is selected from techpostml2
Author: Steven Walton
Heart of Machine Compilation
Edited by: Zenan, Xiaozhou
Top game masterpieces, all of which can be turned on 4K144Hz high special effects, and the power consumption has not increased. Now the pressure and bottlenecks are all on CPU, which is the RTX 40 era.
htmlOn September 21, Nvidia released the GeForce RTX40 series with a new architecture at the GTC conference. The new graphics card has achieved a significant performance improvement due to TSMC's 5nm process, the new ray tracing processor and DLSS 3.
In the early morning of this year, various media released review articles for the flagship version of RTX 4090. The highest version seems to be as powerful as Nvidia claims before, and is "the most cost-effective".
RTX 4090 is based on the AD102 chip and has a size of 608.4 square millimeter , which is about 3% smaller than the GA102 used by RTX 3080 and 3090. The shift from the Samsung 8N process of the previous generation Ampere to TSMC's 4N process has increased the number of 4090 transistors in the Ada Lovelace architecture by 170%, from 28.3 billion to a crazy 76.3 billion.
Compared with RTX 3090 Ti, the 4090 stream processor , CUDA core, Tensor core, RT core and texture units increased by 52%. ROP count increased by 57%, and accelerated clock frequency increased by 35%. The same 21 Gbps GDDR6X memory is used, and there is still 24GB of memory on the 384-bit wide memory bus, so it is the same 1008 GB/s memory bandwidth. GPU also retains the same PCI Express 4.0 x16 interface.
Nvidia claims that the power rating of the RTX 4090 is 450W, the same as the 3090 Ti, while the maximum temperature has been slightly reduced from 93℃ to 90℃. The minimum power supply requirement is 850W, and the power supply of this power was also used during testing.
Someone also released photos compared with X Box X and PS5, which is like this:
The heat dissipation is stable, but I don’t know how to stuff it into the chassis?
In terms of design, the Founders Edition RTX4090 looks similar to the 3090, but there are some pretty obvious changes. The most eye-catching thing is the width of the graphics card. The 3090 has a width of 52mm, while the new RTX 4090 has a width of 61mm. Despite the increase in size, both models weigh the same, both at 2190 grams.
In addition, it is the lack of NVLink connector. NVIDIA has canceled NVLink and instead relies on the PCIe 4.0 bus. Other significant changes are the 16-pin power input that has been upgraded to the PCIe 5.0 specification, also known as the super catch 12VHPWR power connector.
A single PCIe 5.0 power connector provides up to 600W of power, which previously required four 8-pin power connectors. You don't need to replace a PCIe 5.0-compliant PSU, because the RTX 4090 comes with a 4x8 pin to single 16-pin adapter, similar to the 3x8 pin to 16-pin adapter provided by the 3090 Ti.
In addition to increasing the number of various cores, GeForce 40 also introduced DLSS 3, which is a unique feature of the current GeForce 40 series.
This new upgrade / frame rate double technology is exciting. Some results can be shown in the game's tests, but the comprehensive analysis of DLSS 3 will continue to wait for more data. DLSS 3 requires a lot of testing and analysis, which is far beyond the scope of the first reviews.
For testing, all GPUs are set to run in the official clock specifications. The CPU of the test platform is Ryzen 7 5800X3D, which uses 32GB dual-column, dual-channel DDR4-3200 CL14 memory X570S Carbon Max WiFi motherboard on MSI MPG.
Game benchmark review
In the game test of "Watchdogs: Legion" (1440p), the GeForce RTX 4090 doesn't seem to be very stunning, but it is indeed the fastest GPU yet - about 9% faster than the 6950 XT.
RTX 4090 is 22% faster than RTX 3090 Ti, which corresponds to the price difference between the two.In the evaluation, the problem with the RTX 4090 seems to be small, but the CPU 5800X3D of the test platform has become the main bottleneck of the system.
Therefore, the evaluation team raised the resolution to 4K to test how high the frame rate of of RTX 4090 is. It turned out that 4090 is 60% higher than 6950 XT and 64% higher than 3090 Ti. This is the real high refresh rate gaming experience of 4K.
