Keeping their tradition alive of launching a new graphics architecture every two years, this year, NVIDIA introduces its Ampere GPU. The Ampere GPU is built upon the foundation set by Turing. Termed as its biggest generational leap, the GeForce RTX 3090 Founders Edition based on NVIDIA Ampere GPU excel in previous generations at everything.
The Ampere GPU architecture has a lot to be talked about in this review, but so does the new RTX lineup. The Ampere lineup offers faster shader performance, faster ray tracing performance, and faster AI performance. Built on a brand new process node and featuring an architecture designed from the ground up, Ampere is a killer product with lots of numbers to talk about.
The fundamental of Ampere was to take everything NVIDIA learned with its Turing architecture and not only refine it but to use its DNA to form a product in a completely new performance category. Tall claims were made by NVIDIA when they introduced its Ampere lineup earlier this month & we will be finding out whether NVIDIA hit all the ticks with its Ampere architecture as this review will be your guiding path to see what makes Ampere and how it performs against its predecessors.
Today, we will be taking a look at the NVIDIA GeForce RTX 3090 Founders Edition graphics card. The card was provided by NVIDIA for the sole purpose of this review & we will be taking a look at their technology, design, and performance metrics in detail.
NVIDIA GeForce RTX 30 Series Gaming Graphics Cards - The Biggest GPU Performance Leap in Recent History
Turing wasn't just any graphics core, it was the graphics core that was to become the foundation of future GPUs. The future is realized now with next-generation consoles going deep in talks about ray tracing and AI-assisted super-sampling techniques. NVIDIA had a head start with Turing and its Ampere generation will only do things infinitely times better.
The Ampere GPU does many traditional things which we would expect from a GPU, but at the same time, also breaks the barrier when it comes to untraditional GPU operations. Just to sum up some features:
- New Streaming Multiprocessor (SM)
- New Turing Tensor Cores
- New Real-Time Ray Tracing Acceleration
- New Shading Enhancements
- New Deep Learning Features For Graphics & Inference
- New GDDR6X High-Performance Memory Subsystem
- New 2nd Generation NVLINK Interconnect
- New HDMI 2.1 Display Engine & Next-Gen NVENC/NVDEC
The technologies mentioned above are some of the main building blocks of the Ampere GPU, but there's more within the graphics core itself which we will talk about in detail so let's get started.
Let's take a trip down the journey to Ampere. In 2016, NVIDIA announced their Pascal GPUs which would soon be featured in their top to bottom GeForce 10 series lineup. After the launch of Maxwell, NVIDIA gained a lot of experience in the efficiency department which they put a focus on since their Kepler GPUs. Two years go, NVIDIA, rather than offering another standard leap in the rasterization performance of its GPUs took a different approach & introduced two key technologies in its Turing line of consumer GPUs, one being AI-assisted acceleration with the Tensor Cores and the second being hardware-level acceleration for Ray Tracing with its brand new RT cores.
With Ampere and it's brand new Samsung 8nm fabrication process, NVIDIA is adding even more to its gaming graphics lineup. Starting with the most significant part of the Ampere GPU architecture, the Ampere SM, we are seeing an entirely new graphics core. The Ampere SM features the next-gen FP32, INT32, Tensor Cores, and RT cores.
Coming to the new execution units or cores, Ampere has both INT32 and FP32 units which can execute concurrently. This new architectural design allows Turing to execute floating-point and non-floating point operations in parallel which allows for higher throughput in standard floating-point operations. According to NVIDIA, the updated Ampere graphics core delivers up to 1.7x faster traditional rasterization performance and up to 2x faster ray-tracing performance compared to the Turing GPUs.
The Ampere SM is partitioned into four processing blocks, each with 32 FP32 Cores, 16 INT32 Cores, one Tensor Core, one warp scheduler, and one dispatch unit. This is made possible with an updated datapath with one data path offering 16 FP32 execution units while the other offers either 16 FP32 or 16 INT32 execution units. This adds to 128 FP32 Cores, 64 INT 32 Cores,4 Tensor, 4 Wrap Schedulers, and 4 Dispatch Units on a single Ampere SM. Each block also includes a new L0 instruction cache and a 64 KB register file for a total of 256 KB register file per SM.
One of the key design goals for the Ampere 30-series SM was to achieve twice the throughput for FP32 operations compared to the Turing SM. To accomplish this goal, the Ampere SM includes new datapath designs for FP32 and INT32 operations. One datapath in each partition consists of 16 FP32 CUDA Cores capable of executing 16 FP32 operations per clock. Another datapath consists of both 16 FP32 CUDA Cores and 16 INT32 Cores. As a result of this new design, each Ampere SM partition is capable of executing either 32 FP32 operations per clock, or 16 FP32 and 16 INT32 operations per clock. All four SM partitions combined can execute 128 FP32 operations per clock, which is double the FP32 rate of the Turing SM, or 64 FP32 and 64 INT32 operations per clock.
Doubling the processing speed for FP32 improves performance for a number of common graphics and compute operations and algorithms. Modern shader workloads typically have a mixture of FP32 arithmetic instructions such as FFMA, floating point additions (FADD), or floating point multiplications (FMUL), combined with simpler instructions such as integer adds for addressing and fetching data, floating point compare, or min/max for processing results, etc. Performance gains will vary at the shader and application level depending on the mix of instructions. Ray tracing denoising shaders are good examples that might benefit greatly from doubling FP32 throughput.
Doubling math throughput required doubling the data paths supporting it, which is why the Ampere SM also doubled the shared memory and L1 cache performance for the SM. (128 bytes/clock per Ampere SM versus 64 bytes/clock in Turing). Total L1 bandwidth for GeForce RTX 3080 is 219 GB/sec versus 116 GB/sec for GeForce RTX 2080 Super.
Like prior NVIDIA GPUs, Ampere is composed of Graphics Processing Clusters (GPCs), Texture Processing Clusters (TPCs), Streaming Multiprocessors (SMs), Raster Operators (ROPS), and memory controllers.
The GPC is the dominant high-level hardware block with all of the key graphics processing units residing inside the GPC. Each GPC includes a dedicated Raster Engine, and now also includes two ROP partitions (each partition containing eight ROP units), which is a new feature for NVIDIA Ampere Architecture GA10x GPUs. More details on the NVIDIA Ampere architecture can be found in NVIDIA’s Ampere Architecture White Paper, which will be published in the coming days.
The four processing blocks share a combined 128 KB L1 data cache/shared memory. Traditional graphics workloads partition the 128 KB L1/shared memory as 64 KB of dedicated graphics shader RAM and 64 KB for texture cache and register file spill area. In compute mode, the GA10x SM will support the following configurations:
- 128 KB L1 + 0 KB Shared Memory
- 120 KB L1 + 8 KB Shared Memory
- 112 KB L1 + 16 KB Shared Memory
- 96 KB L1 + 32 KB Shared Memory
- 64 KB L1 + 64 KB Shared Memory
- 28 KB L1 + 100 KB Shared Memory
Ampere also ties its ROPs to the HPC and houses a total of 16 ROP units per GPC. The full GA102 GPU feature 112 ROPs while the GeForce RTX 3080 comes with a total of 96 ROPs.
The entire SM works in harmony by using different blocks to deliver high performance and better texture caching, enabling for up to twice as better CUDA core performance when compared to the previous generation.
Many of these Ampere SMs combine to form the Ampere GPU. Each TPC inside the Ampere GPU houses 2 Turing SMs which are linked to the raster engine. There are a total of 6 TPCs or 12 Ampere SM that are arranged inside the GPC or Graphics Processing Cluster. The top configured GA102 GPU comes with 7 GPCs with a total of 42 TPCs and 84 SMs that are connected to 10 MB of L1 and 6 MB of L2 cache, ROPs, TMUs, memory controllers, and NVLINK HighSpeed I/O hub. All of this combines to form the massive Ampere GA102 GPU. The following are some perf figures for the top Ampere graphics cards.
NVIDIA GeForce RTX 3090
- 35.58 TFLOPS of peak single-precision (FP32) performance
- 71.16 TFLOPS of peak half-precision (FP16) performance
- 17.79 TIPS1 concurrent with FP, through independent integer execution units
- 258 Tensor TFLOPS
- 69 RT-TFLOPs
NVIDIA GeForce RTX 3080
- 30 TFLOPS of peak single-precision (FP32) performance
- 60 TFLOPS of peak half-precision (FP16) performance
- 15 TIPS1 concurrent with FP, through independent integer execution units
- 238 Tensor TFLOPS
- 58 RT-TFLOPs
In terms of shading performance which is the direct result of the enhanced core design and GPU architecture revamp, the Ampere GPU offers an uplift of up to 70% better performance per core compared to Turing GPUs.
