However, no matter how powerful it is, it will inevitably roll over. In history, Nvidia has experienced several major rollover events, which have caused a devastating blow to the company's reputation and even the entire industry. At the same time, it also hurt the hearts of many consumers.
1. GTX 480: Barbecue video card
Nvidia once launched the "Great Leap Forward" in the 40nm Fermi Fermi architecture, raising the number of CUDA cores to 512, twice the number of GT200 cores. The single and double precision reached 1:2, and a large number of cores were used for double precision computing. However, the first generation Fermi Core GTX 480 graphics card released in 2010 also rolled over.
Although this video card has a very good run score, it has extremely severe fever, which is so serious that it can hardly be used normally.
At that time, some anchors said, "G means barbecue". In the real machine test, the power consumption of GTX 480 under normal load is far more than 200W, and the temperature is close to 90 degrees; The peak power consumption reaches 300W, and the maximum temperature reaches 98 degrees.
GTX480 is worthy of being the first video card to realize electromagnetic oven cooking. Most happily, it pioneered the video card fried eggs, and Nvidia's follow-up products did not eliminate the fate of fried eggs.
Because of its poor heat dissipation performance, it was also ridiculed as "Thermo" (the combination of Fermi and Thermal) by netizens. It is equivalent to saying "I have super performance, but I use nuclear power".
Later, NVIDIA solved many problems with the Fermi 2 architecture, and the GTX 500 series also eased. However, later, Nvidia's Titan Z, GTX 1080Ti, and RTX 2080Ti were somewhat overheated, and power consumption and hair winding were still the pain consumers could never get around.
2. Titan Z: High price harvesting meets cold
Nvidia's dual core Kawang GeForce GTX Titan Z had two consecutive ticket jumps that year. The first one was because the driver was not ready. The fundamental problem of the second one was that the performance was not up to standard.
At that time, some evaluations showed that even in the actual use process, the core frequency would automatically accelerate significantly to 1058MHz, exceeding the acceleration frequency of 1018MHz of R9 295X2, but usually it could only maintain around 900MHz, and the actual score was significantly lower than R9 295X2, which was obviously unacceptable.
When Titan Z was listed, the asking price was as high as 3000 US dollars, that is to say, 21999 yuan from the Bank of China. Many players and professionals were deterred by the cruel collection of wool. In addition, the Kepler architecture, which was gradually lagging behind at that time, was the main reason for its poor performance in business.
In addition, Titan Z is also inevitably suffering from high temperature. The full load power consumption of Titan Z's two cores is up to 82 ℃ and 79 ℃ respectively. In addition to heating, the fan turns up, just like the helicopter. Many people complain about this.
3. TITAN V: scientific calculation error
Another Nvidia TITAN also rolled over.
Before the arrival of Turing and Ampere architectures, the most powerful single card for mining and playing games under Nvidia was the TITAN V, which cost 22999 yuan for the Bank of China. Such "nuclear bombs" could not avoid bugs.
In 2018, according to TheReg, scientists in some professional fields found that there would be output errors in the simulation calculation of TITAN V, that is to say, the results of multiple calculations of the same equation are actually inconsistent.
The research team bought four TITAN Vs, and two of them rolled over. When the test came down, there was a 10% chance that all of them would be wrong results.
At that time, some insiders in the chip circle analyzed that Titan V may have memory access errors due to its excessively high performance setting, overload of chips, and cutting off the error correction mechanism of video memory.
It was also analyzed that in order to save power consumption, the rounding mode was simplified in general calculation of floating point numbers, so this problem occurred. The problem of avoid can be solved in the upper algorithm.
Nvidia officially said that the design of all our GPU products is accurate, and Tesla has ECC error correction function, which is just to deal with the above high-performance professional computing services. Later, the matter also ended.
With this incident, at that time, the industry began to pop a poem - "Seeking, cold and desolate, sad and sorrowful. When the Titan overtook, it was most difficult to give up. How can the local tyrant, the golden turbine skin, rival others, power consumption upgrade! The Weaver Girl died, sad, but the agricultural enterprise died. The black leather Titan piled up, haggard and damaged, but now it can still be bought? Look at Old Huang from a distance, and castrate alone with a knife! Frequency reduction is more expensive, and in the end, the price is very low. "
4. RTX 3080/90: capacitor jerry building
Shortly after the launch of Nvidia RTX 3080/90, a large number of rollover posts suddenly appeared on the Internet, ranging from blue screen, black screen, flickering or even crashing, to direct burning of power supply. There have been more than 50 cases of this kind in the post bar alone, which has caused a lot of waves in DIY circles.
As the event slowly fermented, all kinds of rumors began to spread, pointing to the brand manufacturers' jerry building. After all, ASUS had nothing to do.