Meanwhile, the RTX 4090 is 60% higher than the 3090 Ti frame rate, and although the gap with the Radeon 6950 XT has been growing, we now see the gap with AMD at 102%.
" Far Cry 6" usually performs better on A cards. At 1440p, the CPU bottleneck effect of the test platform is severe. Using RTX 4090 can only achieve 9% performance improvement, with an average refresh rate of 187 fps. After converting
to 4K resolution, RTX 4090 brings huge performance gain, with an average refresh rate of 164 fps, a 34% increase over the 6950 XT, with a larger increase compared to the RTX 3090 Ti, reaching 50%, and even 60% increase over the standard 3090.
"Assassin's Creed: Valhalla" is also a game that runs well on A cards. The evaluation results show that at 1440p, the RTX 4090 has a 27% higher frame rate than the 6950 XT, which is a good gain, but far less than other comparison improvements, and the RTX 4090 also only improves 37% over the 3090 Ti, rather than the 60% gain at 4K.
Another game, "Cyberpunk 2077", has high requirements for both CPU and GPU. And with the RTX 4090 installed, we are almost certain that the CPU will be subject to 1440p limit. The evaluation results show that the average refresh rate of the RTX 4090 on cyberpunk is 145 fps, 33% higher than the 6950 XT and 36% higher than the 3090 Ti.
However, the performance gain in the 4K situation is not as large as expected, with the RTX 4090 improving by 51% over the 3090 Ti, but the RTX 4090 far exceeding 60 fps, which is far from reaching by other GPUs.
Performance Summary
Techspot's evaluation was tested on 13 games. At 1440p resolution, the average performance of GeForce RTX 4090 is roughly as follows: the output frame rate is 219 fps, 44% higher than the 6950 XT and 45% higher than the 3090 Ti.
These are huge improvements, but they still cannot fully illustrate the excellent performance of RTX 4090, because the system's 5800X3D was limited many times during the test, although 5800X3D is already a very fast gaming CPU. This shows that one of the fastest gaming CPUs at present will still cause serious performance bottlenecks to the RTX 4090 at 1440p.
In 4K games, the GeForce RTX 4090 averages 59% higher than the RTX 3090 Ti and 71% higher than the 6950 XT, which are huge improvements. Also impressive is that the RTX 4090 has an average frame rate of 145 fps at 4K, which is exactly the upper limit for many high-flash monitors. This is indeed the first 4K GPU to deliver a high refresh rate gaming experience.
In addition, when using Ryzen 7 5800X3D at 1080p, the RTX 4090 is 28% higher than the 3090 Ti and 24% higher than the 6950 XT.
Ray Tracing and the new generation DLSS
Cyberpunk 2077 is one of the first games to support DLSS 3. Let’s take a look at its performance below. Please note that this is just a preliminary review.
When using high-quality preset 1440p, the average frame rate of RTX 4090 in cyberpunk is 145 fps, about 35% higher than the 3090 Ti and 6950 XT. However, performance improvements by 90% with DLSS 3 enabled to 276 fps. The real advantage of
DLSS 3.0 can be seen with ultra-high quality ray tracing enabled, here the RTX 4090 has an average frame count of 191 fps at low resolution, which is nearly 70% higher than DLSS 2. Then at 4K, the GeForce RTX 4090 turned on DLSS 3 output 113 fps, a 43% increase over DLSS 2.
power consumption
Moore's Law After the end of , the way to improve chip performance is often to increase power consumption, which is particularly obvious on flagship GPUs.The tester tried to play "Hao: Infinity" at 1440p and observed the total system power consumption, and he saw that the RTX 4090 was not actually as bad as he thought. In fact, it performs 50% more than the RX 3090 Ti, and consumes less power. Both the RTX 4090 and 3090 Ti are 450W graphics cards, so technically they should consume about the same power. The RTX 3090 Ti used in
test increased the total system usage rate by about 50W, because there is no Founders Edition of 3090 Ti on the market, and the non-public version of MSI with a voltage higher than the specification is used. Although it has been downgraded to the specifications of the official version, which has been reduced by several percentage points in terms of performance, higher voltage means that power consumption is still higher than the base model.