It should be pointed out that these are just per core performance gains at the same clock speeds without adding the benefits of other technologies that Ampere comes with. That would further increase the performance in a wide variety of gaming applications.
NVIDIA Ampere "GeForce RTX 30" GPUs Full Breakdown:
| Graphics Card | NVIDIA GeForce RTX 2070 SUPER | NVIDIA GeForce RTX 3070 | NVIDIA GeForce RTX 2080 | NVIDIA GeForce RTX 3080 | NVIDIA Titan RTX | NVIDIA GeForce RTX 3090 |
|---|---|---|---|---|---|---|
| GPU Codename | TU106 | GA104 | TU104 | GA102 | TU102 | GA102 |
| GPU Architecture | NVIDIA Turing | NVIDIA Ampere | NVIDIA Turing | NVIDIA Ampere | NVIDIA Turing | NVIDIA Ampere |
| GPCs | 5 or 6 | 6 | 6 | 6 | 6 | 7 |
| TPCs | 20 | 23 | 23 | 34 | 36 | 41 |
| SMs | 40 | 46 | 46 | 68 | 72 | 82 |
| CUDA Cores / SM | 64 | 128 | 64 | 128 | 64 | 128 |
| CUDA Cores / GPU | 2560 | 5888 | 2944 | 8704 | 4608 | 10496 |
| Tensor Cores / SM | 8 (2nd Gen) | 4 (3rd Gen) | 8 (2nd Gen) | 4 (3rd Gen) | 8 (2nd Gen) | 4 (3rd Gen) |
| Tensor Cores / GPU | 320 (2nd Gen) | 184 (3rd Gen) | 368 | 272 (3rd Gen) | 576 (2nd Gen) | 328 (3rd Gen) |
| RT Cores | 40 (1st Gen) | 46 (2nd Gen) | 46 (1st Gen) | 68 (2nd Gen) | 72 (1st Gen) | 82 (2nd Gen) |
| GPU Boost Clock (MHz) | 1770 | 1725 | 1800 | 1710 | 1770 | 1695 |
| Peak FP32 TFLOPS (non-Tensor) | 9.1 | 20.3 | 10.6 | 29.8 | 16.3 | 35.6 |
| Peak FP16 TFLOPS (non-Tensor) | 18.1 | 20.3 | 21.2 | 29.8 | 32.6 | 35.6 |
| Peak BF16 TFLOPS (non-Tensor) | NA | 20.3 | NA | 29.8 | NA | 35.6 |
| Peak INT32 TOPS (non-Tensor) | 9.1 | 10.2 | 10.6 | 14.9 | 16.3 | 17.8 |
| Peak FP16 Tensor TFLOPS with FP16 Accumulate |
72.5 | 81.3/162.6 | 84.8 | 119/238 | 130.5 | 142/284 |
| Peak FP16 Tensor TFLOPS with FP32 Accumulate |
36.3 | 40.6/81.3 | 42.4 | 59.5/119 | 65.2 | 71/142 |
| Peak BF16 Tensor TFLOPS with FP32 Accumulate |
NA | 40.6/81.3 | NA | 59.5/119 | NA | 71/142 |
| Peak TF32 Tensor TFLOPS | NA | 20.3/40.6 | NA | 29.8/59.5 | NA | 35.6/71 |
| Peak INT8 Tensor TOPS | 145 | 162.6/325.2 | 169.6 | 238/476 | 261 | 284/568 |
| Peak INT4 Tensor TOPS | 290 | 325.2/650.4 | 339.1 | 476/952 | 522 | 568/1136 |
| Frame Buffer Memory Size and Type |
8 GB GDDR6 | 8 GB GDDR6 | 8 GB GDDR6 | 10 GB GDDR6X | 24 GB GDDR6 | 24 GB GDDR6X |
| Memory Interface | 256-bit | 256-bit | 256-bit | 320-bit | 384-bit | 384-bit |
| Memory Clock (Data Rate) | 14 Gbps | 14 Gbps | 14 Gbps | 19 Gbps | 14 Gbps | 19.5 Gbps |
| Memory Bandwidth | 448 GB/sec | 448 GB/sec | 448 GB/sec | 760 GB/sec | 672 GB/sec | 936 GB/sec |
| ROPs | 64 | 96 | 64 | 96 | 96 | 112 |
| Pixel Fill-rate (Gigapixels/sec) | 113.3 | 165.6 | 115.2 | 164.2 | 169.9 | 193 |
| Texture Units | 160 | 184 | 184 | 272 | 288 | 328 |
| Texel Fill-rate (Gigatexels/sec) | 283.2 | 317.4 | 331.2 | 465 | 509.8 | 566 |
| L1 Data Cache/Shared Memory | 3840 | 5888 | 4416 KB | 8704 KB | 6912 KB | 10496 KB |
| L2 Cache Size | 4096 KB | 4096 KB | 4096 KB | 5120 KB | 6144 KB | 6144 KB |
| Register File Size | 10240 KB | 11776 KB | 11776 KB | 17408 KB | 18432 KB | 20992 KB |
| TGP (Total Graphics Power) | 215 Watts | 220W | 225W | 320W | 280W | 350W |
| Transistor Count | 13.6 Billion | 17.4 Billion | 13.6 Billion | 28.3 Billion | 18.6 Billion | 28.3 Billion |
| Die Size | 545 mm2 | 392.5 mm2 | 545 mm2 | 628.4 mm2 | 754mm2 | 628.4 mm2 |
| Manufacturing Process | TSMC 12 nm FFN (FinFET NVIDIA) |
Samsung 8 nm 8N NVIDIA Custom Process |
TSMC 12 nm FFN (FinFET NVIDIA) |
Samsung 8 nm 8N NVIDIA Custom Process |
TSMC 12 nm FFN (FinFET NVIDIA) |
Samsung 8 nm 8N NVIDIA Custom Process |
NVIDIA Ampere GPUs - GA102 & GA104 For The First Wave of Gaming Cards
NVIDIA is first introducing two brand new Ampere GPUs which include the GA102 and the GA104. The GA102 GPU is going to be featured on the GeForce RTX 3090 and GeForce RTX 3080 graphics cards while the GA104 GPU is going to be featured on the GeForce RTX 3070 graphics cards. The Ampere GPUs are based on the Samsung 8nm custom process node for NVIDIA and as such, the resultant GPU dies are slightly smaller than their Turing based predecessors but do come with a denser transistor layout. There will be several variations of each GPU featured across the RTX 30 series lineup. Following is what the complete GA102 and GA104 GPUs have to offer.
NVIDIA Ampere GA102 GPU
The full GA102 GPU is made up of 7 graphics processing clusters with 12 SM units on each cluster. That makes up 84 SM units for a total of 10752 cores in a 28.3 billion transistor package measuring 628.4mm2.
NVIDIA Ampere GA104 GPU
The full GA104 GPU is made up of 6 graphics processing clusters with 8 SM units on each cluster. That makes up 48 SM units for a total of 6144 cores in a 17.4 billion transistor package measuring 392.5mm2.
NVIDIA has also introduced its 3rd Generation Tensor core architecture and 2nd Generation RT cores on Ampere GPUs. Now Tensor cores have been available since Volta and consumers got a taste of it with the Turing GPUs. One of the key areas where Tensor Cores are put to use for AAA games is DLSS. There's a whole software stack that leverages from Tensor cores and that is known as the NVIDIA NGX. These software-based technologies will help enhance graphics fidelity with features such as Deep Learning Super Sampling (DLSS), AI InPainting, AI Super Rez, RTX Voice, and AI Slow-Mo.
While its initial debut was a bit flawed, DLSS in its 2nd iteration (DLSS 2.0) has done wonders to not only improve gaming performance but also image quality. In titles such as Death Stranding and Control, games are shown to offer higher visual fidelity than at native resolution while running at much higher framerates. With Ampere, we can expect an even higher boost in terms of DLSS 2.0 (and DLSS Next-Gen) performance as the deep-learning model continues working its magic in DLSS supported titles. NVIDIA will also be adding 8K DLSS support to its Ampere GPU lineup which would be great to test out with the 24 GB RTX 3090 graphics card.
