Table of Contents
Today we get our first taste of Nvidia’s flagship RTX 3090 GPU, and thanks to MSI, we’ll be testing their RTX 3090 Gaming X Trio. It’s been two years since Nvidia released a new architecture (not that they needed to), so we’re excited to see what Ampere brings to the table. Nvidia claims the RTX 3090 is 1.5X faster than the Titan RTX Turing-based GPU, and if true, this is a substantial generational leap in performance. Per usual, we take that value with a grain of salt because, well, marketing.
Before we get to the performance metrics, we’ll go over some key features of the Ampere architecture and how it differs from the last generation. This should have been done with an RTX 3080 review, but due to some unfortunate events (by a card partner), we missed out on a launch day sample, so we’re including that information here.
Let’s dive in and see for ourselves what Nvidia’s new flagship GPU is capable of:
Specifications and Features
The table below is based on the Nvidia Founders Edition GPUs.
Nvidia RTX 3000 Series Specifications | |||
Model | RTX 3090 | RTX 3080 | RTX 3070 |
Architecture | Ampere (GA102) | Ampere (GA102) | Ampere (GA104) |
Manufacturing | Samsung 8 nm | Samsung 8 nm | Samsung 8 nm |
CUDA Cores | 10,496 | 8,704 | 5,888 |
RT Cores | 82 | 68 | 46 |
Tensor Cores (2nd Gen) | 328 | 272 | 184 |
Texture Units (3rd Gen) | 328 | 272 | 184 |
ROPs | 96 | 88 | 64 |
L2 Cache | 6 MB | 5 MB | 4 MB |
Base Clock | 1,400 MHz | 1,440 MHz | 1,500 MHz |
Boost Clock | 1,700 MHz | 1,710 MHz | 1,730 MHz |
Memory | 24 GB GDDR6X | 10 GB GDDR6X | 8 GB GDDR6 |
Memory Speed | 935.8 GBps | 760 GBps | 512 GBps |
Memory Bus | 384-bit | 320-bit | 256-bit |
Supplementary Power | 2x PCIe 8-pin | 2x PCIe 8-pin | 1x PCIe 8-pin |
Standard Display Connectors | HDMI 2.1, 3x DisplayPort 1.4a | HDMI 2.1, 3x DisplayPort 1.4a | HDMI 2.1, 3x DisplayPort 1.4a |
Max Resolution | 8K (7680 x 4320) | 8K (7680 x 4320) | 8K (7680 x 4320) |
TDP | 350 W | 320 W | 220 W |
Release Date | 09/24/2020 | 09/17/2020 | 10/2020 |
MSRP | $1,499 | $699 | $499 |
Ampere Architecture and Technologies
With Ampere, we see a new silicon fabrication process moving away from TSMC’s 12 nm to Samsung’s 8N (eight nanometer) node. According to Nvidia, this leverages 1.9 times the performance per watt compared to the previous generation Turing architecture. This node shrink allows for a much higher transistor density with Ampere sporting nearly twice as many per square millimeter for a whopping 28 billion in the full GA102 die which measures 628 mm². It does appear as if Nvidia was going for muscle and not power savings with the RTX 3090, judging by the TDP, which is up to 350 W compared to 280 W for the RTX 2080 Ti.
Taking a closer look at the Nvidia RTX 3090, we find it uses the full Ampere GA102 die pictured above minus two SM (stream multiprocessors), which helps to increase yields. The RTX 3090 has a total of 82 SMs enabled with each SM sporting 128 CUDA cores for a total of 10,490. This value is more than double the 4,352 found in the RTX 2080 Ti. The processing power is spread across seven GPCs (Graphics Processing Clusters) with a shared L2 cache and a Giga Thread Engine to coordinate the work between GPCs while utilizing the full 384-bit wide memory interface connecting a massive 24 GB of GDDR6X to the processing cores. All RTX 30XX GPUs feature a PCI-Express 4.0 x16 host interface, but only the RTX 3090 has the updated NVLink multi-GPU interface.
