game streaming is relatively new, but that doesn’t stop it from gaining popularity every year. To make broadcasting easier and better, NVIDIA has developed the Nvidia Broadcast app and added support for the powerful new AV1 codec to the RTX 30-series graphics.
The first problem familiar to anyone who has streamed or even tried to play with friends while talking in voice chat is external noise.
Noise can be anything from keyboard and mouse clicks to extraneous conversations and the clinking of dishes. Even constant microphone background noise can annoy your viewers. Nvidia Broadcast will remove unnecessary sounds using artificial intelligence algorithms.
The second problem is the background behind the player. Not every streamer has the opportunity to furnish their own beautiful studio or stretch a chroma key behind their back. Nvidia Broadcast solves this problem too. The utility will independently identify the streamer and replace the background behind him, and will constantly keep focus on the user’s face. Also, Nvidia Broadcast will remove excess noise from the frame, which always spoils the picture.
Testing
To evaluate the performance of the NVIDIA GeForce RTX 3060 Ti Founders Edition, we used a set of 3DMark and several modern games. From them you can understand the performance of the video card.
| 3DMark test | Results |
| Time Spy | 10 679 |
| Time Spy Extreme | 5 231 |
| Port Royal | 6 831 |
| NVIDIA DLSS Feature Test | 31. 58 (DLSS OFF)
76.58 (DLSS ON) |
The most popular 3DMark Time Spy benchmark scores 10,679 points. This is a high result, which fits into the fork of the average results published on the 3DMark website. According to him, in most cases, NVIDIA GeForce RTX 3060 Ti graphics cards show a score of 10,500 to 11,000 points. Statistics collected on the basis of more than 40 thousand results from around the world.
The latest 3DMark test shows the advantage of DLSS technology. The first run of the test scored 31.58 FPS, and turning on Smart Anti-Aliasing boosted the score by over 100% to 76.58 FPS. Of course, it is worth noting that we are talking about testing in the benchmark and in games, the efficiency of DLSS will be different.
Now let’s move on to games. The tests were run at two resolutions: 1920×1080 and 2560×1440. The first resolution is the most popular, since for a long time Full HD was the ceiling for most monitors, and models with WQHD (the same 2560×1440) were expensive and were not available to most gamers.
Today the situation has not changed much, although the resolution of 2560×1440 is gaining more and more popularity.
During the tests, all settings were set to the maximum value, preset profiles were ignored.
| Game name and settings | Permit | Results |
| Cyberpunk 2077 (Maximum graphics settings, DLSS — Balanced) |
1920×1080 | 75 |
| 2560×1440 | 55 | |
| Red Dead Redemption 2
(Maximum graphics settings) |
1920×1080 | 77 |
| 2560×1440 | 63 | |
| Metro Exodus Enhanced Edition (Maximum graphics settings, DLSS — Balanced) |
1920×1080 | 54 |
| 2560×1440 | 50 | |
| Shadow of the Tomb Raider Maximum graphics settings, DLSS ON) |
1920×1080 | 87 |
| 2560×1440 | 65 | |
| Tom Clancy’s Ghost Recon: Wildlands (Maximum graphics settings) |
1920×1080 | 76 |
| 2560×1440 | 63 | |
| Dirt 5
(maximum setting) |
1920×1080 | 102 |
| 2560×1440 | 82 | |
| Need for Speed Heat
(maximum setting) |
1920×1080 | 76 |
| 2560×1440 | 65 | |
| Resident Evil 3
(Maximum graphics settings) |
1920×1080 | 240 |
| 2560×1440 | 180 |
Released late last year, Cyberpunk 2077 has become a landmark for the industry.
Despite the large number of bugs reported by many players, Cyberpunk 2077 became popular and even today system performance is largely measured by the number of FPS in this game. With maximum graphics settings and DLSS turned on in the «balance» mode at a resolution of 1920×1080 received 75 frames, and at 2560×1440 — already 55 frames, both results can be called playable. But here it is worth saying right away that we are talking about medium frames. In fact, this means that the game will drop from 55 to 35-40 frames in particularly violent moments. That is why it is impossible to focus solely on the middle frame — you need to lay a performance margin for scenes with a large number of particles. Because if you have an average frame of about 40-45, with a big bang you can easily get 25-30 frames, and this is already frank brakes.
Red Dead Redemption 2 is another heavy game for any modern system. Although it has not become as popular as GTA V, RDR 2 is in no hurry to leave the «top 100» rating of popular games on Steam.
At maximum graphics settings, we got results similar to Cyberpunk 2077, namely 77 frames for Full HD and 63 frames for WQHD.
Metro Exodus Enhanced Edition is a new and improved version of the original Metro Exodus that puts a lot of strain on any system. We can say that Metro Exodus Enhanced Edition turned out to be the heaviest game on our test list, and this despite the included DLSS in Balanced mode. At resolution 1920×1080 received 54 frames per second, and at 2560×1440 already 50 FPS. To be honest, the gap is not that big.
Shadow of the Tomb Raider was released in 2018 and was one of the first games to receive NVIDIA DLSS support. Despite such respectable age for the gaming industry, the game requires a powerful graphics card to be able to play comfortably at maximum settings. With DLSS enabled, we only got 87 at 1920×1080 and 65 FPS after upscaling to 2560×1440.
Tom Clancy’s Ghost Recon: Wildlands is a popular tactical shooter with the name of the famous writer Tom Clancy in the title, which can be considered a separate brand in the gaming industry.
The game does not support DLLS, at a resolution of 1920×1080 we got only 76, and when switching to 2560×1400 pixels — 63 frames.
Car simulators were, are and will be one of the most popular and sought after games. For the test, we took Dirt 5, in which we got 102 FPS (1920×1080) and 82 (2560×1400) FPS, as well as Need for Speed Heat, in which the result was lower, namely 76 frames for Full HD and 65 frames for 2560×1440.
In the Resident Evil 3 remaster we got high FPS values of 240 FPS for Full HD and 180 FPS for 2560×1440. The system requirements of the RE 3 remaster are very low, but they heavily load the video memory. We can safely say that eight gigabytes for these modes is more than enough.
Conclusion
Video cards of the 1060/2060/3060 series will always be the most popular and in demand among users. It’s all about the combination of affordable price and the performance they provide. An example is the GTX 1060, which has long been the most popular video card according to Steam statistics.
The video card in question, the NVIDIA GeForce RTX 3060 Ti Founders Edition, can repeat the success of its predecessor, as its performance is comparable to the previous flagship, the GeForce RTX 2080 Super, and in the RTX 30 series lineup, it ranks in the middle place, judging by the combination of price and performance.
NVIDIA did a great job during the creation of the Ampere architecture. Due to the increase in the number of CUDA cores and the ability of INT32 blocks to work with FP32 calculations, the NVIDIA GeForce RTX 3060 Ti Founders Edition performed well during testing at Full HD resolution. The graphics card is perfect for it and even offers some performance headroom. At 2560 x 1440, things are a little more complicated. In new games, the video card passes literally along the lower border, and this is with DLSS Balanced activated. And without it, the values will be much worse. So right now, the RTX 3060 Ti can play most games at 2560 x 1440 at or near max settings, but there’s no headroom for future games at all.