Let’s take a look at how efficient the RTX 4090 is. By locking the frame rate at 90 fps, we can see how much power each GPU is using.
After a fixed frame rate, the RTX 4090 consumes only 215W, which means that for the same level of performance, the 3090 Ti requires 93% more power, while the 6950 XT requires 40% more power. So while people are talking about the power explosion of the RTX 4090, it is actually very impressive in terms of efficiency.
heat dissipation
In terms of cooling, the hot spot of the FE version of the RTX 4090 reached a peak of 83°C after one hour of high load operation in a 21°C room installed in the ATX chassis and the door is closed. The average GPU temperature reaches a peak of 72℃ and the memory temperature reaches a peak of 84℃, all of which are acceptable considering what is running.
In addition, the fan speed is only 1600 RPM and the running volume is only 42 dBA, which is quieter than most high-end and even mid-range graphics cards tested in the past. In this test, the core clock speed was fairly stable at 2730 MHz, memory was running at 21 Gbps, and the last GPU power consumption averaged 415W.
cost-effective
For cost-effective, our use per-frame cost analysis, it looks like the RTX 4090 is roughly comparable to the Radeon 6950 XT and 6900 XT levels, but is more than 30% higher than what we think the reasonable choice cost of the previous generation of high-end GPUs in the RTX 3080 and 6800 XT — in fact it is about 50% more expensive than the 6800 XT.
If you compare the actual selling price, the attractiveness of RTX 4090 will be reduced. The cost-effectiveness is slightly worse than the 3090 Ti at 1440p resolution, making it the most expensive GPU per frame and overall price on the market.
090 is obviously more efficient at 4K resolution, here we see it is slightly better than the 6950 XT and slightly better than the discounted RTX 3090, which is pretty good for premium products.
However, mid-to-high-end graphics cards such as 3070 can always provide higher cost-effectiveness.
Summary
GeForce RTX 4090 is obviously a very fast GPU, and people's main concerns about it are pricing and power consumption. It is obvious that it has no increase in power consumption compared to the previous generation. On the other hand, due to the large increase in performance, its efficiency has also increased. You can see that when limiting frame rates, the RTX 4090 consumes nearly 50% less power when gaming than the 3090 Ti and nearly 30% less than the 6950 XT.
But this is just the beginning of the new generation: before more RTX 40 Ada Lovelace GPUs and of course the competitive AMD RDNA3 GPUs appear, it is hard to say whether the RTX 4090 is the most cost-effective one.
But if you don’t care whether you pay the IQ tax, the RTX 4090 is worthy of appreciation as a graphics card. This is the first time we can enjoy a real 4K high refresh rate gaming experience without affecting the visual effect. Ray tracing has finally become a carefree option, and while DLSS is still important, enjoying ultra-high quality ray tracing games in 4K, such as Cyberpunk 2077, Watchdog, while maintaining frame rates above 60 fps is a special experience.
DLSS 3 is an exciting new feature, but we need to dig into it before we can further comment on its value.
Of course, this is just an evaluation of the public version. We ran 4090 normally with only the original Corsair RM850x 850w PSU. The FE card runs very quietly at full load for a long time, not noisy than a high-quality mid-range graphics card, and does not consume much power.
This is not to say that RTX 4090 is suitable for everyone. It costs more than 10,000 yuan, but if you decide to buy 4090, there shouldn't be much trouble, and the inventory is obviously sufficient.
For most people, it is normal to enjoy the game with half the price, or a quarter price GPU, we can wait for the RTX 4070, and of course AMD's upcoming RDNA 3 series.
Reference content:
https://www.theverge.com/23398201/nvidia-rtx-4090-review-test-benchmark
https://www.techspot.com/review/2544-nvidia-geforce-rtx-4090/