The Ampere Streaming Multiprocessor
Nvidia’s engineers concentrated much of their effort into improving the throughput of the SM units for Ampere. These stream multiprocessors are the main engine of Nvidia’s GPUs, responsible for some serious number crunching, and they claim the Ampere SM has double the capacity of the Turing SM. Not surprising when we see the CUDA core count per SM has been doubled from 64 to 128 when compared to the RTX 2080 Ti, with half of these (64) capable of concurrent floating-point and integer data paths, much like the Turing architecture.
With the introduction of the ray tracing (RT moving forward) cores specifically designed to process performance-intensive ray tracing workloads, Nvidia brought real-time video game ray tracing to the masses. The second-generation RT core, according to Nvidia, has double the triangle intersection rates of the first generation RT core, in essence doubling its performance. Nvidia is also introducing Ray-Traced motion blur with the addition of a new logic block, which can help determine whether or not all the ray intersection need to be recalculated during movement, which in turn boosts the performance of the RT core.
Nvidia’s third-generation tensor core is purpose-built for AI deep-learning. How does that benefit your average gamer? DLSS (deep learning supersampling) leverages these tensor cores for real-time rendering through AI-based super-resolution rendering fewer pixels and then using AI to construct sharp, higher resolution images. With DLSS enabled titles, the GPU engine will produce lower resolution images, increasing frame rates.
DLSS has to be incorporated into the game title since there is a training process used which compares output images to ultra-high quality 16K images, and the differences are reported back to the network. This is repeated tens of thousands of times using supercomputers until the network reliably outputs high quality, high-resolution images. Once the network is trained, the AI model for that specific title is delivered via Nvidia’s Game-Ready drivers. This time around, Nvidia claims that DLSS 2.0 will enable 8K gaming with the RTX 3090 GPU.
GDDR6X
Last, is the Micron and Nvidia developed GDDR6X RAM. This new memory works off the GDDR6 platform but introduces a new technology called PAM4 (Pulse-Amplitude Modulation 4). PAM4 enables four values per cycle using different voltage levels allowing it to deliver double the data per cycle compared to GDDR6. This required new data encoding to translate the new signaling technique consistently. We also have a new, more robust memory training algorithm that can adapt to environmental changes, such as temperature or EMI interference. Bringing it all together, the 24 GB of GDDR6X on the RTX 3090 is capable of up to 19.5 Gbps data transfer and a staggering 940 GB/s bandwidth.
MSI RTX 3090 Gaming X Trio
The RTX 3090 Gaming X Trio from MSI comes with their updated Tri Frozr 2 cooling system. MSI’s Frozr series coolers have always been exceptional, and indeed, the Tri Frozr 2 is no exception. It all starts with their “Core Pipes,” which are precision machined heat pipes for maximum contact of the GPU surface. These pipes snake their way through an improved fin array with deflectors for added surface area and air-flow control along with Wave-curved 2.0 fin edges, which improve harmonics and reduce noise. Pushing air through the cooler are three large Torx 4.0 fans designed for maximum flow and minimum noise, which sit on top of dual ball bearings for years of quiet operation.
MSI is also using a full metal backplate, which incorporates close quarters heat pipes which add additional cooling for the memory modules located on the back of the PCB. The PCB is custom-built for the RTX 3090 Gaming X Trio for added power with three 8-pin PCIe power connections, more security from accidental electrical damage with additional fuses, and two ounces of thickened copper for increased conductivity and improved heat dissipation. Considering the RTX 3090 has a TDP of 350 W, or more, all this attention to cooling will surely pay off.
MSI has included a couple of other additions that work to maintain the integrity of the card. First, we have a rigid anti-bending strap incorporated into the card’s construction, which spreads support over the length of the card eliminating weak areas. The second is an add-in support bracket that can be attached to the PC case to help take any strain off of the PCIe slot connection.
We also have the usual MSI software starting with Dragon Center, which can be used to monitor, tweak, and optimize your card in real-time. Mystic light for control of the RGB LED operation and coordination with other mystic light compatible peripherals. Last but not least, MSI Afterburner, for complete control of the card’s operation, including overclocking the core and memory, custom fan curves as well as monitoring thermals.
