@Esot Technology is composed of several old cannons who do not want to lie flat, the main content includes mobile phones, computers, photographic equipment, digital Coldplay, looking forward to your attention and interaction.
Author: Happy Fat House Lab
Readers familiar with the rhythm of NVIDIA's new product release should have experience, generally speaking, Nvidia's each generation of game card architecture life cycle is two years, within one year after the release of the new architecture is the time period for launching new products from top to bottom to complete the product line, into the third year after the product release needs to gradually clean up the old architecture products waiting for the new architecture to be listed, at this time, even if the new product is generally launched, it will be a strange castrated version of the waste film utilization type (such as GTX1650 GDDR6). But after all, everyone understands, the market rhythm of the past two years can not be inferred according to common sense, first the low-end market has been completely abandoned, even according to the MSRP of the RTX3050, there has never been a certain generation of graphics card its lowest-end model suggested retail price of up to 1899 this situation appears (1030, 1650 are hundreds of products when it comes out); the high-end rhythm is also very chaotic, first RTX3080Ti changed the specifications several times before the release, from 24GB Various claims such as 16Gbps GDDR6 to 20GB 19Gbps GDDR6X have been finalized with the 12GB 384bit GDDR6X. Logically, the high-end product line of the GA102 series has ended, and the complete core + large display memory specifications of titan positioning have been replaced by the RTX A6000, which actually does not require additional product layout.
But in this year, Lao Huang was very surprised to update the GA102 series high-end product line twice, 3080 12GB this is not to mention, the performance and 3080ti close to let me suspect that this is the 3080ti knife under the crooked after the new regret drug. Today's protagonist, 3090Ti, originally thought that it was only symbolically completed the two sets of SM castrated above 3090, and then the official over-frequency gave the full core obsessive-compulsive disorder an option to choose from. But after actually getting the 3090Ti test, I feel that its positioning should be more than that, which is why this article is not called 3090Ti "first review" - the details of this card that are worth mining and commenting on are far more worthy of saying than its performance test, the performance test is very boring, people who pay attention to the graphics card should be able to guess eight or nine even if they use py, everyone is ampere architecture, the scale is not bad to where to go, you can still go to heaven. However, if it is unexpected, everyone is not happy, so put the performance test behind, if you only need to see the 3090Ti performance test, you can directly turn it back.
Again, let's start with the look of the new card to comment on what's so special about the 3090Ti and why I think it's an "exploratory version."
This time I got a rainbow of RTX3090Ti Vulcan OC, from the frontal appearance and the previous 30 series of Vulcan is no different, is still a large 3 fans, angular appearance plus the top LCD display design.
The logo in the middle does not rotate with the fan and the screen that can customize the display content is exactly the same as the previous 30 series Vulcan.
Look at the side to see some differences, the thickness of the 3090Ti Vulcan is thicker than other models of the 30 series, the overall thickness reaches about 2.75 slots, even if the 3 slot card is no problem.
In addition, it can be seen that this time the RTX3090Ti uses a new 12Pin power supply interface similar to (but not the same) as the previous RTX 30 FE, and not only the Rainbow Vulcan, but actually all RTX3090Ti uses a 12VHPWR power input interface.
Through the new interface design, the 12VHPWR power supply interface can support a maximum of 600W input power consumption, which should actually be carefully said to be the specification of PCIe 5.0, just like the old 6Pin power input interface is the official specification of PCIe 2.0 - older players should still have the impression that when there was no such interface, the graphics card with large 4Pin as auxiliary power supply was used. However, it is a pity that although a power specification in PCIe 5.0 is used, but the RTX3090Ti itself does not support the bandwidth of PCIe 5.0, whether it is the GPU-Z display or the actual transmission bandwidth test is PCIe 4.0, fortunately, the bandwidth of PCIe 4.0 x16 is far more excessive at this stage, and the GPU of 3090Ti is unlikely to re-tape after changing the PCIe controller, which is also extenuating.
However, the next generation of graphics cards should be the full use of PCIe 5.0 specifications to design and manufacture, AMD there is no specific situation now said, Nvidia is to confirm the 12VHPWR as the next generation of graphics card standard power interface. Then since the replacement of the new interface will certainly have people worried that the high-power heirloom power supply they bought before can not be passed down, this is not to worry, for different power supplies there are different solutions.
If you are using a non-module normal power supply:
As in the past, when the 6Pin/8Pin external power supply interface first came out, the graphics card with the 12VHPWR interface will be attached with a 2×8 or 3×8Pin to 12VHPWR adapter according to its own power consumption needs, and it is also possible to connect the power supply with 3 8Pin terminals.