The foreign media igorslab also reported at that time that the large area of game crash may be related to the type and number of capacitors used by the chip on the back of the graphics card. Nvidia selected MLCC+POSCAP when determining the chip capacitance scheme. MLCC is a multilayer ceramic chip capacitor (as shown in the green part of the figure below), and POSCAP is a conductive polymer titanium solid capacitor. Relatively speaking, MLCC has better performance and POSCAP is cheaper.
At that time, most emphasized that the existing design, especially the capacitor configuration, did not have any problems, but some manufacturers were quietly upgrading to ensure safety.
Then, Nvidia updated the GeForce 456.55 driver, which improved the stability, and the problem of collapse seemed to be alleviated to a great extent. However, according to the test of netizens, it seems to be at the expense of frequency.
What is more outrageous is that careful netizens found that after experiencing the "capacitance gate", the manufacturer began to secretly upgrade the capacitance design of the graphics card completely changed. For example, MSI's RTX 3080 GAMING X TRIO (Magic Dragon Version) is like this. The combination of 5 POSCAP capacitors and 10 MLCC capacitors (5+1) was originally used on the back of the graphics card chip, and the combination of 4 POSCAP capacitors and 20 MLCC capacitors (4+2) was changed.
Of course, since NVIDIA released the drive, the 3080/90 has never seen a large-scale rollover, and then there has never been any related voice.
5. RTX 4080 12GB: renamed after being sprayed
NVIDIA's RTX 4080 12GB version was "sprayed" by netizens to be cancelled shortly after its release, which is also a "long time to see" in the industry.
This video card is connected with RTX 4090 and RTX 4080 1 6G B launched at the same time, but its market positioning and performance did not meet user expectations.
It was analyzed that Nvidia's RTX 4080 12GB version and RTX 4080 16GB version appear to belong to the same series at first sight, but they are actually two cores, with very different specifications, and only 192 bit memory width.
What's more, it was revealed that the so-called RTX 4080 12GB was originally planned to be called RTX 4070 Ti, or even RTX 4070, but it was renamed at the last moment before the release.
In order to deal with this situation, Nvidia subsequently adopted a repositioning strategy, replacing the RTX 4080 12GB version with a new waistcoat - RTX 4070 Ti, and then re launched it to the market.
Obviously, consumers will not buy it, which is just that it has changed back to its original name. So far, the impact of this matter has not been completely erased, and negative emotions still haunt consumers.
6. RTX 4090: Your video card is on fire
In recent years, the power consumption of graphics cards is getting higher and higher, and natural events are also inevitable. In the 4090 generation, due to design defects, various spontaneous combustion events have occurred, which is the famous connector melting event.
On the RTX3090Ti, Nvidia launched the 12VHPWR interface based on the ATX3.0/PCIE 5.0 power supply specification for the first time, and this design has also continued to 4090. However, the interface has some design problems:
First, the plug and unplug life is short. For example, the 12VHPWR adapter line attached by Sotai can be connected/disconnected for up to 30 times, which is outrageous.
Second, there are strict requirements for the installation of 12VHPWR power supply wire, and no one knows where the problem occurs, which may be Cable It may not be bent, or the connector may not be too loose or tight, or it may be inserted or pulled too many times. In short, as long as you do not plug it in properly, it will overheat or even burn out due to uneven load.
For example, a netizen named reggie_gakil was playing Wild Escort: Redemption 2 when the power supply interface of the RTX4090 video card suddenly caught fire. Soon after this incident, at least five similar incidents occurred in succession.
The problem is the same. The 16Pin power supply socket terminal was burned, and some plastic parts were melted and deformed by heat. Look, your video card is burning!
Later, NVIDIA introduced an improved 12V-2 × 6 connector for RTX 4090, but this problem still seems to exist. Moreover, because of the above events, everyone will be afraid of the mention of 12VHPWR. What is more serious is that everyone has begun to be alert to RTX 5090.
Nvidia RTX 4090 smokes
7. RTX 4070 Super: The initial sales volume is low
At the beginning of this year, the RTX4070 Super graphics card was officially launched, which is 10~15% higher than the non S model, superior to the previous generation RTX3090, with performance close to 4070Ti and a price of $200 cheaper.
Although Nvidia believes that it will not increase the price if it increases the quantity, everyone will buy and buy. The result was surprising. Many retailers reported that the RTX4070 Super graphics card did not perform well after its debut, and it is likely to become Nvidia's worst selling graphics card so far. In some regions of Europe, the price of the RTX4070 Super is lower than its suggested retail price.
In Poland, the official price of the standard frequency version of Gengsheng RTX4070 Super Ghost is 2999 zloty, but the retail platform X-Kom has made a break price of 2849 zloty, a decrease of 5%. In CSV, Germany, Yingzhong RTX4070 Super TWIN X2 sold for 659 euros, but now it has dropped to 643 euros, a decrease of nearly 3%.
Some people say that the initial sales of this video card is even worse than the RTX 4080. Moreover, some retailers said that not only players were not interested in this video card, but scalpers did not want it, and even cancelled a large number of pre orders.