Retail Packaging and Accessories
MSI’s retail packaging for the card is mostly black with Nvidia green. The card’s make is located in the bottom right corner with MSI’s model, “Gaming X Trio”, directly above. Placed prominently in the middle is a picture of the card itself. The back of the package lists some specifications and features showing off the Tri Frozr 2 cooling capabilities.
Opening up the box, owners are greeted by the accessories envelope. Below that we’ll find the card sitting comfortably and securely inside an anti-static bag and form-fitting foam. MSI includes a quick user’s guide and “Lucky the Dragon” booklet for installation and MSI Afterburner directions. One other addition is the support bracket to help support the GPU. The card itself is quite large, measuring 13.1 x 5.5 x 2.2″ (335 x 140 x 56mm) and weighing just over 3.3 lbs (1.5 kg), so this extra bracket may be handy. Be sure to check your case to ensure this card will fit.
Meet the MSI RTX 3090 Gaming X Trio
Our first look at the card up close shows the three large Torx 4.0 fans with the MSI Dragon on the hub as the primary feature. The shroud is black with angled grey accents around the fans. At diagonals from the center, fan are translucent strips for the RGB LEDs. There is also an RGB LED strip along the edge of the backplate. The MSI RTX 3090 Gaming X Trio, with its black and grey shroud and full-length backplate, should easily fit in with any build theme.
A Closer Look
The RTX 3090 Gaming X Trio includes the typical outputs for Ampere based cards with three DisplayPorts (1.4a) and one HDMI (2.1) output with a maximum resolution of 7680×4320 (8K). The IO plate is vented in order to let some airflow out of the case. Power is handled by three 8-pin PCIe power leads for a total of 525 W available power, including the PCIe slot.
Taking the heatsink off the card exposes the custom 18 phase VRM using OnSemiconductor NCP30245 45 A MOSFETs. The heatsink itself has seven heat pipes weaving its way through the fin array and coming together in the nickel-plated base. The rear bank of memory is fully covered by the backplate, which uses “close quarters heat pipes” for additional cooling. The main memory and the MOSFETs make direct contact with the heatsink for best cooling.
Below is a closeup of the power bits, Micron GDDR6X, and the GA102-300-A1 DIE and control ICs.
As you can see from the GPUz shot above, the MSI RTX Gaming X Trio has a stock operating clock of 1395 MHz with a 1785 MHz Boost. The full boost speed from this sample managed to read 2025 MHz under load.
MSI RTX 3090 Gaming X Trio on the test bench…
Test System and Benchmark Methods
Test System Components | |
Motherboard | ASUS ROG Maximus XII Extreme, ASRock Z490 PG Velocita |
CPU | Intel i9-10900K @ stock |
CPU Cooler | EVGA CLC 240 |
Memory | 2×8 GB G.Skill Royal 3600 MHz CL16-16-16-36 |
SSD | Gigabyte Aorus 2 TB NVMe Gen4 (OS + Applications) |
Power Supply | EVGA 750W G3 |
Video Card | MSI RTX 3090 Gaming X Trio @ Stock, Nvidia 456.38 Win10 64-bit WHQL drivers |
Our test system is based on the latest mainstream Intel z490 platform and uses the i9-10900K 10/20t CPU. The CPU is overclocked to 4.9 GHz on all cores/threads with cache set to 4.3 GHz. The clock speeds used provides a good base to minimize any limitations the CPU may have on our titles, particularly when using the lower resolutions, and should be attainable with a good air cooler or better. The DRAM is in a 2×8 GB configuration at 3600 MHz with CL16-16-16-36-2T timings, a middle of the road option balancing performance and cost.
We have made some significant changes since the last update adding a few new titles and dropping some of the older games. More details can be found in the GPU Testing Procedure article, which we have updated with our latest benchmarks. Below is a quick summary for easy reference.