With Ampere, Tensor cores add INT8 and INT4 precision in addition to FP16 which is still fully supported. NVIDIA has been at the helm of the deep learning revolution by supporting it since its Kepler generation of graphics cards. Today, NVIDIA has some of the most powerful AI graphics accelerators and a software stack that is widely adopted by this fast-growing industry.
For its 3rd Gen Tensor cores, NVIDIA is using the same sparsity architecture that they've used on the Ampere HPC line of GPUs. While Ampere features 4 Tensor cores per SM compared to Turing's 8 tensor cores per SM, they are not only based on the new 3rd Generation design but also get an increased count with the larger SM array. The Ampere GPU can execute 128 FP16 FMA operations per tensor core utilizing its entire INT16 cores and with sparsity, it can do up to 256. The total FP16 FMA operations per SM are increased to 512 and 1024 with sparsity. That's a 2x increase over the Turing GPU in terms of inference performance with the updated Tensor design.
2nd Gen RT Cores, RTX, and Real-Time Ray Tracing Dissected
Next up, we have the RT Cores which are what will power Real-Time Raytracing. NVIDIA isn't going to distance themselves from traditional rasterization-based rendering, but instead following a hybrid rendering model. The new 2nd Generation RT cores offer increased performance and offer double the ray/triangle intersection testing rate over Turing RT cores.
There's one RT core per SM and all of them combined accelerate Bounding Volume Hierarchy (BVH) traversal and ray/triangle intersection testing (ray casting) functions. RT Cores work together with advanced denoising filtering, a highly-efficient BVH acceleration structure developed by NVIDIA Research, and RTX compatible APIs to achieve real-time ray tracing on a single Turing GPU.
RT Cores traverse the BVH autonomously, and by accelerating traversal and ray/triangle intersection tests, they offload the SM, allowing it to handle another vertex, pixel, and compute shading work. Functions such as BVH building and refitting are handled by the driver, and ray generation and shading are managed by the application through new types of shaders.
To better understand the function of RT Cores, and what exactly they accelerate, we should first explain how ray tracing is performed on GPUs or CPUs without a dedicated hardware ray tracing engine. Essentially, the process of BVH traversal would need to be performed by shader operations and take thousands of instruction slots per ray cast to test against bounding box intersections in the BVH until finally hitting a triangle and the color at the point of intersection contributes to the final pixel color (or if no triangle is hit, the background color may be used to shade a pixel).
Ray tracing without hardware acceleration requires thousands of software instruction slots per ray to test successively smaller bounding boxes in the BVH structure until possibly hitting a triangle. It’s a computationally-intensive process making it impossible to do on GPUs in real-time without hardware-based ray tracing acceleration.
The RT Cores in Ampere can process all the BVH traversal and ray-triangle intersection testing, saving the SM from spending the thousands of instruction slots per ray, which could be an enormous amount of instructions for an entire scene. The RT Core includes two specialized units. The first unit does bounding box tests, and the second unit does ray-triangle intersection tests.
The SM only has to launch a ray probe, and the RT core does the BVH traversal and ray-triangle tests, and return a hit or no hit to the SM. Also unlike the last generation, Ampere SM can process two compute workloads simultaneously, allowing ray-tracing & graphics/compute workloads to be done concurrently.
In a visual demonstration, NVIDIA has shown how RT and Tensor cores help speed up ray tracing and shader workloads significantly. A fully ray-traced frame from Wolfenstein Youngblood was taken as an example. The last-gen RTX 2080 SUPER will take 51ms to render the frame if it does it all with its shaders (CUDA Cores). With RT cores and shaders working in tandem, the processing times are reduced to just 20ms or less than half the time. Adding in Tensor cores to help reduce the rendering time even lower to just 12ms (~83 FPS).
However, with Ampere, each standard processing block receives a huge performance uplift. With an RTX 3080, the same frame can be rendered within 37ms on the Shader cores alone, 11ms with the RT+Shader cores, and 6.7ms (150 FPS) with all three core technologies working together. That's half the time of what Turing took to render the same scene.
The Micron GDDR6X memory brings a lot of new stuff to the table. It is faster, doubles the I/O data rate, and is the first to implement PAM4 multi-level signaling in memory dies. With the Geforce RTX 3090 class products, Micron's GDDR6X memory achieves a bandwidth of up to 1 TB/s which is used to power the next-generation gaming experiences at high-fidelity resolutions such as 8K.
Micron GDDR6X graphics memory doubles input/output (I/O) performance while minimizing the cost of memory. Working with AI-innovation leader NVIDIA, Micron delivers higher bandwidth by enabling multi-level signaling in the form of four-level pulse amplitude modulation (PAM4) technology in this memory device via Micron
The new GDDR6X SGRAM:
- Doubles the data rate of SGRAM at a lower power per transaction while enabling breaking of the 1 Terabyte per second (TB/s) system memory bandwidth boundary for graphics card applications;
- Is the first discrete graphics memory device that employs PAM4 encoded signaling between the processor and the DRAM, using four voltage levels to encode and transfer two bits of data per interface clock.
- Can be designed and operated stably at high speeds, and built-in mass-production.
As mentioned, GDDR6X features the brand new PAM4 multilevel signaling techniques which helps transfer data much faster, doubles the I/O rate, pushing the capability of each memory dies from 64 GB/s to 84 GB/s. The Micron GDDR6X memory dies are also the only graphics DRAM that can be mass-produced while feature PAM4 signaling.
What is interesting is that Micron quotes that its GDDR6X memory can hit speeds of up to 21 Gbps whereas we have only got to see 19.5 Gbps in action on the GeForce RTX 3090. It is likely that AIBs could utilize higher binned dies as they are available. Micron also confirms that they plan to offer speeds higher than 21 GB/s moving in 2021 but we will have to wait and see whether any cards will utilize them.
It's not just faster speeds but Micron's GDDR6X provides higher bandwidth while sipping in 15% lower power per transferred bit compared to the previous generation GDDR6 memory. PAM4 signaling is a big upgrade from the two-level NRZ signaling on the GDDR6 memory.
Instead of transmitting two binary bits of data each clock cycle (one bit on the rising edge and one bit on the falling edge of the clock), PAM4 sends two bits each clock edge, encoded using four different voltage levels. The voltage levels are divided into 250 mV steps with each level representing two bits of data - 00, 01, 10, or 11 sent on each clock edge (still DDR technology).
Micron GDDR6X Memory
| Feature | GDDR5 | GDDR5X | GDDR6 | GDDR6X |
|---|---|---|---|---|
| Density | From 512Mb to 8Gb | 8Gb | 8Gb, 16Gb | 8Gb, 16Gb |
| VDD and VDDQ | Either 1.5V or 1.35V | 1.35V | Either 1.35V or 1.25V | Either 1.35V or 1.25V |
| VPP | N/A | 1.8V | 1.8V | 1.8V |
| Data rates | Up to 8 Gb/s | Up to 12Gb/s | Up to 16 Gb/s | 19 Gb/s, 21 Gb/s, >21 Gb/s |
| Channel count | 1 | 1 | 2 | 2 |
| Access granularity | 32 bytes | 64 bytes 2x 32 bytes in pseudo 32B mode |
2 ch x 32 bytes | 2 ch x 32 bytes |
| Burst length | 8 | 16 / 8 | 16 | 8 in PAM4 mode 16 in RDQS mode |
| Signaling | POD15/POD135 | POD135 | POD135/POD125 | PAM4 POD135/POD125 |
| Package | BGA-170 14mm x 12mm 0.8mm ball pitch |
BGA-190 14mm x 12mm 0.65mm ball pitch |
BGA-180 14mm x 12mm 0.75mm ball pitch |
BGA-180 14mm x 12mm 0.75mm ball pitch |
| I/O width | x32/x16 | x32/x16 | 2 ch x16/x8 | 2 ch x16/x8 |
| Signal count | 61 - 40 DQ, DBI, EDC - 15 CA - 6 CK, WCK |
61 - 40 DQ, DBI, EDC - 15 CA - 6 CK, WCK |
70 or 74 - 40 DQ, DBI, EDC - 24 CA - 6 or 10 CK, WCK |
70 or 74 - 40 DQ, DBI, EDC - 24 CA - 6 or 10 CK, WCK |
| PLL, DCC | PLL | PLL | PLL, DCC | DCC |
| CRC | CRC-8 | CRC-8 | 2x CRC-8 | 2x CRC-8 |
| VREFD | External or internal per 2 bytes | Internal per byte | Internal per pin | Internal per pin 3 sub-receivers per pin |
| Equalization | N/A | RX/TX | RX/TX | RX/TX |
| VREFC | External | External or Internal | External or Internal | External or Internal |
| Self refresh (SRF) | Yes Temp. Controlled SRF |
Yes Temp. Controlled SRF Hibernate SRF |
Yes Temp. Controlled SRF Hibernate SRF VDDQ-off |
Yes Temp. Controlled SRF Hibernate SRF VDDQ-off |
| Scan | SEN | IEEE 1149.1 (JTAG) | IEEE 1149.1 (JTAG) | IEEE 1149.1 (JTAG) |
With each new generation of graphics cards, NVIDIA delivers a new range of display technologies. This generation is no different and we see some significant updates to not only the display engine but also the graphics interconnect. With the adoption of faster GDDR6X memory which provides higher bandwidth, faster compression, and more cache, Gaming applications can now run at higher resolutions, supporting more details on the display.