If you are using the old full module power supply:
12VHPWR just changed the shape of the interface, but it is still using 12V voltage, then only need to connect the correct +12V DC and COM to the correct location, saying that people can find a custom line to do a 12VHPWR module line (some manufacturers should also have official cables to launch, you can pay attention to the information on the manufacturer's official website). So the full module power supply this kind of thing as long as the power is enough, the device is not aging, the heirloom can still be heirloom, +12V this standard to change the things involved can be too much, as long as the +12V standard is still there, nothing more than a problem of how to connect the module line.
Of course, it is still necessary to remind you that it is necessary to choose the manufacturer's official cable or a module line with excellent quality, 12VHPWR requires a current of 9.2A per pin, compared with 6Pin terminals per pin 6.75A for a heat increase of 85%, if the wire is not passed, it is easy to have safety hazards.
If you are brand new installed to buy power supply:
There are already some power supplies on the market that are prepared in advance, providing a native 12VHPWR solution, such as the Asus ROG Thor II 1000W used this time:
Personally, I feel that the appearance of this silver glossy, front-facing display mirror is much better than the pure black of the Thor generation, and it is not as easy to imagine fingerprints or mirror reflections affecting the viewing of power indicators. Of course, appearance is not the focus, the power supply design takes into account the power consumption requirements of this generation and the next generation of graphics cards, and directly provides a sufficiently high peak power and native 12VHPWR version of the cable. Of course, because the 1000W version was listed relatively early, the 12VHPWR has not been fully finalized, so the current commercial version should still be the old 12Pin terminal (that is, the one used on the NVIDIA 30 series public version), and the subsequent batch and the subsequent 1200W high-power version will be replaced by the standard 16Pin 600W 12VHPWR cable shown in the figure. ASUS said that the previous power supply can also apply for 12VHPWR cable, and then you can pay attention to the official notice, which should be similar to the process of applying for LGA1700 fasteners.
If it is a new installation, and intend to directly start with the RTX3090Ti top configuration, you can consider directly selecting the power supply that natively supports 12VHPWR, anyway, this is the specification standard of PCIe 5.0, sooner or later everyone has to change it, and prepare in advance to make yourself completely OK when you change the configuration later.
Having said that the power supply problem continues to look down on this rainbow 3090 Ti Vulcan other details.
Remove the radiator, before tossing the 30 series graphics card video memory heat dissipation audience should be able to see at a glance that this video memory thermal pad is so familiar, the original direct use of Laird 90000 card I am really the first time to see, it is estimated that the manufacturer is also scared by GDDR6X.
The problem is that as the size of gpu cores skyrockets, the demand for bandwidth will only get higher and higher, and the next generation of video memory will be hotter. Simply take advantage of this NVIDIA to give 3090Ti the opportunity to change video memory (and maybe the next generation card uses the same video memory), "explore" the improved video memory heat dissipation scheme, Rainbow here is directly used Lai 90,000 and directly welded on the base of the heat vapor plate on the memory conductive plate, compared to the old 30 series graphics card to make further improvements. As for the effect after such an improvement, the temperature test will show you later.
Speaking of thermal memory heat dissipation, by the way, the biggest difference is that the new memory is replaced, avoiding the problem of how to dissipate heat on the back of the double-sided memory, and the frequency is increased to 21G.
From the official website of Magnesium, the code of the video memory can be found, 512M per particle×32/8 = 2GB, the equivalent rate is 21Gbps, and the operating voltage is 1.35V. With the previous 19Gbps GDDR6X looks completely consistent in the package, I don't know if it can be pin to pin direct replacement.
Let's move on to pcbs and thermal designs.
The alien PCB should have been seen in the first review of 3080/3090, but at that time it was still limited to the FE old yellow version, the RE reference design version and the AIC non-public version designed by each manufacturer itself were still square PCBs. All models of this generation of 3090Ti use this special-shaped PCB - and not only the shape of the PCB (why the shape is designed like this, I will say next), the power supply design, layout and standard of the PCB are all decided by the old Huang, and all kinds of non-public can only be fine-tuned in terms of materials and wiring. The previous RE is a reference design, everyone looks at using or comparing to doing one, the reference design on the 3090Ti has become a mandatory standard, we must do PCB according to this standard, and it is not known why Lao Huang wants to do so and whether the 40 series graphics card (especially the high-end model) will take similar regulations.