- UL 3DMark Time Spy – Default settings
- UL 3DMark Fire Strike (Extreme) – Default settings
- UL 3DMark Port Royal – Default Settings (Ray Tracing capable cards only)
- Unigine Superposition – Performance, 1080p High
- Shadow of the Tomb Raider – DX12, “Highest” preset (will add RTX when it has been patched)
- The Division 2 – DX12, Ultra preset, VSync Off
- Assassin’s Creed: Odyssey – Ultra High preset, VSync Off
- Far Cry New Dawn – Ultra defaults
- F1 2020 – DX12, Very High defaults, TAA, and x16 AF, Australia track, do not show FPS counter
- Metro: Exodus – DX12, Ultra defaults
Synthetic Benchmarks
Our first set of benchmarks hail from Underwriters Laboratories, who acquired Futuremark back in 2014. Earlier in 2018, a rebrand occurred, and since that time, Futuremark is now UL. The benchmarks have not changed, just the name. We chose to stick with 3DMark Fire Strike (Extreme) and 3DMark Time Spy as these tests give users a good idea of performance on modern titles.
3DMark Fire Strike (Extreme) is a DX11-based test that runs at 1080p resolution. UL says the graphics are rendered with detail and complexity far beyond other DX11 benchmarks and games. 3DMark Time Spy is a DX12 benchmark designed for Windows 10 PCs. It supports new API features such as asynchronous compute, explicit multi-adapter, multi-threading, and runs at 2560×1440 resolution. 3DMark Port Royal is the first Ray Tracing benchmark designed for Windows 10 PCs and graphics cards with Microsoft DirectX Raytracing capabilities.
No real surprises here. The RTX 3090 Gaming X Trio easily surpassed the previous “king of the hill,” the RTX 2080 Ti, in all of our synthetic benchmarks.
Gaming Benchmarks
For gaming benchmarks, we have updated our testing suite to bring more modern titles into the mix. Gone are Battlefield V, F1 2018, Far Cry 5, AOTS:e, and World of Tanks, which were replaced with Metro Exodus, Assassin’s Creed: Odyssey, F1 2020, and Far Cry: New Dawn. We kept The Division 2 and Shadow of the Tomb Raider. The games should provide a good view of the overall performance of the card. Many of these are DX12 games.
The gaming benchmark results all show the RTX 3090 easily outpacing the RTX 2080 Ti in all but Far Cry New Dawn. We’re not sure what is going on here, but one would think the RTX 3090 would score more than one FPS higher than a 2080 TI. The game was deleted and reinstalled, but the results remained consistent. We even ran these benchmarks on an AMD system with a clean OS and game installations, but the results for Far Cry New Dawn were still similarly disappointing.
AMD PCIe 4.0 Results
Since the new Ampere cards are PCIe 4.0 compliant, we wanted to test them on a PCIe 4.0 capable system, so we swapped out the CPU for a Ryzen 9 3900X at stock, and the motherboard was switched to the ROG Crosshair VII WiFi all other components remained the same as the original test system.
Title | 1080p | 1440p | 4K |
Assassins Creed: Odyssey | 99 | 90 | 70 |
The Division 2 | 172 | 148 | 86 |
F1 2020 | 228 | 211 | 139 |
Far Cry: New Dawn | 113 | 109 | 102 |
Metro: Exodus | 110 | 107 | 78 |
Shadow of the Tomb Raider | 142 | 138 | 95 |
If you compare the results above, you’ll find that PCIe 4.0 had very little to no effect. That could be related to CPU speed as the 3900X clocks slower than the i9 10900K. This is quite apparent in the 1080p results, which are considerably lower. Once we move to 4K, the CPU plays a smaller role, and the PCIe 4.0 system topped our test system in SOTR by a couple of FPS and AC:O by five FPS, which is nearly 10%. It’s difficult to compare across systems like this as there are many variables. The advantage could have been due to the extra cores and not PCIe 4.0.