The Ampere Display Engine supports two new display technologies, HDMI 2.1 and DisplayPort 1.4a with DSC 1.2a. HDMI 2.1 allows for up to 48 Gbps of total bandwidth and allows for up to 4K 240Hz HDR and 8K 60Hz HDR.
DisplayPort 1.4a allows for up to 8K resolutions with 60Hz refresh rates and includes VESA's display stream compression 1.2 technology with visually lossless compression. You can run up to two 8K displays at 60 Hz using two cables, one for each display. In addition to that, Ampere also supports HDR processing natively with tone mapping added to the HDR pipeline.
Ampere GPUs also ships with the Fifth Generation NVDEC decoder unit that adds AV1 hardware decode support. Ampere's new NVDEC decoder has also been updated to support the decoding of MPEG-2, VC-1, H.264 (AVCHD), H.265 (HEVC), VP8, VP9, and AV1.
Ampere also adds the 7th Generation NVENC encoder by offering seamless hardware-accelerated encoding of up to 4K on H.264 and 8K on HEVC.
NVIDIA RTX IO - Blazing Fast Read Speeds With GPU Utilization
As storage sizes have grown, so has storage performance. Gamers are increasingly turning to SSDs to reduce game load times: while hard drives are limited to 50-100 MB/sec throughput, the latest M.2 PCIe Gen4 SSDs deliver up to 7 GB/sec. With the traditional storage model, game data is read from the hard disk, then passed from the system memory and CPU before being passed to the GPU.
Historically games have read files from the hard disk, using the CPU to decompress the game image. Developers have used lossless compression to reduce install sizes and to improve I/O performance. However, as storage performance has increased, traditional file systems and storage APIs have become a bottleneck. For example, decompressing game data from a 100 MB/sec hard drive takes only a few CPU cores, but decompressing data from a 7 GB/sec PCIe Gen4 SSD can consume more than twenty AMD Ryzen Threadripper 3960X CPU cores!
Using the traditional storage model, game decompression can consume all 24 cores on a Threadripper CPU. Modern game engines have exceeded the capability of traditional storage APIs. A new generation of I/O architecture is needed. Data transfer rates are the gray bars, CPU cores required are the black/blue blocks.
NVIDIA RTX IO is a suite of technologies that enable rapid GPU-based loading and decompression of game assets, accelerating I/O performance by up to 100x compared to hard drives and traditional storage APIs. When used with Microsoft’s new DirectStorage for Windows API, RTX IO offloads dozens of CPU cores’ worth of work to your RTX GPU, improving frame rates, enabling near-instantaneous game loading, and opening the door to a new era of large, incredibly detailed open-world games.
Object pop-in and stutter can be reduced, and high-quality textures can be streamed at incredible rates, so even if you’re speeding through a world, everything runs and looks great. In addition, with lossless compression, game download and install sizes can be reduced, allowing gamers to store more games on their SSD while also improving their performance.
NVIDIA RTX IO plugs into Microsoft’s upcoming DirectStorage API which is a next-generation storage architecture designed specifically for state-of-the-art NVMe SSD-equipped gaming PCs and the complex workloads that modern games require. Together, streamlined and parallelized APIs specifically tailored for games allow dramatically reduced IO overhead and maximize performance/bandwidth from NVMe SSDs to your RTX IO-enabled GPU.
Specifically, NVIDIA RTX IO brings GPU-based lossless decompression, allowing reads through DirectStorage to remain compressed and delivered to the GPU for decompression. This removes the load from the CPU, moving the data from storage to the GPU in a more efficient, compressed form, and improving I/O performance by a factor of two.
GeForce RTX GPUs will deliver decompression performance beyond the limits of even Gen4 SSDs, offloading potentially dozens of CPU cores’ worth of work to ensure maximum overall system performance for next-generation games. Lossless decompression is implemented with high performance compute kernels, asynchronously scheduled. This functionality leverages the DMA and copy engines of Turing and Ampere, as well as the advanced instruction set, and architecture of these GPU’s SM’s.
The advantage of this is that the enormous compute power of the GPU can be leveraged for burst or bulk loading (at level load for example) when GPU resources can be leveraged as high performance I/O processor, delivering decompression performance well beyond the limits of Gen4 NVMe. During streaming scenarios, bandwidths are a tiny fraction of the GPU capability, further leveraging the advanced asynchronous compute capabilities of Turing and Ampere. Microsoft is targeting a developer preview of DirectStorage for Windows for game developers next year, and NVIDIA Turing & Ampere gamers will be able to take advantage of RTX IO enhanced games as soon as they become available.
NVLINK For GeForce RTX 3090 And Titan Class Products Only!
NVIDIA has said farewell to their SLI (Scale Link Interface) interconnect for consumer graphics cards. They will now be using the NVLINK interconnect which has already been featured on their Turing GPUs. The reason is that SLI was simply not enough to feed higher bandwidth to Ampere GPUs.
A single x8 NVLINK channel provides 25 GB/s peak bandwidth. There are 4 x4 links on the GA102 GPU. The GA102 GPU features 50 GB/s of bandwidth in parallel and 100 GB/s bandwidth bi-directionally. Using NVLINK on high-end cards would be beneficial in high-resolution gaming but there's a reason NVIDIA still restricts users from doing 3 and 4 way SLI.
Multi-GPU still isn't optimized so you won't see many benefits unless you are running the highest-end graphics cards. That's another reason why the RTX 3080 & RTX 3070 are deprived of NVLINK connectors. The NVLINK connectors cost $79 US each and are sold separately.
The NVIDIA GeForce RTX 3090 is a force to be reckoned with. It is the ultimate gaming GPU that is surprisingly much faster than the GeForce RTX 2080 Ti and Titan RTX which are its Turing based predecessor but is also coming in a bit higher than the RTX 2080Ti but much lower than the Titan RTX in pricing. The GeForce RTX 3090 carries more cores, more memory (same as Titan RTX), higher performance efficiency, and also carries next-generation ray-tracing and tensor cores that make this a truly next-generation graphics card.
NVIDIA designed the GeForce RTX 3090 not just for any gamer but the professional creator who also wants to have the best graphics performance at hand to power the next-generation of AAA gaming titles with superb visuals and insane fluidity. It's not just the FPS that matters these days, its visuals, and a smoother frame rate too and this is exactly what the GeForce RTX 30 series is made to excel at. There's a lot to talk about regarding NVIDIA's flagship Ampere gaming graphics cards so let's start off with the specifications.
Marvels of NVIDIA Ampere Architecture - 2nd Generation RTX
Enabling the blistering performance of the new RTX 30 Series GPUs and the NVIDIA Ampere architecture are cutting-edge technologies and over two decades of graphics R&D, including:
- New streaming multiprocessors: The building block for the world’s fastest, most efficient GPU, delivering 2x the FP32 throughput of the previous generation, and 30 Shader-TFLOPS of processing power.
- Second-gen RT Cores: New dedicated RT Cores deliver 2x the throughput of the previous generation, plus concurrent ray tracing and shading and compute, with 58 RT-TFLOPS of processing power.