(The three in the upper right corner are not powered, that is the RLC primary filter with a +12V input)
At this stage because of this rule, each 3090Ti in addition to the heat dissipation design and PCB layout adjustment in the power supply capacity and materials are completely consistent, all are 28 groups of × 70A absolute redundancy design, power supply capacity does not need to worry about the slightest (I even suspect that it is because these device manufacturers have made a bunch of 70, 80A and even 105A integrated MOS, we can so fly themselves in terms of chip power consumption), the advantage is not to worry about what shrinkage to buy, The downside is that it feels like DIY fun is a little less fun.
On the back, there was some debate about the capacitor problem when the 30 series was first launched, and then the mine owners who did not pick the capacitors were silent after entering the market (consumers with more money and less money were not welcome anywhere). 3090Ti above also uses a similar back patch + core 4+ 2 group with the original FE version, pleasing to the eye, but it is quite pleasing to the eye, that is, if this integration degree is broken, if you want to repair it yourself, think about it...
Of course, AIC can still add some functions on the basis of the reference (strong) examination (system) design, such as one-click OC (dual BIOS) and its own personalized lighting effects, etc., but it is limited to this (it seems that the playability of the graphics card itself is in addition to power supply/heat dissipation, right?). ）
Next look at the heat dissipation, in fact, after looking at the radiator, you should be able to understand the reason for the special-shaped PCB - you can not design a 600W power supply interface, the energy is conserved consumption of a joule of electricity will produce a joule of heat, you have to dissipate these heat to maintain the normal operation of the system. In the experience of using the CPU radiator, we all know that the side blowing tower radiator effect of the air duct is much better than the downward pressure dissipation of the same scale, so this penetrating heat dissipation is to change the original downward pressure type blown to the PCB into a side blown type, and the same wind pressure after hollowing can provide more air flow, which is convenient for blowing through the heat sink that is thickened to 2.75 slots.
However, the penetrating heat dissipation although the graphics card is comfortable, but...
I think, this kind of thing, the water directly discharged heat is the only way out, the chassis stuffed with four or five hundred five hundred or six hundred watts of things, other accessories are really a bit unbearable.
So our conclusion came, don't look at this card is just one more Ti, more than two sets of SM, a little more frequency, but all aspects of the design can be said to be completely redone, if it is just for such a model then directly on the 3090 or even 3080 PCB core replacement can be, 3090/3080 non-public PCB power supply is extremely redundant everywhere, you can continue to be compatible with 3×8Pin external power supply, why bother. And these parameters, such as PCIe 5.0 standard 12VHPWR power supply, standardized PCB design, up to 600W power consumption and heat dissipation requirements and the like are highly similar to the previous Ada Lovelace 40 series graphics card, so according to these information reasoning conjecture, 3090Ti may be a 40 series graphics card with a GA102 core, first to the new standard new design trip to the mine (it is estimated that the previous two months of delay is because of the mine clearance), By the way, it is entirely possible to share a little research and development and material costs. So rather than saying that it is the final chapter of the 30 series graphics card, it is an exploratory version of the 40 series card, in addition to exploring the new design of the old yellow, you can also explore your power supply and chassis, if it is no problem to cope with it, there is no problem to cope with the new card of the 40 series (fog).
The introduction and analysis of the RTX3090Ti is here, and then look at the most boring part of the article - performance testing, because it is still an ampere architecture, no new design has been introduced, and the performance gap is a pitifully small scale difference and a slightly more frequency difference. The convention is to put the picture first:
In addition to the 21Gbps video memory, the core frequency is also a little over, which is basically the frequency of the previous top non-public version 3080Ti/3090.
GPU-Z can now also detect the complete information, which is consistent with the specification table above.
Power wall information, default 450W, maximum 480W. Power walls up to 510W can be obtained by switching to the Boost BIOS.