2560×1440 and 4K UHD Results
Below are the higher resolution results starting with 2560×1440 and the gaining in popularity 3840×2160 (4K UHD). These resolutions prove to be a bit of a stretch for many cards, but the RTX 3090 Gaming X Trio handled them well, even 4K UHD.
Moving up in resolution to 2560×1440, the RTX handles that with ease. At 4K, we were still able to maintain over 60 FPS in all titles with Far Cry and F1 well over 100 FPS.
RTX and DLSS Testing
Below we did some testing in Metro Exodus and Shadow of the Tomb Raider with ray tracing (RTX) on at 1440p and 4K. On the right side are results with DLSS enabled, and as you can see, it did improve frame rates, but they were still below our previous results without ray tracing enabled.
Overclocking
For overclocking, we used MSI Afterburner 4.6.3 beta 2. With a bit of testing and patience, the RTX 3090 Gaming X Trio was able to increase a meager +64 on the GPU core, which yielded a boost speed of 2100 MHz. The GDDR6X memory was also overclocked +545. All overclocking was done by simply raising the power limit to its maximum, which, I assume, Nvidia has limited to 102% with the voltage and fan speeds were left on auto. This gave us a slight performance boost, but to be honest, the few FPS we gained was minimal, and the card is strong enough on its own without any overclock.
Overclocking Software
MSI’s latest version of their MSI Afterburner software is version 4.6.3 beta 2. This version includes the OC scanner and the latest features to support Ampere cards. It also comes with a revamped GUI, as seen below. MSI AB still does everything it used to, such as monitoring and overclocking; it just looks slightly different.
Temperatures and Power Use
We test power consumption by running through the game benchmarks of Shadow of the Tomb Raider and F1 2020 at stock speeds and while overclocked. We monitor temperatures throughout this testing, with the peak temperature being what is listed in the data below. To more accurately simulate real gaming conditions, the benchmarks are extended (time) to allow the card to settle.
Temperatures on the MSI RTX 3090 GXT during testing were well-controlled, peaking at 75 °C while overclocked and 73°C at stock. The fans ramped up slightly but were inaudible using the auto setting.
Power use on this 350 W card peaked at 615 W (system) overclocked and 604 W while at stock. A quality 750 W PSU will be adequate, including overclocking both CPU and GPU and still allowing for headroom and quiet operation, but this is a power-hungry graphics card. If you noticed in the overclocked GPUz screenshot above, the board power usage while overclocked came in around 420 W. Comparing it to our 2080 Ti results at 416 W system power usage in SOTR, it’s almost a full 200 W higher. That’s a significant jump in power, showing just how thirsty this GPU is.
Conclusion
Nvidia has pulled a big card out of their hat with the RTX 3090, but it is eerily reminiscent of AMD’s RDNA. They took a node shrink from 12 nm to 8 nm, which according to them, gives nearly twice the performance per watt and pushed it as far as it can go. Out of the box, they didn’t leave any headroom for overclocking, and luckily this sample didn’t require a special power adapter.
Next up, we have the price, which in all honesty, is far beyond the reach of many people for an upgrade to their gaming rig. The Nvidia FE has an MSRP of $1500 USD and our MSI is listed for $1589 at Newegg. Comparatively, this is a small premium over the MSRP for an improved PCB, power handling and cooling.
Overall, Nvidia has set the bar very high for AMD. The RTX 3090 is one heck of a GPU with exceptional performance. Even with the upcoming launch of RDNA2, I believe Nvidia can rest easy. Judging by the “leaks” we have seen, it doesn’t appear that AMD will take the performance crown this time around unless they have been holding something tight to their chest, but that seems unlikely. Aside from price and power usage, if we base this purely on the performance, we can easily give the RTX 3090 the Overclockers stamp of approval!
– Shawn Jennings (Johan45)
22 replies
Loading new replies...
Benching Team Leader
Gulper Nozzle Co-Owner
Senior Golfer
Member
Dreadnought Class Senior
Premium Member
Senior Golfer
Member
Gulper Nozzle Co-Owner
Member
Join the full discussion at the Overclockers Forums →