- Third-gen Tensor Cores: New dedicated Tensor Cores, with up to 2x the throughput of the previous generation, making it faster and more efficient to run AI-powered technologies, like NVIDIA DLSS, and 238 Tensor-TFLOPS of processing power.
- NVIDIA RTX IO: Enables rapid GPU-based loading and game asset decompression, accelerating input/output performance by up to 100x compared with hard drives and traditional storage APIs. In conjunction with Microsoft’s new DirectStorage for Windows API, RTX IO offloads dozens of CPU cores’ worth of work to the RTX GPU, improving frame rates and enabling near-instantaneous game loading.
- World’s fastest graphics memory: NVIDIA has worked with Micron to create the world’s fastest discrete graphics memory for the RTX 30 Series, GDDR6X. It provides data speeds of close to 1TB/s system memory bandwidth for graphics card applications, maximizing game and app performance.
- Next-gen process technology: New 8N NVIDIA custom process from Samsung, which allows for higher transistor density and more efficiency.
NVIDIA GeForce RTX 3080 Graphics Card Specifications - GA102 GPU & 10 GB GDDR6X Memory
At the heart of the NVIDIA GeForce RTX 3090 graphics card lies the GA102 GPU. The GA102 is one of the many Ampere GPUs that we will be getting on the gaming segment. The GA102 GPU is the fastest gaming GPU that NVIDIA has produced. The GPU is based on Samsung's 8nm custom process node designed specifically for NVIDIA and features a total of 28 Billion transistors. It measures at 628mm2 which makes it the 2nd biggest gaming GPU ever produced right below the Turing TU102 GPU.
The new shader core on the NVIDIA Ampere architecture is 2.7x faster, the new RT cores are 1.7x faster while the new Tensor cores are up to 2.7x faster than the previous generation Turing GPUs. The 2nd Generation RT core delivers dedicated hardware-accelerated ray-tracing performance & features twice the ray/triangles intersection with concurrent RT graphics and compute operations.
For the GeForce RTX 3090, NVIDIA has enabled a total of 72 SM units on its flagship which results in a total of 10,496 CUDA cores. In addition to the CUDA cores, NVIDIA's GeForce RTX 3080 also comes packed with next-generation RT (Ray-Tracing) cores, Tensor cores, and brand new SM or streaming multi-processor units. The GPU runs at a base clock speed of 1395 MHz and a boost clock speed of 1695 MHz. The card has a TDP of 350W.
In terms of memory, the GeForce RTX 3090 comes packed with 240 GB of memory and that too the next-generation GDDR6X design. With Micron's latest and greatest graphics memory dies, the RTX 3090 can deliver GDDR6X memory speeds of 19.5 Gbps. That along with a bus interface of 384-bit will deliver a cumulative bandwidth of 936 Gbps.
NVIDIA GeForce RTX 30 Series 'Ampere' Graphics Card Specifications:
| Graphics Card Name | NVIDIA GeForce RTX 3060 Ti | NVIDIA GeForce RTX 3070 | NVIDIA GeForce RTX 3080 | NVIDIA GeForce RTX 3090 |
|---|---|---|---|---|
| GPU Name | Ampere GA104-200 | Ampere GA104-300 | Ampere GA102-200 | Ampere GA102-300 |
| Process Node | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm |
| Die Size | 395.2mm2 | 395.2mm2 | 628.4mm2 | 628.4mm2 |
| Transistors | 17.4 Billion | 17.4 Billion | 28 Billion | 28 Billion |
| CUDA Cores | 4864 | 5888 | 8704 | 10496 |
| TMUs / ROPs | TBA | TBA | 272 / 96 | TBA |
| Tensor / RT Cores | 152 / 38 | 184 / 46 | 272 / 68 | 328 / 82 |
| Base Clock | TBA | 1500 MHz | 1440 MHz | 1400 MHz |
| Boost Clock | TBA | 1730 MHz | 1710 MHz | 1700 MHz |
| FP32 Compute | TBA | 20 TFLOPs | 30 TFLOPs | 36 TFLOPs |
| RT TFLOPs | TBA | 40 TFLOPs | 58 TFLOPs | 69 TFLOPs |
| Tensor-TOPs | TBA | 163 TOPs | 238 TOPs | 285 TOPs |
| Memory Capacity | 8 GB GDDR6 | 8/16 GB GDDR6 | 10/20 GB GDDR6X | 24 GB GDDR6X |
| Memory Bus | 256-bit | 256-bit | 320-bit | 384-bit |
| Memory Speed | 14 Gbps | 14 Gbps | 19 Gbps | 19.5 Gbps |
| Bandwidth | 448 Gbps | 448 Gbps | 760 Gbps | 936 Gbps |
| TDP | 180W? | 220W | 320W | 350W |
| Price (MSRP / FE) | $399 US? | $499 US | $699 US | $1499 US |
| Launch (Availability) | October 2020 | 15th October | 17th September | 24th September |
NVIDIA GeForce RTX 3090 Graphics Card Cooling & Design- Next-Gen NVTTM Founders Edition Design
NVIDIA has developed one of their best and most powerful Founders Edition cooling design to date for the GeForce RTX 30 series graphics cards. NVIDIA explained that higher performance requires a new form of cooling solution and as such, it has prepared a unique cooling solution for its next-gen cards which will keep GPUs running cool while staying quiet by utilizing several new & existing tech.
The Founders Edition cooling makes use of a full aluminum alloy heatsink which makes use of a hybrid vapor chamber with dual-sided axial-tech based fans. The cooler heatsink is coated with a nano-carbon coating and should do a really good job at keeping the temperatures in control.
The design is interesting in the sense that not only does it goes all out with a fin and heat pipe design. This is the first design of its kind since the original Founders Edition GeForce GTX 780 that makes use of a much larger heatsink area.
It also comes with a unique fan placement, one on the front and one at the bottom. This push & pull fan configuration which as it is referred to is said to push heat out of the exhaust vents much more effectively. There will be some air that will be blown out inside the case from the back of the card itself but that shouldn't be a major cause of concern as modern CPU Air or Liquid coolers do a really good job venting out air from within the case.
Acoustically, the new Founders Edition design is quieter than traditional dual axial coolers, while still delivering nearly 2x the cooling performance of previous-generation solutions. The aforementioned NVLink and power design changes help here, creating more space for airflow through the largest fin stack seen to date, and the larger bracket vents improve airflow in concert with individually shaped shroud fins. In fact, wherever you look, every aspect of the Founders Edition cards are designed to maximize airflow, minimize temperatures, and enable the highest levels of performance with the least possible noise.
NVIDIA GeForce RTX 3090 Graphics Card PCB & 12-Pin Power Input
One of the biggest changes on the Founders Edition GeForce RTX 30 series graphics cards is the PCB design. The GeForce RTX 3090 & GeForce RTX 3080 comes with a unique & compact PCB package that is unlike anything we've seen in the consumer space before. But being compact doesn't mean that the cards don't pack a punch. There's some serious horsepower on these compact PCBs that NVIDIA has designed.
The PCB features over 18 power chokes which put it is a more premium design than the flagship non-reference RTX 20 series cards. The GeForce RTX 3080 is powered by an 18 phase design that is insane and designed to be overclocked with unprecedented GPU overclock headroom that most users can leverage from to gain even faster performance.
In addition to that, GeForce RTX 30 series Founders Edition cards will be featuring the 12-pin Micro-Fit 3.0 power connectors. These connectors don't require a power supply upgrade as the cards will ship with bundled 2x 8-pin to 1x 12-pin connectors so you can run your latest graphics card without any compatibility issues.
The placement of the 12-pin connector on the PCB is also noteworthy. It is placed in a vertical position and judging by the PCB design, we can tell why NVIDIA moved to a single 12-pin plug instead of the standard dual 8-pin design. There's limited room on the PCB to do stuff and as such, it was necessary to go for a more small and compact power input.
NVIDIA GeForce RTX 3080 Graphics Card Price & Availability - Both Custom & Reference Designs at Launch
The NVIDIA GeForce RTX 3080 is being announced today and will be launching to consumers on the 17th of September, 2020. The first wave of graphics cards to hit the market would be the reference Founders Edition variant which will cost $699 US. The NVIDIA GeForce RTX 3080 will feature a price of $699 (MSRP) however custom models will vary depending on their design and the extra horse-power that they have to offer.