Before the performance test, let's talk about the test platform configuration:
CPU: Intel Core i9 12900K 5.3G/4.0G Large Core Hyperthreading Shutdown
Motherboard: ASUS Maximus Z690 Hero (it is still the same set that was first tested at the previous 12900K)
Memory: Zhiqi Xuanfeng DDR5 6000c36 @ 6200c32
SSD: Pucott M10P 1TB
Power supply: ASUS ROG Thor II 1000W direct connection to 12VHPWR terminals
Although it is a non-public card, the default BIOS of the Rainbow RTX3090Ti Vulcan is exactly the same as the public version of the 3090 in terms of frequency and power limitations, so let's compare the results of the public version of the RTX3090 first:
According to the weightless average to calculate the performance of the public version of RTX3090Ti is 14.35% stronger than the public version of RTX3090, there was also a rumor before that 3090Ti was 14% stronger than 3090, really it is no problem to say so, but to be honest, the reason why everyone doubts this data is that the public version of 3090 only gives 350W power consumption, which does not represent the actual level of 3090 in the hands of users (all buy 3090 I am disgusted by the entire version of the 3090)? Not to mention that this 350W power consumption is also divided by 24 video memory, and the actual game frequency of the public version 3090 is unbearable:
Although it is true that the public-to-public comparison is correct, I think the reference value of this data is really a bit..., and taking a step back, using 28.5% power consumption for 14.3% performance improvement is not a very commendable thing.
However, I don't have a non-public version of the 3×8Pin 3090 in my hand, so I found an RTX3080Ti Super Dragon to compare. On the one hand, for high-end gamers, the possibility of choosing 3080Ti or even 3080 12GB should be much greater than 3090, on the other hand, 3080Ti Super Dragon is also a single-sided GDDR6X video memory, only a little less 440W power consumption cap, I personally feel that compared to the 3090 public version, this card is more like 3090Ti, and the reference value for high-end gamers whether it is necessary to upgrade will be greater.
Almost the power consumption wall of 30890Ti and 3090Ti game performance comparison, certainly not surprising is that 3090Ti is still leading, but the leading range is only 5.6% on average, the previously rumored performance improvement can not really say how strong 3090Ti is, but the 3090 public version is too weak, can it be stronger than 3080 12GB have to play a question mark (the back will do GA102 full series of game cards comparison test, when to say it again... ）
So if you only need to go to the 4K position, then it is clear that the 3090Ti belongs to a choice that can go further but the cost performance is not too high, after all, this scene can not actually play its biggest advantage - 24GB of video memory. So for me and other stinky game users what can 24GB video memory do, there must be people who will think of 8K games, but the current 8K display equipment is expensive to choose rare and a bunch of problems, but a variety of 4K high-brush high-quality screens can now be bought at a more reasonable price, so I thought of a thing - DLDSR.
4K screen to open DLDSR The actual rendering resolution is 6K, 6K through DLSS interpolation to 8K and then zoom back to 4K, should be the best solution to use the current more mature 4K screen experience under the "8K" game texture, the requirements are not as perverted as 8K rendering, so in this scene 24GB video memory can play some role?
In the actual test, even 6K rendering in some games also requires 24GB of video memory, 12GB of video memory direct PPT, frame performance 3090Ti in the 6K DLDSR 8K setting under the Killer 3 and Tomb Shadow can even be close to 60 frames, can be compared with the flow of play. The remaining thirty or forty frames of the game are not completely unplayable for players who are accustomed to cinematic graphics, and should be the most appropriate way to experience "8K games" at present.
Finally, power dissipation and overclocking
25 ° C room temperature test 20 laps of Time Spy, stability of 99.4%, core maximum temperature of 70.6 ° C, this cooling system to cope with 450W feeling there is still a surplus, of course, this is an open platform, after packing the temperature should be higher, really to cope with the legendary 600W I think it is still a bit reluctant, but for this 3090Ti itself is already a good temperature performance.
In terms of the effect of the new memory replacement, the maximum power consumption of the memory is 86.7W, which is similar to the 3080Ti 12 1GB 19Gbps video memory, and the process should be somewhat improved. The maximum memory temperature is 78 ° C, which is lower than the 3090 back memory that is not clear to the point of being moving.
Overclocking in terms of the frequency of the new video memory can run up to 23Gbps (running less than 24Gbps always feel a little obsessive-compulsive outbreak), the core Boost up to 2145MHz, then the Time Spy score is almost less than 23000, compared to the default state to increase 4.5%, the value is not worth so tossing and seeing it, some users think that the original official help you ash good on the line, some users think that I have spent money not to squeeze clean that is not a loss, are OK, Just decide according to your own habits.
Finally, there is the purchase advice and summary, to be honest, I think this card does not need any purchase advice at all, its corresponding consumer group is too clear, just and the Intel Core i9 12900KS released in a few days to get together. Its biggest significance is still as an "exploration version", telling us what the design of the graphics card in the future of the era when the semiconductor process dividend disappears and the inevitable performance can only be exchanged for performance with power consumption. Perhaps the remaining problems are how much more high-power power supplies need to spend, and the impact on indoor air conditioning effects in summer.