There aren't any performance numbers that NVIDIA is sharing right now but from what has been showcased, the GeForce RTX 3070 is faster than an RTX 2080 Ti, the RTX 3080 is a good bit ahead of the RTX 2080 Ti and the RTX 3090 is about as much as 50% faster than the RTX 2080 Ti which is very impressive for the full lineup stack. The NVIDIA GeForce RTX 3080 itself is twice as fast as the RTX 2080 and is considerably faster than the RTX 2080 Ti making it a perfect 60 FPS 4K gaming graphics card.
NVIDIA has developed one of their best and most powerful Founders Edition cooling design to date for the GeForce RTX 30 series graphics cards. NVIDIA explained that higher performance requires a new form of cooling solution and as such, it has prepared a unique cooling solution for its next-gen cards which will keep GPUs running cool while staying quiet by utilizing several new & existing tech.
The Founders Edition cooling makes use of a full aluminum alloy heatsink which makes use of a hybrid vapor chamber with dual-sided axial-tech based fans. The cooler heatsink is coated with a nano-carbon coating and should do a really good job at keeping the temperatures in control.
The design is interesting in the sense that not only does it goes all out with a fin and heat pipe design. This is the first design of its kind since the original Founders Edition GeForce GTX 780 that makes use of a much larger heatsink area.
It also comes with a unique fan placement, one on the front and one at the bottom. This push & pull fan configuration which as it is referred to is said to push heat out of the exhaust vents much more effectively. There will be some air that will be blown out inside the case from the back of the card itself but that shouldn't be a major cause of concern as modern CPU Air or Liquid coolers do a really good job venting out air from within the case.
Acoustically, the new Founders Edition design is quieter than traditional dual axial coolers, while still delivering nearly 2x the cooling performance of previous-generation solutions. The aforementioned NVLink and power design changes help here, creating more space for airflow through the largest fin stack seen to date, and the larger bracket vents improve airflow in concert with individually shaped shroud fins. In fact, wherever you look, every aspect of the Founders Edition cards are designed to maximize airflow, minimize temperatures, and enable the highest levels of performance with the least possible noise.
In terms of cooler noise and performance, the GeForce RTX 3080 operates at a peak temperature of 78C when hitting its peak TBP of 320W with a noise output of just 30dBA. For comparison, the Turing Founders Edition coolers peak out at 81C with a noise output of 32dBA when hitting their TBP of 240W (RTX 2080 SUPER). In NVIDIA's own testing, they reveal that the GeForce RTX 3080 averages at around 1920 MHz with a GPU power draw of 310W and a peak temperature of 76C.
This is also where NVIDIA gets its 1.9x efficiency figure from as the RTX 3080 can deliver over 100 FPS while being cooler and quiet versus the 60 FPS of its Turing gen predecessor.
NVIDIA GeForce RTX 3090, RTX 3080, RTX 3070 Founders Edition Gallery:
NVIDIA GeForce RTX 3090 & RTX 3080 Graphics Card PCB & Power - Designed To Be Overclocked!
One of the biggest changes on the Founders Edition GeForce RTX 3090 graphics cards is the PCB design. The GeForce RTX 3090 & GeForce RTX 3080 comes with a unique & compact PCB package that is unlike anything we've seen in the consumer space before. But being compact doesn't mean that the cards don't pack a punch. There's some serious horsepower on these compact PCBs that NVIDIA has designed.
The PCB features over 20 power chokes which put it is a more premium design than the flagship non-reference RTX 20 series cards. The GPU is powered by 18 phases while the memory receives power from 2 phases. NVIDIA touts this PCB as an overclocking marvel with unprecedented GPU overclock headroom that most users can leverage from to gain even faster performance. But as pointed out earlier by us, the Founders Edition PCB is not the reference design and that will come with a standard rectangular PCB. Water block manufacturers have also confirmed this which we reported here.
In addition to that, GeForce RTX 30 series Founders Edition cards will be featuring the 12-pin Micro-Fit 3.0 power connectors. These connectors don't require a power supply upgrade as the cards will ship with bundled 2x 8-pin to 1x 12-pin connectors so you can run your latest graphics card without any compatibility issues.
The placement of the 12-pin connector on the PCB is also noteworthy. It is placed in a vertical position and judging by the PCB design, we can tell why NVIDIA moved to a single 12-pin plug instead of the standard dual 8-pin design. There's limited room on the PCB to do stuff and as such, it was necessary to go for a more small and compact power input.
NVIDIA GeForce RTX 3090 Founders Edition Box and presentation
The packaging of the GeForce RTX 3090 is the same as the GeForce RTX 3080, but bigger to accommodate the 3-slot wide load monster that is the GeForce RTX 3090 Founders Edition.
The bottom side of the card retains the same styling as the RTX 3080 but again, much bigger. The infinity surround makes a much more pronounced X pattern this time around and the fans are significantly larger.
The top (or backplate) area of the card brings the RTX 3090 branding into full view and looks strikingly similar to the other side. Again at the rear (front of the case) we see the return of the reverse swept fan that will pull air through the heatsink. And from this angle, you can see the cutout for the only card in the Ampere gaming lineup to feature NVLink support.
To get a real sense of scale we lined up the GeForce RTX 3080 Founders edition with the GeForce RTX 3090. All of these images are the cards at the same distance from the camera to try and get a sense of just how much bigger the RTX 3090 Founders Edition really is, it's massive folks.
How about installed into a case that isn't exactly compact?
We used the following test system for comparison between the different graphics cards. The latest drivers that were available at the time of testing were used from AMD and NVIDIA on an updated version of Windows 10. All games that were tested were patched to the latest version for better performance optimization for NVIDIA and AMD GPUs.
Test System
| Components | X570 |
|---|---|
| CPU | Ryzen 9 3900X 4.3GHz All Core Lock (disable one CCD for 3600X Results) |
| Memory | 32GB Hyper X Predator DDR4 3600 |
| Motherboard | ASUS TUF Gaming X570 Plus-WiFi |
| Storage | TeamGroup Cardea 1TB NVMe PCIe 4.0 |
| PSU | Cooler Master V1200 Platinum |
| Windows Version | Latest verion of windows at the time of testing |
| Hardware-Accelerated GPU Scheduling | On if supported by GPU and driver. |
Graphics Cards Tested
| GPU | Architecture | Core Count |
Clock Speed | Memory Capacity |
Memory Speed |
|---|---|---|---|---|---|
| NVIDIA RTX 3080 FE | Ampere | 8704 | 1440/1710 | 10GB GDDR6X | 19Gbps |
| NVIDIA RTX 2080ti FE | Turing | 4352 | 1350/1635 | 11GB GDDR6 | 14Gbps |
| NVIDIA RTX 2080 SUPER FE | Turing | 3072 | 1650/1815 | 8GB GDDR6 | 15.5Gbps |
| NVIDIA GTX 1080 FE | Pascal |
2560 | 1607/1733 | 8GB GDDR5X | 10Gbps |
| AMD Radeon RX 5700XT | Navi 10 | 2560 | 1605/1755/1905 | 8GB GDDR6 | 14Gbps |
Drivers Used
| Drivers | |
|---|---|
| Radeon Settings | 20.8.3 |
| GeForce | 456.16 |
- All games were tested at 3840x2160 (4K) resolution.
- Image Quality and graphics configurations are provided with each game description.
- The "reference" cards are the stock configs.
Firestrike
Firestrike is running the DX11 API and is still a good measure of GPU scaling performance, in this test we ran the Extreme and Ultra versions of Firestrike which runs at 1440p and 4K and we recorded the Graphics Score only since the Physics and combined are not pertinent to this review.
Time Spy
Time Spy is running the DX12 API and we used it in the same manner as Firestrike Extreme where we only recorded the Graphics Score as the Physics score is recording the CPU performance and isn't important to the testing we are doing here.
Port Royal
Port Royal is another great tool in the 3DMark suite, but this one is 100% targeting Ray Tracing performance. It loads up ray traced shadows, reflections, and global illumination to really tax the performance of the graphics cards that either have hardware-based or software-based ray tracing support.
Thermals
Thermals were measured from our open test bench after running the Time Spy graphics test 2 on loop for 30 minutes recording the highest temperatures reported. The room was climate controlled and kept at a constant 22c throughout the testing.
Forza Horizon 4
Forza Horizon 4 carries on the open-world racing tradition of the Horizon series. The latest DX12 powered entry is beautifully crafted and amazingly well executed and is a great showcase of DX12 games. We use the benchmark run while having all of the settings set to non-dynamic with an uncapped framerate to gather these results.
Shadow of the Tomb Raider
Shadow of the Tomb Raider, unlike its predecessor, does a good job putting DX12 to use and results in higher performance than the DX11 counterpart in this title and because of that, we test this title in DX12. I do use the second segment of the benchmark run to gather these numbers as it is more indicative of in-game scenarios where the foliage is heavy.
Rainbow 6 Siege
Rainbow 6 Siege has maintained a massive following since its launch and it consistently in Steams Top Ten highest player count game. In a title where the higher the framerate the better in a tactical yet fast-paced competitive landscape is essential, we include this title despite its ludicrously high framerates. We use the Vulkan Ultra preset with the High Defenition Texture Pack as well and gather our results from the built-in benchmarking tool.
DOOM Eternal
DOOM Eternal brings hell to earth with the Vulkan powered idTech 7. We test this game using the Ultra Nightmare Preset and follow our in-game benchmarking to stay as consistent as possible.
Gears Tactics
Gears Tactics is the latest in the Gears franchise and takes things in a completely different direction with the gameplay design. It is built on a DX12 based Unreal Engine 4 build. We used the Maximum settings allowed but refrained from enabling Variable Rate Shading as all cards are not capable of supporting this feature.
Ghost Recon Breakpoint
Ghost Recon Breakpoint is powered by the latest iteration of the Anvil Next 2.0 game engine. This is the same engine that was used in Assassin's Creed Odyssey but in Breakpoint has been updated to support the Vulkan API. We performed our tests using the High Preset with the Vulkan API.
Call of Duty Modern Warfare (2019)
Call of Duty Modern Warfare is back and this time on a new engine running DX12 to allow for some sick DXR Ray Traced Shadows, those results are in the RT section We tested in the 'Fog of War' mission. At 4K we set the settings all to High.
Resident Evil 3
The Resident Evil 3 Remake has surpassed the RE2 Remake in visuals and is the latest use of the RE Engine. While it does have DX12 support the DX11 implementation is far superior and because of that, we will be sticking to DX11 for this title. We use the cutscene where Jill and Carlos enter the subway car for the first time and a 2 minute capture at that point.
Borderlands 3
Borderlands 3 has made its way into the test lineup thanks to strong demand by gamers and simply delivering MORE Borderlands. This game is rather intensive after the Medium preset but since we're testing the 'Ultimate UW 1440p' card, High it is. We tested using the built-in benchmark utility
Total War Saga: Troy
Total War Saga: Troy is powered by their TW Engine 3 (Total War Engine 3) and in this iteration, they have stuck to a strictly DX11 release. We tested the game using the built-in benchmark using the Dynasty model that represents a battle with many soldiers interacting at once and is more representative of normal gameplay.
Shadow of the Tomb Raider
Shadow of the Tomb Raider, unlike its predecessor, does a good job putting DX12 to use and results in higher performance than the DX11 counterpart in this title, and because of that, we test this title in DX12. I do use the second segment of the benchmark run to gather these numbers as it is more indicative of in-game scenarios where the foliage is heavy. SotTR features Ray Traced Shadows as well as DLSS and we used both in the benchmarks with the game set to the 'Highest' preset and RT Shadows at Ultra with DLSS enabled.
Modern Warfare
Call of Duty Modern Warfare is back and this time on a new engine running DX12 to allow for some sick DXR Ray Traced Shadows. We tested in the 'Fog of War' mission where we tested our RT performance run. At 4K we set the settings all to High with ray-traced shadows enabled.
Control
Control is powered by Remedy's Northlight Storytelling Engine but severely pumped up to support multiple functions of ray-traced effects. We ran this through our test run in the cafeteria with all ray tracing functions on high and the game set to high. DLSS was enabled for this title in the quality setting.
Battlefield V
Battlefield V was one of the earlier games in the RTX 20 Series lifecycles to receive a DXR update. Battlefield V was tested on the opening sequence of the Tirailleur war story as it's been consistently one of the more demanding scenes for ray-traced reflections that are featured in this game. DLSS was enabled for this game.
Metro Exodus
Metro Exodus was the third entry into the Metro series and as Artym ventures away from the Metro he, and you, are able to explore the world with impressive RT Global Illumination. RTGI has proven to be quite an intense feature to run. Metro Exodus also supports DLSS so it was used in our testing. Advanced PhysX was left disabled, but Hairworks was left on.
Minecraft
Minecraft, yup Minecraft. When it comes to ray tracing Minecraft has it all. The Minecraft with RTX update has recently been updated to DXR1.1 so it gets the latest treatment in that regard. But, we're talking a fully path traced version of Minecraft here. We set up a run in the RTX world of Crystal Palace and set the Chunks to the maximum of 24, up from the default 8 in order to really turn the wrenches. DLSS was enabled for this game.
Quake 2 RTX
Quake II RTX is much like Minecraft with RTX in the sense that it is fully path-traced, so no rasterization here. This one however doesn't support DLSS so you're going to have to brute force it to acceptable framerates. Thankfully if these numbers don't do it for you then you can always adjust the resolution slider and enjoy a healthy performance boost.
Boundary
Boundary is a multiplayer tactical shooter...in space. It's not out yet so treat this one as more of a synthetic benchmark as there are likely to be quite a few improvements but for now, we had access to the benchmark and it's a doozy to run. Featuring full raytracing effects for the benchmark as well as DLSS, we ran that in Quality mode.
Bright Memory
Bright Memory is an action shooter that is currently in early access on Steam, will later be called Bright Memory Infinite when it fully releases. A one-man team has turned this game into a showstopper and now it features RT reflections as well as DLSS. We ran it at the High preset with DLSS set to Balanced for our testing.
Amid Evil
Amid Evil is a high energy old school shotoer that seems like an unlikely recipient of RT features, but here we are with insane DXR support in a modern retro shooter. Feature RT Reflections, RT Shadows, and NVIDIA's DLSS support we had to put this one through the rounds and see how things went. The RTX version of this game is still in beta but publicly available for those who want to try it. We tested with all RT features on and DLSS enabled.
Death Stranding
Sam Porter Bridges has delivered one of PS4's most anticipated games to the PC community and opened a whole new wold of possibilities. This was the first game to feature the Decima Engine on PC and unarguably did it the best. Death Stranding may not feature ray tracing effects but it does showcase that DLSS can be used effectively even when RT isn't around. We tested this one just like we did in our launch coverage with DLSS enabled.
V-Ray Next
Chaos Group's V-RAY is a commercial plugin used across various 3D Modeling suites from Maya to Cinema 4D allowing for path tracing, photon mapping, irradiance maps, and directly computed global illumination. The plugin is used from video game creation all the way to film and industrial design.
Thankfully they have their own benchmark utility so you don't have to go into 10+ different suites to find out how well CPUs or graphics cards perform. We used just the GPU portion for our results and according to Chaos Group, there should be a linear performance improvement based on the power and scaling of the GPU.
OctaneRender
OctaneRender is touted as the world's first and fastest unbiased spectrally correct GPU render engine and is RTX accelerated to bring 2-5x faster render speeds to NVIDIA's raytracing GPU. So we put the OctaneBench to test here with and without RTX Acceleration and then followed up with an actual workload (provided by NVIDIA) to test out how everything response outside of the benchmark and in a heavily loaded scene.
OctaneBench
In the actual OctaneRender scene we measured the time to completion as well as memory allocation. We set the Out-of-System memory up to 8GB from the default 4GB and when running the final render we can measure how much of the GPUs memory is being used as well as system resource overflow. In the render time results the 0 time for the RTX 2080 and the GTX 1080Ti are results from a render failure, so they were removed from the memory results.
OctaneRender
As a note here, the 0 measurement indicates that all memory was able to be handled on the GPU itself without having to offload to system memory during the rendered scene.
Davinci Resolve
Black Magic Davinci Resolve is a great place to work with RED 8K Raw footage, but it is not the easiest thing in the world to work with. If you're looking for a performance chart on here you're not really going to find one since it's either a pass/fail and experiential workflow experience. We were provided an 8K Raw video sample and played around with it to find that cards outside of the Titan RTX and the RTX 3090 the experience was quite subpar resulting in much lower playback speeds and constant crashes. Being able to playback the color corrected and modified 8K Raw footage made for a much smoother workflow. And if you're the kind of content creator that would be utilizing that kind of equipment I could easily see why you would be in the market for a 24GB capable graphics card.
Blender Render
Blender is a household name at this point for those in the 3D modeling and rendering scene and is used by many aspiring creators. They updated it last year with experimental an ongoing support for NVIDIA OptiX. Optix is an application framework for boosting ray tracing performance on the NVIDIA RTX lineup of cards. There is also an Optix AI denoiser as well so in our testing of the provided scene we did a pass using the OptiX enhancements vs the traditional CUDA based rendering with Open Image Denoiser applied. This scene did cause issues for lower VRAM cards and is signified by a 0 performance metric showing it did not finish. This is another instance where we can see that the raw performance of the RTX 3080 shown in synthetics and gaming vs the RTX 3090 just don't hold up when there isn't enough VRAM for these professional applications.
Graphics cards and power draw have always been quite synonymous with each other in terms of how much performance they put out for the power they take in. Measuring this has not always been the most straight forward when it comes to accuracy and methods for reviewers and end-users. NVIDIA has developed their PCAT system, or Power Capture Analysis Tool in order to be able to capture direct power consumption from ALL graphics cards that plug into the PCIe slot so that you can get a very clear barometer on actual power usage without relying on hacked together methods
The Old Way
The old method, for most anyway, was to simply use something along the lines of a Kill-A-Watt wall meter for power capture. This isn't the worst way, but as stated in our reviews it doesn't quite capture the amount of power that the graphics card alone is using. This results in some mental gymnastics to figure out how much the graphics card is using by figuring the system idle, CPU load, and the GPU load and estimating about where the graphics card lands, not very accurate to say the least.
Another way is to use GPU-z. This is the least reliable method as you have to rely entirely on the software reading from the graphics card. This is a poor method as the graphics cards vary in how they report to software when it comes to power usage. Some will only send out what the GPU core itself is using and not consider what the memory is drawing or any other component.
The last way I'll mention is the use of a multi-meter amperage clamp across the PCIe slot by way of a riser cable with separate cables then more power clamps on all the PCIe power cables going into the graphics card. This method is very accurate for graphics card power but is also very cumbersome and typically results in you having to watch the numbers and document them as you see them rather than plotting them across a spreadsheet.
The PCAT Way
This is where PCAT (power capture analysis tool) comes into play. NVIDIA has developed quite a robust tool for measuring graphics card power at the hardware level and taking the guesswork out of the equation. The tool is quite simple to set up and get going, as far as components used there are; a riser board for the GPU with a 4-pin Dupont cable, the PCAT module itself that everything plugs into with an OLED screen attached, 3 PCI-e cables for when a card calls for more than 2x 8-pin connectors, and a Micro-USB cable that allows you to capture the data on the system you're hooked up to or a secondary monitoring system.
Well, that's what it looks like when all hooked up on a test bench, you're not going to want to run this one in a case for sure. Before anyone gets worried, performance is not affected at all by this and the riser board is fully compliant with PCIe Gen 4.0. I'm not so certain about those exposed power points however, I will be getting the hot glue gun out soon for that. Now, what does this do at this point? Well, two options: Plug it into the computer that it's all running on and let FrameView include the metrics, but that's for NVIDIA cards only so a pass, OR (what we do) plug it into a separate monitoring computer and observe and capture during testing scenarios.
The PCAT Power Profile Analyzer is the software tool provided to use to capture and monitor power readings across the PCI Express Power profile. The breadth of this tool is exceptionally useful for us here on the site to really explore what we can monitor. The most useful metric on here to me is the ability to monitor power across all sources, PCIe power cables (individually), and the PCIe slot itself.
Those who rather pull long-form spreadsheets to make their own charts are fully able to do so and even able to quickly form performance per watt metrics. We've found a very fun metric to monitor is actually Watts per frame, how many watts does it take for the graphics card to produce one frame at a locked 60FPS in various games, we'll get into that next.
Control Power
Control was the first game that we wanted to take a look at running at 1440p with RT and DLSS on, and then again with DLSS off, this is the game that NVIDIA used when showcasing the performance per watt improvements of Ampere, and well..they were right in the claim there.
From these results for Control is shows that NVIDIAs measurements and claims of improvements were accurate, but it's not always the case. We tested Forza Horizon 4 in a spot to test the same way again but this time at 4K and looking at when we target at 4K60 scene in this game
'Titan Class', that's what NVIDIA referred to the GeForce RTX 3090 as, not as the flagship but a 'Titan Class'. What does that even mean? The GeForce RTX 3080 was called the flagship, but this is also a GTX 30 Series and before the Titans were distinguished by that, the Titan branding, so what's the deal here? Well, it's not a dual GPU like the 90 class of the past so why the 90? Well, from my experience with this card it's occupying a very strange place in the market that I'm not sure would have made sense before this year. I'll elaborate.
More than ever before are people working from home in many different industries, that's just a fact of the world right now. Many of those people working from home are editors, designers, and modelers who also play games. Typically the Titan would come in as that between consumer and professional card, a bit less expensive than the Quadro line but carrying the higher VRAM capabilities and still being able to push the envelope as the absolute absurd top tier gaming card for those who are crazy enough to shell out for it. Well, the GeForce RTX 3090 is more along the lines of catering to that person but also doing so at a much lower price point than what they did the last go around.
What you're getting here is an 82 SM packed GPU with 10,496 CUDA Cores, 328 3rd Gen Tensor Cores, 82 2nd Gen RT Cores, and a mammoth 24GB of GDDR6X all for a grand total of $1499. To put that in a bit of market perspective the GeForce RTX 2080Ti 11GB came to retail at $1199 and in order to get a 5-6% performance boost along with 24GB of GDDR6 you were going to spend $2499 for it. The performance jump from the RTX 3080 to the RTX 3090 in gaming applications fell between 10-15% which is modest but better than the previous upgrade and coming in quite a bit less expensive. I'll wrap up on those concerns toward the end here.
Where the GeForce RTX 3090 shines is in the world of creative professionals who know they can put that massive VRAM pool to use for more than just gaming. Working with 8K Raw Footage in Davinci Resolve was a breeze on the RTX 3090 as well as the Titan RTX, but the RTX 3090 was able to present that experience for a savings of a smooth grand. VRAY performance on the new Ampere architecture is through the roof, same goes for OctaneRender so long as you have the VRAM to support it otherwise you'll find yourself stumbling. Blender really benefits from the architectural improvements and shows quite the speedup over the Titan RTX when the other cards just don't have the VRAM to keep up.
I want to take a moment to bring things back around to the cooler. They call this a BFGPU and that's quite appropriate. It's massive, it's effective, and it's quiet. It does a better job keeping the 350w RTX 3090 cooler and quieter than the 320w RTX 3090 and allows the card's clocks to stay comfortably past its boost rating. But, you're going to want to make sure you have the space for this, it overhangs in every directly and basically filled my Cooler Master H500P Mesh when I tossed it in there out of curiosity.
What about 8K gaming? Well, I don't have access to an 8K panel but I don't think you're going to have to look far to find reviewers who do like Anthony at TweakTown. But for 4K gaming, the GeForce RTX 3090 is simply unmatched. And that's kind of what NVIDIA went for here. While the GeForce RTX 3090 was consistently 50% faster, or more, over the GeForce RTX 2080Ti the Achilles Heel of the RTX 3090 isn't really its price so much as it is the RTX 3080. The GeForce RTX 3080 comes within spitting distance of the GeForce RTX 3090 and does so for much less of your money. So why would you go for the RTX 3090 instead? It's simple, those who know they can use what the card offers will likely find great value in it, others might just want to have the absolute best on the market and past Titan sales have proven that person does exist. But folks like me who are simply PC Gaming enthusiasts might just find themselves better suited with the likes of the Ampere Flagship gaming card instead. If you're loaded up on a 24+ core Threadripper and 128GB or more of memory then you very well are likely to be in the group of unique users who can truly benefit from the flexible nature and performance that the GeForce RTX 3090 offers.
The post NVIDIA GeForce RTX 3090 24 GB “Ampere” Founders Edition Review – A True BFGPU by Keith May appeared first on Wccftech.
source https://wccftech.com/review/nvidia-geforce-rtx-3090-24-gb-ampere-founders-edition-review-a-true-bfgpu/